Stop two tests failing on what is around them

The macOS path test set XDG_CONFIG_HOME to "/c" and asserted that "/c" was nowhere in the answer, but the home these run under is a mkdtemp path, so any TMPDIR with a "/c" in it failed the test. macOS is exactly where that happens: its temporary directories are /var/folders/<two letters>/, and one letter in thirty-six starts with c. The needle is now a word no directory can be called.

The other is the `ask` verb reading what was piped into it. Nothing was, but the runner's own stdin is not nothing either, so the verb read whatever the runner left there. It now reads an empty stream, whichever runner is in use.

README.md

@@ -115,9 +115,13 @@ or runs it on the spot.
 Speech to text and cleanup each pick a provider in the settings window, and both
 run here by default, on models of your own. The cloud is the other option:
 speech to text on **OpenAI**, **Groq** or **OpenRouter** (`gpt-4o-transcribe`),
-cleanup on OpenRouter (`google/gemini-3.5-flash-lite`) or, when either is
+cleanup on OpenRouter (`google/gemini-3.5-flash-lite`), on **Google AI Studio**
-installed, on Claude Code or Codex. The keys fall back to `OPENAI_API_KEY`,
+(`gemini-3.5-flash-lite`), on **OpenCode Go** (`deepseek-v4-flash`) or, when one
-`GROQ_API_KEY` and `OPENROUTER_API_KEY`, and are stored in
+of them is installed, on Claude Code, Codex or Antigravity. The first three are
+a single HTTP request; the three CLIs each open a whole session to do it, which
+is where their few extra seconds go. The keys fall back to `OPENAI_API_KEY`,
+`GROQ_API_KEY`, `OPENROUTER_API_KEY`, `GEMINI_API_KEY` and `OPENCODE_API_KEY`,
+and are stored in
 `~/.config/dikte/config.json`, mode 600, or in
 `~/Library/Application Support/Dikte` on a Mac. Cleanup can be switched off, in
 which case the raw transcript is pasted, and a thinking model's effort can be
@@ -189,11 +193,13 @@ running.
   what comes of it: the answer, or a sentence saying what was done. It is the
   session you would have opened yourself, so your skills and connected services
   are there, which is what makes "put that in my calendar on Thursday at three"
-  a thing you can say to a window that is not Claude. Codex (`codex exec`) runs
+  a thing you can say to a window that is not Claude. Codex (`codex exec`) and
-  the same way, and OpenRouter is there as a plain question-and-answer fallback
+  Antigravity (`agy -p`) run the same way, though Antigravity takes neither a
-  for a machine with neither CLI on it. Provider, model, permissions and working
+  permission mode nor a sandbox from Dikte: what it may do without asking is
-  directory are under Settings → Agent, and commands close together stay in one
+  whatever its own allow-rules say. OpenRouter or OpenCode Go is there as a
-  conversation.
+  plain question-and-answer fallback for a machine with no CLI on it. Provider,
+  model, permissions and working directory are under Settings → Agent, and
+  commands close together stay in one conversation.
 - **Meetings** are recorded from the microphone and the speaker output at the
   same time, which settles who said what by the channel a voice arrived on
   instead of guessing at it. The two sides are transcribed separately and
@@ -240,9 +246,9 @@ cli.py            the command line: every verb, and what it answers with
 ipc.py            one request and one reply over the local socket
 audio.py          PCM capture: pw-record for dictation, ffmpeg for a meeting
 meeting.py        channel split, speaker labelling, cleanup, minutes
-assistant.py      running a dictation through Claude Code, Codex or OpenRouter
+assistant.py      handing a dictation to Claude Code, Codex, agy or a chat model
 api.py            transcription and cleanup requests (stdlib only)
-cleanup.py        who rewrites the transcript: OpenRouter, here, Claude or Codex
+cleanup.py        who rewrites the transcript: a hosted model, one here, a CLI
 ggml.py           whisper.cpp and llama.cpp here: fetch, verify, keep serving
 hub.py            what GitHub and Hugging Face have on offer today
 update.py         whether a newer release is out, and the page it is on

README.tr.md

@@ -112,10 +112,14 @@ Sesi yazıya çevirme ve temizleme, ayarlar penceresinde ayrı ayrı sağlayıc
 seçer; ikisi de varsayılan olarak burada, kendi modellerinle çalışır. Bulutu
 seçersen sesi yazıya çevirme **OpenAI**, **Groq** ya da **OpenRouter**'da
 (varsayılan `gpt-4o-transcribe`), temizleme OpenRouter'da
-(`google/gemini-3.5-flash-lite`) ya da kuruluysa Claude Code veya Codex'te
+(`google/gemini-3.5-flash-lite`), **Google AI Studio**'da
-çalışır. Anahtarları boş bırakırsan `OPENAI_API_KEY`, `GROQ_API_KEY` ve
+(`gemini-3.5-flash-lite`), **OpenCode Go**'da (`deepseek-v4-flash`) ya da
-`OPENROUTER_API_KEY` kullanılır; anahtarlar `~/.config/dikte/config.json`
+kuruluysa Claude Code, Codex veya Antigravity'de çalışır. İlk üçü tek bir HTTP
-içinde, izinler 600, Mac'te ise `~/Library/Application Support/Dikte` altında.
+isteği; üç CLI ise bunun için birer oturum açar, fazladan giden birkaç saniye de
+oradan gelir. Anahtarları boş bırakırsan `OPENAI_API_KEY`, `GROQ_API_KEY`,
+`OPENROUTER_API_KEY`, `GEMINI_API_KEY` ve `OPENCODE_API_KEY` kullanılır;
+anahtarlar `~/.config/dikte/config.json` içinde, izinler 600, Mac'te ise
+`~/Library/Application Support/Dikte` altında.
 Temizlemeyi tamamen kapatabilirsin, o zaman ham transkript yapıştırılır; modelin
 yanındaki kutudan düşünme seviyesini de seçebilirsin.
 
@@ -184,9 +188,11 @@ olmasını ister.
   yapıştırır: cevabı ya da ne yapıldığını söyleyen bir cümle. Kendi açacağın
   oturumun aynısıdır, yani skill'lerin ve bağlı servislerin oradadır; "bunu
   perşembe üçe takvime koy" cümlesini Claude olmayan bir pencerede söyleyebilir
-  olmanı sağlayan da budur. Codex (`codex exec`) da aynı şekilde çalışır;
+  olmanı sağlayan da budur. Codex (`codex exec`) ile Antigravity (`agy -p`) da
-  OpenRouter ise ikisi de kurulu olmayan bir makinede düz soru cevap için
+  aynı şekilde çalışır; ama Antigravity'ye Dikte bir izin kipi ya da sandbox
-  duruyor. Sağlayıcı, model, izinler ve çalışma dizini Ayarlar → Ajan
+  veremiyor, sormadan ne yapabileceğini kendi allow-rule'ları belirliyor.
+  OpenRouter ya da OpenCode Go ise hiçbiri kurulu olmayan bir makinede düz soru
+  cevap için duruyor. Sağlayıcı, model, izinler ve çalışma dizini Ayarlar → Ajan
   sekmesinde; arka arkaya verilen komutlar tek bir konuşmada kalır.
 - **Toplantılar** mikrofonla hoparlör çıkışından aynı anda kaydedilir; kimin ne
   dediği tahmin edilmez, sesin hangi kanaldan geldiğiyle belli olur. İki taraf
@@ -233,9 +239,9 @@ cli.py            komut satırı: bütün fiiller ve verdikleri cevap
 ipc.py            yerel sokette bir istek, bir cevap
 audio.py          PCM kaydı: diktede pw-record, toplantıda ffmpeg
 meeting.py        kanal ayırma, konuşmacı etiketi, temizleme, tutanak
-assistant.py      dikteyi Claude Code, Codex ya da OpenRouter'dan geçirme
+assistant.py      dikteyi Claude Code, Codex, agy ya da sohbet modelinden geçirme
 api.py            transkript ve temizleme istekleri (yalnız stdlib)
-cleanup.py        transkripti kim temizler: OpenRouter, burası, Claude ya da Codex
+cleanup.py        transkripti kim temizler: bulutta bir model, burası, bir CLI
 ggml.py           whisper.cpp ve llama.cpp'yi indirip burada çalıştırma
 hub.py            GitHub ve Hugging Face'te bugün ne olduğu
 update.py         yeni sürüm çıkmış mı, çıkmışsa hangi sayfada

dikte/__init__.py

@@ -10,4 +10,4 @@ business loading Qt to answer one question.
 # both the .dmg's Info.plist and the AppImage's file name are built from it. A
 # build off master rather than off a tag appends the commit to it, so that a
 # bug report from someone running "latest" names a commit.
-__version__ = "1.0.2"
+__version__ = "1.1.0"

dikte/api.py

@@ -1,10 +1,17 @@
-"""OpenAI, Groq, OpenRouter and this machine, stdlib only.
+"""OpenAI, Groq, OpenRouter, Google AI Studio and this machine, stdlib only.
 
-Transcription runs on any of the four: Groq and OpenRouter both mirror OpenAI's
+Transcription runs on the first three and on this machine: Groq and OpenRouter
-/audio/transcriptions endpoint field for field, and ggml.py starts whisper.cpp
+both mirror OpenAI's /audio/transcriptions endpoint field for field, and ggml.py
-on that same path, so one multipart request serves all of them and only the key,
+starts whisper.cpp on that same path, so one multipart request serves all of
-the base URL and the model id change. llama.cpp answers /chat/completions the way
+them and only the key, the base URL and the model id change. llama.cpp answers
-OpenRouter does, so cleanup here is the same request too.
+/chat/completions the way OpenRouter does, so cleanup here is the same request
+too.
+
+Google AI Studio is here for cleanup and nothing else. Its OpenAI-compatible
+endpoint answers /chat/completions and /models, but there is no
+/audio/transcriptions behind it: audio only goes in as base64 inside a chat
+message, and what comes back has none of the segment times a subtitle file or a
+meeting transcript is built out of.
 
 What is on this machine has no key, and its base URL is not known until a server
 is up, which is the one thing this module has to fill in for it.
@@ -31,6 +38,7 @@ USER_AGENT = f"dikte/1.0 (+{APP_URL})"
 OPENAI_URL = "https://api.openai.com/v1"
 GROQ_URL = "https://api.groq.com/openai/v1"
 OPENROUTER_URL = "https://openrouter.ai/api/v1"
+GEMINI_URL = "https://generativelanguage.googleapis.com/v1beta/openai"
 
 # The floor for a local request. The timeouts elsewhere are sized for a hosted
 # API, where a slow answer is a bill running; here the only thing being spent is
@@ -254,10 +262,23 @@ def _request(url, data, headers, timeout=120, aborter=None):
 
 
 def _extract_error(body):
+    """The line worth showing out of a failed request's body.
+
+    Whatever comes back, this has to end in a string: it is called while an
+    ApiError is being raised, and an exception thrown here would escape the
+    `except ApiError` every caller is holding and lose the dictation the raw
+    transcript would otherwise have been pasted from.
+    """
     try:
         payload = json.loads(body)
     except json.JSONDecodeError:
         return body[:300]
+    if isinstance(payload, list):
+        # Google answers some failures with an array holding the object the
+        # other providers send on its own.
+        payload = next((item for item in payload if isinstance(item, dict)), None)
+    if not isinstance(payload, dict):
+        return body[:300]
     err = payload.get("error")
     if isinstance(err, dict):
         return err.get("message") or json.dumps(err)[:300]
@@ -448,14 +469,24 @@ def transcribe_segments(target, audio_path, language="", prompt="", timeout=300,
     return out
 
 
+# The settings window offers OpenRouter's ladder, and Google has neither end of
+# it: "none" is refused outright with a 400, and there is nothing above "high".
+# Both ends land on the nearest rung that does exist, which costs the cleanup
+# rather than the dictation when it is wrong. "minimal" is where "off" goes, and
+# it is the quickest of them by a wide margin, which is what cleanup wants
+# anyway.
+GEMINI_EFFORT = {"none": "minimal", "xhigh": "high", "max": "high"}
+
+
 def _thinking(payload, provider, reasoning):
     """Ask for as much thinking as this provider understands, or for none.
 
     An empty level means "whatever the model does on its own", so nothing is
-    sent. The two mean opposite things by that, which is why the setting is kept
+    sent. The three mean opposite things by that, which is why the setting is
-    per provider: OpenRouter's cleanup models answer straight away, while a local
+    kept per provider: OpenRouter's cleanup models answer straight away, while a
-    model that was trained to think will think, and cleanup is punctuation rather
+    local model that was trained to think will think, and a Gemini Flash left to
-    than a job worth thinking about.
+    itself thinks too. Cleanup is punctuation rather than a job worth thinking
+    about.
     """
     if not reasoning:
         return
@@ -463,6 +494,13 @@ def _thinking(payload, provider, reasoning):
         # What llama.cpp passes to the chat template. The models that think read
         # it; the ones that do not ignore it.
         payload["chat_template_kwargs"] = {"enable_thinking": reasoning != "none"}
+    elif provider == "gemini":
+        # Google's compatibility layer takes OpenAI's flat field rather than
+        # OpenRouter's object, and it has no word for off, so "none" is asked
+        # for as the lowest rung it has rather than skipped: a Flash model left
+        # to decide for itself thinks, and thinking about a comma is the second
+        # this provider was chosen to save.
+        payload["reasoning_effort"] = GEMINI_EFFORT.get(reasoning, reasoning)
     elif reasoning != "none":
         # The thinking itself is never shown, so ask for it to be left out.
         payload["reasoning"] = {"effort": reasoning, "exclude": True}
@@ -524,15 +562,16 @@ def cleanup(text, api_key, model, system_prompt, reasoning="",
 
 
 def chat(messages, api_key, model, system_prompt, reasoning="",
-         base_url=OPENROUTER_URL, timeout=180):
+         base_url=OPENROUTER_URL, timeout=180, provider="openrouter",
+         service="OpenRouter"):
     """A conversation, rather than one transcript rewritten.
 
     The messages are the whole history and come back unchanged; the caller keeps
-    them, because there is no session on OpenRouter's side to resume.
+    them, because there is no session on the provider's side to resume.
     """
     if not api_key:
         raise ApiError(t("{service} API key is empty. Add it in Settings.",
-                         service="OpenRouter"))
+                         service=service))
     payload = {
         "model": model,
         "messages": [{"role": "system", "content": system_prompt}] + list(messages),
@@ -543,11 +582,11 @@ def chat(messages, api_key, model, system_prompt, reasoning="",
         data = _request(
             f"{base_url.rstrip('/')}/chat/completions",
             json.dumps(payload).encode("utf-8"),
-            _headers("openrouter", api_key, "application/json"),
+            _headers(provider, api_key, "application/json"),
             timeout=timeout,
         )
     except ApiError as exc:
-        raise explain(exc, "OpenRouter") from None
+        raise explain(exc, service) from None
     choices = data.get("choices") or []
     if not choices:
         raise ApiError(_extract_error(json.dumps(data)))
@@ -611,6 +650,37 @@ def openrouter_models(api_key="", transcription=False):
     return sorted(m["id"] for m in models if m.get("id"))
 
 
+# What a `gemini` id can be besides a model that answers a chat request: an
+# embedding, a picture, or a voice. None of them is any use to cleanup.
+NOT_CHAT = ("embedding", "-image", "-tts", "-audio")
+
+
+def gemini_models(api_key, base_url=GEMINI_URL):
+    """The Gemini models Google AI Studio will answer a chat request with.
+
+    Google serves its embedding, image and speech models out of the same list
+    and names them all `gemini` too, so the prefix alone is not the question;
+    none of those can clean up a sentence. The listing has also been known to
+    hand the ids back in their long form, `models/gemini-3.5-flash`, while a
+    request wants the short one; taking the prefix off costs nothing and is
+    right whichever form arrives.
+    """
+    service = "Google AI Studio"
+    if not api_key:
+        raise ApiError(t("{service} API key is empty. Add it in Settings.",
+                         service=service))
+    try:
+        data = _get_json(
+            f"{base_url.rstrip('/')}/models",
+            {"Authorization": f"Bearer {api_key}", "User-Agent": USER_AGENT},
+        )
+    except ApiError as exc:
+        raise explain(exc, service) from None
+    ids = [(m.get("id") or "").removeprefix("models/") for m in data.get("data", [])]
+    return sorted(i for i in ids
+                  if i.startswith("gemini") and not any(w in i for w in NOT_CHAT))
+
+
 def openai_models(api_key, base_url=OPENAI_URL, service="OpenAI"):
     """The audio models of anything that speaks OpenAI's /models, Groq included.
 

dikte/app.py

@@ -163,11 +163,13 @@ class Dikte:
         self.front_before = None
         self._front_watch = None
 
-        self.overlay = Overlay(self.conf["overlay_corner"])
+        self.overlay = Overlay(self.conf["overlay_corner"],
+                               screen_name=self.conf["overlay_screen"])
         # The agent's indicator sits on top of the dictation one when both are
         # up, and drops into the corner when it is alone there.
         self.ask_overlay = Overlay(self.conf["overlay_corner"], below=self.overlay,
-                                   dismissable=True)
+                                   dismissable=True,
+                                   screen_name=self.conf["overlay_screen"])
         self.recorder = audio.Recorder()
         self.pipeline = Pipeline(self.conf)
         self.ask_pipeline = Pipeline(self.conf)
@@ -1296,7 +1298,9 @@ class Dikte:
 
     def _apply_settings(self):
         self.overlay.corner = self.conf["overlay_corner"]
+        self.overlay.screen_name = self.conf["overlay_screen"]
         self.ask_overlay.corner = self.conf["overlay_corner"]
+        self.ask_overlay.screen_name = self.conf["overlay_screen"]
         self._apply_local()
         self._build_tray()
         self._refresh_tray()

dikte/assistant.py

@@ -1,22 +1,29 @@
 """Handing a dictation to an agent as a command, and pasting back its answer.
 
-Three of them, because not everyone has the same one installed:
+Five of them, because not everyone has the same one installed:
 
   Claude Code   `claude -p`, the session you would have opened yourself
   Codex         `codex exec`, the same idea from the other shop
+  Antigravity   `agy -p`, Google's, with a browser of its own attached
   OpenRouter    a plain chat request, over the key that is already configured
+  OpenCode Go   a plain chat request, over a subscription to open coding models
 
-The first two are the whole machine: they run commands, read files, and reach
+The first three are the whole machine: they run commands, read files, and reach
 whatever skills and services you have connected, which is what makes "put that
-in my calendar on Thursday" a thing you can say. OpenRouter cannot touch any of
+in my calendar on Thursday" a thing you can say. The two chat requests cannot
-that, and is there so that a question still gets an answer on a machine with
+touch any of that, and are there so that a question still gets an answer on a
-neither CLI installed.
+machine with no CLI installed at all.
+
+What each of the three is allowed to do without asking is settled where that
+program keeps its own permissions, not here. Dikte hands Claude Code the mode
+chosen in Settings because it has a flag for one; Codex gets a sandbox for the
+same reason; Antigravity has neither, and reads its own allow-rules instead.
 
 Whichever it is, the reply is pasted exactly where the transcript would have
 been, and the conversation carries across dictations so that "and move that to
 Friday" knows what "that" is.
 
-The two CLIs are read as they stream rather than waited out. A command that
+The three CLIs are read as they stream rather than waited out. A command that
 reaches for the calendar or the web takes long enough that a still indicator is
 indistinguishable from a hang, so every tool they pick up is named in the corner
 while they work.
@@ -38,11 +45,16 @@ from . import paths
 from .i18n import t
 
 SESSION_FILE = cfg.DATA_DIR / "assistant.json"
-PROVIDERS = ("claude", "codex", "openrouter")
+PROVIDERS = ("claude", "codex", "agy", "openrouter", "opencode")
 
-# How many messages of an OpenRouter conversation are carried forward. The two
+# What each one is called where a person reads it: the tray, the corner of
-# CLIs keep their own history and need no such number; here every turn is resent
+# the screen, and the line an error is written in.
-# in full, so the window has to end somewhere.
+SERVICES = {"claude": "Claude", "codex": "Codex", "agy": "Antigravity",
+            "openrouter": "OpenRouter", "opencode": "OpenCode Go"}
+
+# How many messages of a chat provider's conversation are carried forward. The
+# two CLIs keep their own history and need no such number; here every turn is
+# resent in full, so the window has to end somewhere.
 MAX_HISTORY = 24
 
 # What to say in the indicator for a tool, keyed by the name the CLI uses.
@@ -70,6 +82,29 @@ CODEX_ITEMS = {
     "patch_apply": "Editing a file…",
     "todo_list": "Planning…",
 }
+# Antigravity carries a browser around with it, so the handful of names below
+# stand in for the couple of dozen browser_* tools it can pick up; being told
+# which mouse button moved is not what the corner of the screen is for.
+AGY_TOOLS = {
+    "run_command": "Running a command…",
+    "command_status": "Running a command…",
+    "send_command_input": "Running a command…",
+    "view_file": "Reading…",
+    "read_url_content": "Reading a web page…",
+    "list_dir": "Looking through files…",
+    "find_by_name": "Looking through files…",
+    "grep_search": "Searching the files…",
+    "search_web": "Searching the web…",
+    "replace_file_content": "Editing a file…",
+    "multi_replace_file_content": "Editing a file…",
+    "sed_file": "Editing a file…",
+    "notebook_edit": "Editing a file…",
+    "write_to_file": "Writing a file…",
+    "generate_image": "Drawing…",
+    "manage_task": "Planning…",
+    "invoke_subagent": "Handing it to a subagent…",
+    "browser_subagent": "Handing it to a subagent…",
+}
 
 
 # How hard to think, in each provider's own vocabulary. The setting is one
@@ -85,6 +120,12 @@ CLAUDE_EFFORT = {"none": "low", "minimal": "low", "low": "low",
 CODEX_EFFORT = {"none": "low", "minimal": "low", "low": "low",
                 "medium": "medium", "high": "high", "xhigh": "high",
                 "max": "high"}
+# agy has three rungs and no word for off, so the bottom of the ladder lands on
+# "low" and the top two on "high". Shared with cleanup, which runs the same
+# program for the smaller job.
+AGY_EFFORT = {"none": "low", "minimal": "low", "low": "low",
+              "medium": "medium", "high": "high", "xhigh": "high",
+              "max": "high"}
 
 
 class AssistantError(Exception):
@@ -102,12 +143,30 @@ def provider(conf):
 
 def executable(name):
     """The CLI a provider runs, or "" when it needs none."""
-    return {"claude": "claude", "codex": "codex"}.get(name, "")
+    return {"claude": "claude", "codex": "codex", "agy": "agy"}.get(name, "")
+
+
+def model(conf):
+    """Which model answered, for the history to record.
+
+    Each provider keeps its own setting, and the one a CLI is left on has no id
+    to report, only a name — the same arrangement cleanup.model() makes.
+    """
+    name = provider(conf)
+    if name == "codex":
+        return conf["assistant_codex_model"].strip() or "codex"
+    if name == "agy":
+        return conf["assistant_agy_model"].strip() or "agy"
+    if name == "openrouter":
+        return conf["assistant_openrouter_model"]
+    if name == "opencode":
+        return conf["assistant_opencode_model"]
+    return conf["assistant_model"]
 
 
 def display_name(conf):
     """What to call the thing being asked, in the tray and in the corner."""
-    return {"claude": "Claude", "codex": "Codex"}.get(provider(conf), "OpenRouter")
+    return SERVICES.get(provider(conf), "OpenRouter")
 
 
 # --- the conversation -----------------------------------------------------
@@ -197,8 +256,8 @@ def ask(prompt, conf, on_stage=None, should_stop=None):
     one, and only the denial explains why it did not do what it was asked to.
     """
     name = provider(conf)
-    if name == "openrouter":
+    if name in ("openrouter", "opencode"):
-        return _ask_openrouter(prompt, conf, on_stage)
+        return _ask_chat(name, SERVICES[name], prompt, conf, on_stage)
 
     binary = executable(name)
     if not shutil.which(binary):
@@ -207,7 +266,7 @@ def ask(prompt, conf, on_stage=None, should_stop=None):
             "Settings → Agent.", binary=binary,
         ))
 
-    run = _ask_claude if name == "claude" else _ask_codex
+    run = {"claude": _ask_claude, "codex": _ask_codex, "agy": _ask_agy}[name]
     session = read_session(name, conf["assistant_session_minutes"] * 60)
     try:
         return run(prompt, conf, session, on_stage, should_stop)
@@ -259,7 +318,7 @@ def _ask_claude(prompt, conf, session, on_stage, should_stop):
             found["warning"] = _denial_warning(event)
 
     code, stderr = _stream(cmd, conf, on_event, should_stop)
-    return _conclude(found, code, stderr, session, "Claude")
+    return _conclude(found, code, stderr, session, "claude")
 
 
 def _claude_label(block):
@@ -327,7 +386,7 @@ def _ask_codex(prompt, conf, session, on_stage, should_stop):
                                 else str(error)) or t("Codex ended with an error.")
 
     code, stderr = _stream(cmd, conf, on_event, should_stop)
-    return _conclude(found, code, stderr, session, "Codex")
+    return _conclude(found, code, stderr, session, "codex")
 
 
 def _codex_label(item):
@@ -366,29 +425,125 @@ def codex_models():
     return [row["slug"] for row in rows]
 
 
-# --- OpenRouter -----------------------------------------------------------
+# --- Antigravity ----------------------------------------------------------
 
-def _ask_openrouter(prompt, conf, on_stage):
+def _ask_agy(prompt, conf, session, on_stage, should_stop):
-    """No tools, no files, no calendar: a question and an answer.
+    # Antigravity takes no system prompt of its own either, so the instruction
+    # rides in front of the command, kept apart from it so the two are not read
+    # as one.
+    body = f"{conf.assistant_prompt()}\n\n---\n\n{prompt}"
+    cmd = [
+        "agy", "-p", body,
+        "--output-format", "stream-json",
+        # agy stops after five minutes unless it is told otherwise, which is
+        # shorter than the timeout this setting offers.
+        "--print-timeout", f"{conf['assistant_timeout']}s",
+    ]
+    # One or the other, always: left with neither, agy picks up whichever
+    # project it was last in and works in that project's directory rather than
+    # the one _stream is about to start it in.
+    cmd += ["--conversation", session] if session else ["--new-project"]
+    if conf["assistant_agy_model"].strip():
+        cmd += ["--model", conf["assistant_agy_model"].strip()]
+    effort = AGY_EFFORT.get(conf["assistant_reasoning"], "")
+    if effort:
+        # Most of agy's own model ids carry the effort in their suffix already;
+        # this is for the ones that do not.
+        cmd += ["--effort", effort]
 
-    It is the fallback for a machine with neither CLI on it, so it says what it
+    found = {"answer": "", "warning": "", "session": "", "failure": ""}
-    knows and nothing else. The conversation is ours to keep here, since there
+
-    is no session on the other end to resume.
+    def on_event(event):
+        kind = event.get("event")
+        if kind == "init":
+            found["session"] = event.get("conversation_id") or found["session"]
+        elif kind == "step_update":
+            step = event.get("step_update") or {}
+            # A tool is reported twice, once when it starts and once when it is
+            # done; the corner wants the first of those.
+            if (on_stage and step.get("step_type") == "tool"
+                    and step.get("state") == "ACTIVE"):
+                on_stage(_agy_label(step))
+        elif kind == "result":
+            result = event.get("result") or {}
+            found["session"] = result.get("conversation_id") or found["session"]
+            answer = (result.get("response") or "").strip()
+            if result.get("status") == "SUCCESS":
+                found["answer"] = answer
+            else:
+                found["failure"] = answer or t("{service} ended with an error.",
+                                               service="Antigravity")
+
+    code, stderr = _stream(cmd, conf, on_event, should_stop)
+    return _conclude(found, code, stderr, session, "agy")
+
+
+def _agy_label(step):
+    name = step.get("tool_name", "")
+    if name in AGY_TOOLS:
+        return t(AGY_TOOLS[name])
+    if name.startswith("browser_") or name.startswith("capture_browser"):
+        return t("Working in the browser…")
+    if name == "call_mcp_tool":
+        server = (step.get("tool_info") or {}).get("parameters") or {}
+        return t("Using {name}…", name=server.get("server") or "a tool")
+    return t("Using {name}…", name=name or "a tool")
+
+
+def agy_models():
+    """The models Antigravity itself would offer right now, in its own order.
+
+    `agy models` prints one `id<TAB>display name` line per model, so the list
+    is as current as the account behind the CLI. Unlike Codex it asks Google
+    rather than a cache on disk, a couple of seconds the caller spends off the
+    interface thread. A machine without agy, or a call that fails, answers
+    with nothing and the caller keeps its built-in list.
+    """
+    if not shutil.which("agy"):
+        return []
+    try:
+        proc = subprocess.run(["agy", "models"],
+                              capture_output=True, text=True, timeout=30)
+    except (OSError, subprocess.SubprocessError):
+        return []
+    if proc.returncode != 0:
+        return []
+    ids = []
+    for line in (proc.stdout or "").splitlines():
+        model_id, tab, _ = line.partition("\t")
+        if tab and model_id.strip():
+            ids.append(model_id.strip())
+    return ids
+
+
+# --- OpenRouter and OpenCode Go -------------------------------------------
+
+def _ask_chat(name, service, prompt, conf, on_stage):
+    """A plain question and answer, over a chat provider's key.
+
+    No tools, no files, no calendar. It is the fallback for a machine with
+    neither CLI on it, so it says what it knows and nothing else. The
+    conversation is ours to keep here, since there is no session on the other
+    end to resume.
     """
     if on_stage:
         on_stage(t("Thinking…"))
-    history = read_messages("openrouter", conf["assistant_session_minutes"] * 60)
+    history = read_messages(name, conf["assistant_session_minutes"] * 60)
     messages = history + [{"role": "user", "content": prompt}]
+    model = (conf["assistant_openrouter_model"] if name == "openrouter"
+             else conf["assistant_opencode_model"])
+    base_url = (conf["openrouter_base_url"] if name == "openrouter"
+                else conf["opencode_base_url"])
+    key = conf.openrouter_key() if name == "openrouter" else conf.opencode_key()
     try:
         answer = api.chat(
-            messages, conf.openrouter_key(), conf["assistant_openrouter_model"],
+            messages, key, model, conf.assistant_prompt(),
-            conf.assistant_prompt(), reasoning=conf["assistant_reasoning"],
+            reasoning=conf["assistant_reasoning"], base_url=base_url,
-            base_url=conf["openrouter_base_url"],
+            timeout=conf["assistant_timeout"], provider=name, service=service,
-            timeout=conf["assistant_timeout"],
         )
     except api.ApiError as exc:
         raise AssistantError(str(exc)) from exc
-    write_session("openrouter",
+    write_session(name,
                   messages=messages + [{"role": "assistant", "content": answer}])
     return answer, ""
 
@@ -469,8 +624,9 @@ _API_TROUBLE = re.compile(
     r"\b(401|403|429|5\d\d)\b")
 
 
-def _conclude(found, code, stderr, session, service):
+def _conclude(found, code, stderr, session, name):
     """Turn what the stream said into an answer, or into the reason there is none."""
+    service = SERVICES.get(name, name)
     if code != 0 and not found["answer"]:
         # A resumed run that died with nothing to show is treated as the
         # session being gone, whatever the wording: this code used to look for
@@ -489,7 +645,7 @@ def _conclude(found, code, stderr, session, service):
     if not found["answer"]:
         raise AssistantError(t("{service} answered with nothing.", service=service))
     if found["session"]:
-        write_session("claude" if service == "Claude" else "codex", found["session"])
+        write_session(name, found["session"])
     return found["answer"], found["warning"]
 
 

dikte/cleanup.py

@@ -1,7 +1,8 @@
 """Who rewrites the transcript once it has been heard.
 
-Normally a small model on OpenRouter: one request, a second, a few tenths of a
+Normally a small model over one HTTP request: a second, and a few tenths of a
-cent. A machine with Claude Code or Codex on it is already paying for a model
+cent on OpenRouter or nothing at all on Google AI Studio's free tier. A machine
+with Claude Code, Codex or Antigravity on it is already paying for a model
 though, and the subscription that answers "put that in my calendar on Thursday"
 can just as well take the "eee"s out of a sentence. No second key, no second
 bill. It costs seconds rather than one, because a CLI opens a whole session to
@@ -10,7 +11,11 @@ do it, which is the trade.
 Whoever does it, the job is the same one: no tools, no files, no memory of the
 last dictation. There is nothing here to look up and nothing to carry over, and
 a transcript is text from a microphone rather than an instruction, so the less
-the agent can reach while it reads one, the better.
+the agent can reach while it reads one, the better. Claude Code is handed an
+empty tool list and Codex a read-only sandbox. Antigravity has neither switch,
+and this is worth saying plainly rather than implying parity: there the
+transcript is read by an agent that could go and do something. What can be done
+is done — a project of its own, the home directory, and its slash commands off.
 """
 
 import os
@@ -24,7 +29,7 @@ from . import ggml
 from . import paths
 from .i18n import t
 
-PROVIDERS = ("openrouter", "local", "claude", "codex")
+PROVIDERS = ("openrouter", "gemini", "opencode", "local", "claude", "codex", "agy")
 
 
 class CleanupError(api.ApiError):
@@ -43,7 +48,7 @@ def provider(conf):
 
 def executable(name):
     """The CLI a provider runs, or "" when it needs none."""
-    return {"claude": "claude", "codex": "codex"}.get(name, "")
+    return {"claude": "claude", "codex": "codex", "agy": "agy"}.get(name, "")
 
 
 def model(conf):
@@ -57,13 +62,20 @@ def model(conf):
         # Codex is left on whatever it is set to unless a model is typed in, so
         # here there is only the name of the thing that did it.
         return conf["cleanup_codex_model"].strip() or "codex"
+    if name == "agy":
+        # The same arrangement as Codex, and the same reason for it.
+        return conf["cleanup_agy_model"].strip() or "agy"
+    if name == "gemini":
+        return conf["cleanup_gemini_model"]
+    if name == "opencode":
+        return conf["cleanup_opencode_model"]
     return conf["cleanup_model"]
 
 
 def run(text, conf, system_prompt, timeout=180, aborter=None):
     """Hand the transcript to whoever is set to clean it up.
 
-    `aborter` is only of use to the two that answer over HTTP; a CLI is stopped
+    `aborter` is only of use to the three that answer over HTTP; a CLI is stopped
     between blocks instead, which is close enough when a block is seconds.
     """
     name = provider(conf)
@@ -74,9 +86,23 @@ def run(text, conf, system_prompt, timeout=180, aborter=None):
             base_url=conf["openrouter_base_url"], timeout=timeout,
             aborter=aborter,
         )
+    if name == "gemini":
+        return api.cleanup(
+            text, conf.gemini_key(), conf["cleanup_gemini_model"], system_prompt,
+            reasoning=conf["cleanup_reasoning"],
+            base_url=conf["gemini_base_url"], timeout=timeout,
+            provider="gemini", service="Google AI Studio", aborter=aborter,
+        )
+    if name == "opencode":
+        return api.cleanup(
+            text, conf.opencode_key(), conf["cleanup_opencode_model"], system_prompt,
+            reasoning=conf["cleanup_reasoning"],
+            base_url=conf["opencode_base_url"], timeout=timeout,
+            provider="opencode", service="OpenCode Go", aborter=aborter,
+        )
     if name == "local":
         return _local(text, conf, system_prompt, timeout, aborter)
-    runner = _claude if name == "claude" else _codex
+    runner = {"claude": _claude, "codex": _codex, "agy": _agy}[name]
     return runner(text, conf, system_prompt, timeout)
 
 
@@ -172,6 +198,40 @@ def _codex(text, conf, system_prompt, timeout):
     return answer
 
 
+# --- Antigravity ----------------------------------------------------------
+
+def _agy(text, conf, system_prompt, timeout):
+    # Antigravity takes no system prompt of its own either, so the rules ride in
+    # front of the transcript, kept apart from it so the two are not read as one.
+    body = f"{system_prompt}\n\n---\n\n{_wrap(text)}"
+    cmd = [
+        "agy", "-p", body,
+        "--output-format", "text",       # the answer, and nothing around it
+        # Left to itself agy picks up whichever project it was last in and works
+        # in that project's directory rather than this one. A dictation belongs
+        # to no project, so each one starts on a project of its own.
+        "--new-project",
+        # A transcript that happens to begin with a slash is still a transcript.
+        "--disable-slash-commands",
+        # agy gives up after five minutes of its own accord, which would have it
+        # killed from outside rather than answering.
+        "--print-timeout", f"{timeout}s",
+    ]
+    if conf["cleanup_agy_model"].strip():
+        cmd += ["--model", conf["cleanup_agy_model"].strip()]
+    effort = assistant.AGY_EFFORT.get(conf["cleanup_reasoning"], "")
+    if effort:
+        # agy's own model ids carry the effort in their suffix, so this only
+        # matters for the ones that do not, and for a model typed in by hand.
+        cmd += ["--effort", effort]
+
+    answer = _output(cmd, timeout, "Antigravity")
+    if not answer:
+        raise CleanupError(t("{service} answered with nothing.",
+                             service="Antigravity"))
+    return answer
+
+
 def _read(path):
     try:
         with open(path, encoding="utf-8", errors="replace") as fh:

dikte/cli.py

@@ -216,7 +216,9 @@ def cmd_ask(opts):
         conf["assistant_provider"] = opts.provider
     if opts.model:
         key = {"claude": "assistant_model", "codex": "assistant_codex_model",
-               "openrouter": "assistant_openrouter_model"}[assistant.provider(conf)]
+               "openrouter": "assistant_openrouter_model",
+               "agy": "assistant_agy_model",
+               "opencode": "assistant_opencode_model"}[assistant.provider(conf)]
         conf[key] = opts.model
     if opts.dir:
         conf["assistant_dir"] = opts.dir
@@ -255,7 +257,8 @@ def cmd_ask(opts):
         "cleanup_error": warning,
         "mode": "ask",
         "question": text,
-        "assistant_model": conf["assistant_model"],
+        "assistant": assistant.provider(conf),
+        "assistant_model": assistant.model(conf),
         "raw": text,
         "text": answer,
     })
@@ -508,7 +511,8 @@ def cmd_history_clear(opts):
 
 # --- settings ---------------------------------------------------------------
 
-SECRET_KEYS = ("openai_api_key", "groq_api_key", "openrouter_api_key")
+SECRET_KEYS = ("openai_api_key", "groq_api_key", "openrouter_api_key",
+               "gemini_api_key", "opencode_api_key")
 
 
 def _mask(key, value):
@@ -698,6 +702,22 @@ def cmd_test_key(opts):
             results[name] = {"ok": True, "message": message}
         except api.ApiError as exc:
             results[name] = {"ok": False, "message": str(exc)}
+    if opts.which in ("gemini", "all"):
+        try:
+            count = len(api.gemini_models(conf.gemini_key(),
+                                          conf["gemini_base_url"]))
+            message = f"connection works, {count} models visible"
+            results["gemini"] = {"ok": True, "message": message}
+        except api.ApiError as exc:
+            results["gemini"] = {"ok": False, "message": str(exc)}
+    if opts.which in ("opencode", "all"):
+        try:
+            count = len(api.openai_models(conf.opencode_key(),
+                                          conf["opencode_base_url"], "OpenCode Go"))
+            results["opencode"] = {"ok": True,
+                                   "message": f"connection works, {count} models visible"}
+        except api.ApiError as exc:
+            results["opencode"] = {"ok": False, "message": str(exc)}
     everything_ok = all(item["ok"] for item in results.values())
     lines = [f"{'✓' if item['ok'] else '✗'} {name}: {item['message']}"
              for name, item in results.items()]
@@ -862,8 +882,17 @@ def cmd_doctor(opts):
     # and marking them by the key they do not use reported every fully local
     # setup as broken.
     transcribe_ready = conf.transcribe_ready()
-    if cleaner == "openrouter":
+    # Only the ones that answer over HTTP have a key worth looking at. A CLI
-        cleanup_ready = bool(conf.openrouter_key())
+    # has a program to find instead, and the model on this machine has neither,
+    # so "no key" there has to read as beside the point rather than as one that
+    # has gone missing.
+    cleanup_service, cleanup_key = {
+        "openrouter": ("OpenRouter", conf.openrouter_key()),
+        "gemini": ("Google AI Studio", conf.gemini_key()),
+        "opencode": ("OpenCode Go", conf.opencode_key()),
+    }.get(cleaner, ("", ""))
+    if cleanup_service:
+        cleanup_ready = bool(cleanup_key)
     elif cleaner == "local":
         cleanup_ready = conf.local_llm_ready()
     else:
@@ -875,7 +904,7 @@ def cmd_doctor(opts):
                           "ready": transcribe_ready},
         "cleanup": {"enabled": conf["cleanup_enabled"], "provider": cleaner,
                     "model": cleanup.model(conf),
-                    "key": bool(conf.openrouter_key()),
+                    "key": bool(cleanup_key) if cleanup_service else None,
                     "ready": cleanup_ready},
         "agent": {"provider": assistant.provider(conf),
                   "directory": assistant.working_dir(conf)},
@@ -887,9 +916,9 @@ def cmd_doctor(opts):
     else:
         transcribe_line = (f"{'✓' if transcribe_ready else '✗'} {target.service} "
                            f"key, transcribing on {target.model}")
-    if cleaner == "openrouter":
+    if cleanup_service:
-        cleanup_line = (f"{'✓' if cleanup_ready else '✗'} OpenRouter key, "
+        cleanup_line = (f"{'✓' if cleanup_ready else '✗'} {cleanup_service} key, "
-                        f"cleaning up on {conf['cleanup_model']}")
+                        f"cleaning up on {cleanup.model(conf)}")
     elif cleaner == "local":
         cleanup_line = (f"{'✓' if cleanup_ready else '✗'} Local model, "
                         f"cleaning up on {conf['local_llm_model'] or 'no model yet'}")
@@ -990,7 +1019,7 @@ def build_parser():
     ask = leaf(subs, "ask", "put a command to the agent")
     ask.add_argument("text", nargs="*", help="the command; read from stdin, or "
                                              "recorded when there is none")
-    ask.add_argument("--provider", choices=("claude", "codex", "openrouter"),
+    ask.add_argument("--provider", choices=assistant.PROVIDERS,
                      help="just for this run")
     ask.add_argument("--model", help="just for this run")
     ask.add_argument("--dir", help="working directory, just for this run")
@@ -1116,7 +1145,7 @@ def build_parser():
     models.set_defaults(func=cmd_models)
     test = leaf(subs, "test-key", "check the API keys")
     test.add_argument("which", nargs="?", default="all",
-                      choices=("all", *cfg.TRANSCRIBERS))
+                      choices=("all", *cfg.TRANSCRIBERS, "gemini", "opencode"))
     test.set_defaults(func=cmd_test_key)
     leaf(subs, "doctor", "keys, programs, and what is missing").set_defaults(func=cmd_doctor)
 

dikte/config.py

@@ -387,6 +387,12 @@ DEFAULTS = {
     "groq_base_url": "https://api.groq.com/openai/v1",
     "openrouter_api_key": "",
     "openrouter_base_url": "https://openrouter.ai/api/v1",
+    "gemini_api_key": "",
+    # Google's OpenAI-compatible endpoint. Cleanup only: there is no
+    # /audio/transcriptions behind it, so it is not one of the TRANSCRIBERS.
+    "gemini_base_url": "https://generativelanguage.googleapis.com/v1beta/openai",
+    "opencode_api_key": "",
+    "opencode_base_url": "https://opencode.ai/zen/go/v1",
     "transcribe_provider": "local",  # "local", or a key of TRANSCRIBERS
     "transcribe_model": "gpt-4o-transcribe",           # used when provider is openai
     "groq_transcribe_model": "whisper-large-v3-turbo",
@@ -412,6 +418,9 @@ DEFAULTS = {
     "cleanup_model": "google/gemini-3.5-flash-lite",
     "cleanup_claude_model": "haiku",   # Claude Code: an alias, or a full model id
     "cleanup_codex_model": "",         # empty -> whatever Codex is set to
+    "cleanup_gemini_model": "gemini-3.5-flash-lite",
+    "cleanup_agy_model": "",           # empty -> whatever Antigravity is set to
+    "cleanup_opencode_model": "deepseek-v4-flash",
     "cleanup_reasoning": "",        # empty -> whatever the model does by default
 
     # --- llama.cpp, on this machine -----------------------------------------
@@ -457,6 +466,7 @@ DEFAULTS = {
     "pause_shortcut": "",
     "evdev_hotkey": False,
     "overlay_corner": "bottom-left",
+    "overlay_screen": "",
     "keep_audio": False,
     "history_limit": 200,
     # A look at the releases page once a day, and nothing more than a look:
@@ -484,12 +494,14 @@ DEFAULTS = {
 
     # --- speaking a command to an agent -------------------------------------
     "assistant_shortcut": "",       # empty -> tray only
-    "assistant_provider": "claude",  # claude | codex | openrouter
+    "assistant_provider": "claude",  # claude | codex | agy | openrouter
     "assistant_model": "sonnet",    # Claude Code: an alias, or a full model id
     "assistant_permission_mode": "auto",
     "assistant_codex_model": "",    # empty -> whatever Codex is set to
     "assistant_codex_sandbox": "workspace-write",
     "assistant_openrouter_model": "google/gemini-3.5-flash",
+    "assistant_agy_model": "",      # empty -> whatever Antigravity is set to
+    "assistant_opencode_model": "deepseek-v4-flash",
     "assistant_reasoning": "",      # empty -> the model's own default
     "assistant_dir": "",            # empty -> the home directory
     "assistant_prompt": "",         # empty -> language-specific default
@@ -632,6 +644,12 @@ class Config:
     def openrouter_key(self):
         return self.api_key("openrouter_api_key")
 
+    def gemini_key(self):
+        return self.api_key("gemini_api_key")
+
+    def opencode_key(self):
+        return self.api_key("opencode_api_key")
+
     def transcribe_target(self):
         """Key, endpoint and model for whichever provider does speech to text.
 

dikte/ggml.py

@@ -83,6 +83,10 @@ LLAMA = Program("llama", "ggml-org/llama.cpp", "llama-server", "/health")
 WHISPER_MODELS_REPO = "ggerganov/whisper.cpp"
 LLM_AUTHOR = "ggml-org"
 
+# The file llama.cpp attaches to its version releases in place of the binaries:
+# a line naming the nightly tag those are published under.
+NIGHTLY_TAG = "nightly-tag.txt"
+
 # What the whisper repository holds besides models: Core ML encoders for Apple
 # hardware and the odd loose file.
 WHISPER_PREFIX = "ggml-"
@@ -277,6 +281,65 @@ def _wanted_assets(program):
     return (f"bin-ubuntu-{arch}.tar.gz",)
 
 
+def _matching_asset(program, assets):
+    """The archive this machine wants out of one release's files, or None."""
+    for ending in _wanted_assets(program):
+        item = next((a for a in assets if a.name.endswith(ending)), None)
+        if item:
+            return item
+    return None
+
+
+def _pick_asset(program, tag="", refresh=False):
+    """(tag, Item) for the release archive to install. Item is None when there
+    is none for this machine.
+
+    A named tag is taken as given. For the newest, what GitHub answers is not
+    always where the builds are: llama.cpp's latest release is a version marker
+    carrying a single nightly-tag.txt, which names the tag the archives are
+    actually attached to, and those are prereleases that "latest" never points
+    at. The pointer is followed when it is there, and when it is not, the newest
+    release that does carry a build for this machine is taken instead.
+    """
+    named = bool(tag) and tag != "latest"
+    missing = None
+    try:
+        tag, assets = hub.release(program.repo, tag or "latest", refresh=refresh)
+    except hub.HubError as exc:
+        # A release carrying no files at all is the case the search below exists
+        # for, not a reason to stop before it: the build for this machine may be
+        # attached to a prerelease that "latest" never points at. The failure is
+        # kept rather than dropped, because an unreachable GitHub arrives here
+        # the same way and that one is the message the caller wants.
+        if named:
+            raise
+        missing, assets = exc, []
+    item = _matching_asset(program, assets)
+    if item or named:
+        return tag, item
+    # Best effort from here on: a machine this project publishes nothing for is
+    # not a failed lookup, and the caller's message about that is the useful
+    # one. Whatever goes wrong while looking further leaves it standing.
+    try:
+        pointer = next((a for a in assets if a.name == NIGHTLY_TAG), None)
+        if pointer:
+            nightly = hub.text(pointer.url).strip()
+            if nightly:
+                found, assets = hub.release(program.repo, nightly, refresh=refresh)
+                item = _matching_asset(program, assets)
+                if item:
+                    return found, item
+        for found, assets in hub.releases(program.repo, refresh=refresh):
+            item = _matching_asset(program, assets)
+            if item:
+                return found, item
+    except hub.HubError:
+        pass
+    if missing is not None:
+        raise missing
+    return tag, None
+
+
 def _install_record(program):
     return BIN_DIR / program.name / "installed.json"
 
@@ -375,15 +438,10 @@ def install_program(program, tag="", on_progress=None, should_stop=None,
     whisper.cpp has one.
     """
     try:
-        tag, assets = hub.release(program.repo, tag or "latest", refresh=refresh)
+        tag, item = _pick_asset(program, tag, refresh=refresh)
     except hub.HubError as exc:
         raise LocalError(str(exc)) from exc
 
-    item = None
-    for ending in _wanted_assets(program):
-        item = next((a for a in assets if a.name.endswith(ending)), None)
-        if item:
-            break
     if item is None:
         # Nothing to download and nothing to install for you: whisper.cpp
         # publishes no macOS binary, and Homebrew's whisper-cpp is configured

dikte/hub.py

@@ -121,6 +121,13 @@ def _digest(value):
     return value.split(":", 1)[1] if value.startswith("sha256:") else value
 
 
+def _assets(data):
+    return [Item(a.get("name") or "", a.get("browser_download_url") or "",
+                 int(a.get("size") or 0), _digest(a.get("digest")))
+            for a in (data.get("assets") or [])
+            if a.get("browser_download_url")]
+
+
 def release(repo, tag="latest", refresh=False):
     """(tag, [Item]) for one GitHub release, newest when no tag is given."""
     where = "latest" if tag in ("", "latest") else f"tags/{tag}"
@@ -128,10 +135,47 @@ def release(repo, tag="latest", refresh=False):
                   f"{GITHUB_API}/repos/{repo}/releases/{where}", refresh=refresh)
     if not isinstance(data, dict) or not data.get("assets"):
         raise HubError(t("{repo} has no downloadable release.", repo=repo))
-    assets = [Item(a.get("name") or "", a.get("browser_download_url") or "",
+    return data.get("tag_name") or tag, _assets(data)
-                   int(a.get("size") or 0), _digest(a.get("digest")))
+
-              for a in data["assets"] if a.get("browser_download_url")]
+
-    return data.get("tag_name") or tag, assets
+def releases(repo, limit=20, refresh=False):
+    """[(tag, [Item])] for the recent releases, newest first, with their files.
+
+    "latest" is one release and this is the list behind it, prereleases
+    included: a project that attaches its builds to a prerelease is invisible
+    to release() above, and its newest usable build is in here.
+    """
+    data = _fetch(f"gh-list-{repo}-{limit}",
+                  f"{GITHUB_API}/repos/{repo}/releases?per_page={limit}",
+                  refresh=refresh)
+    if not isinstance(data, list):
+        raise HubError(t("{repo} has no downloadable release.", repo=repo))
+    out = []
+    for entry in data:
+        tag, items = entry.get("tag_name") or "", _assets(entry)
+        if tag and items:
+            out.append((tag, items))
+    return out
+
+
+def text(url, limit=4096, timeout=20):
+    """A small text file from a release, as a string.
+
+    Not cached and not checksummed, because what it carries is a pointer: a few
+    bytes naming the release the actual archives are attached to, read once on
+    the way to a download that is checked in full.
+    """
+    request = urllib.request.Request(url, headers={"User-Agent": USER_AGENT})
+    try:
+        with urllib.request.urlopen(request, timeout=timeout) as response:
+            return response.read(limit).decode("utf-8", "replace")
+    except urllib.error.HTTPError as exc:
+        exc.close()
+        raise HubError(t("{url} answered HTTP {code}.",
+                         url=urllib.parse.urlsplit(url).netloc, code=exc.code)) from exc
+    except (urllib.error.URLError, OSError, ValueError) as exc:
+        raise HubError(t("Could not reach {url}: {error}",
+                         url=urllib.parse.urlsplit(url).netloc, error=exc)) from exc
 
 
 def newest_release(repo, refresh=False):

dikte/i18n.py

@@ -40,8 +40,16 @@ def t(text, /, **kwargs):
 # by the sentence, so it arrives already inflected. English takes the name as it
 # is and puts the preposition in the sentence, where it belongs.
 _TR_CASES = {
-    "dative": {"Claude": "Claude'a", "Codex": "Codex'e", "OpenRouter": "OpenRouter'a"},
+    "dative": {
-    "accusative": {"Claude": "Claude'u", "Codex": "Codex'i", "OpenRouter": "OpenRouter'ı"},
+        "Claude": "Claude'a", "Codex": "Codex'e", "OpenRouter": "OpenRouter'a",
+        "Google AI Studio": "Google AI Studio'ya", "Antigravity": "Antigravity'ye",
+        "OpenCode Go": "OpenCode Go'ya",
+    },
+    "accusative": {
+        "Claude": "Claude'u", "Codex": "Codex'i", "OpenRouter": "OpenRouter'ı",
+        "Google AI Studio": "Google AI Studio'yu", "Antigravity": "Antigravity'yi",
+        "OpenCode Go": "OpenCode Go'yu",
+    },
 }
 
 
@@ -151,6 +159,7 @@ TR = {
     # --- settings: tabs and general ------------------------------------
     "Dikte Settings": "Dikte Ayarları",
     "General": "Genel",
+    "Display": "Ekran",
     "API and models": "API ve modeller",
     "Cleanup rules": "Temizleme kuralları",
     "Audio file": "Ses dosyası",
@@ -179,6 +188,9 @@ TR = {
         "macOS bu ilk gönderildiğinde Erişilebilirlik izni ister.",
     "Restore the previous clipboard after pasting":
         "Yapıştırdıktan sonra eski pano içeriğini geri koy",
+    "Indicator screen": "Gösterge ekranı",
+    "Follow the mouse pointer": "Fare imlecini takip et",
+    "{name} (not connected)": "{name} (bağlı değil)",
     "Indicator corner": "Gösterge köşesi",
     "bottom-left": "sol-alt",
     "bottom-right": "sağ-alt",
@@ -220,26 +232,25 @@ TR = {
     "sk-… (falls back to OPENAI_API_KEY)": "sk-… (boşsa OPENAI_API_KEY kullanılır)",
     "gsk_… (falls back to GROQ_API_KEY)": "gsk_… (boşsa GROQ_API_KEY kullanılır)",
     "sk-or-… (falls back to OPENROUTER_API_KEY)": "sk-or-… (boşsa OPENROUTER_API_KEY kullanılır)",
+    "(falls back to GEMINI_API_KEY)": "(boşsa GEMINI_API_KEY kullanılır)",
+    "(falls back to OPENCODE_API_KEY)": "(boşsa OPENCODE_API_KEY kullanılır)",
     "Test": "Test et",
     "Trying…": "Deneniyor…",
     "Runs on OpenRouter.": "OpenRouter üzerinde çalışır.",
+    "Runs on Google AI Studio.": "Google AI Studio üzerinde çalışır.",
+    "Runs on OpenCode Go.": "OpenCode Go üzerinde çalışır.",
     "Connection works. {count} audio models visible.":
         "Bağlantı tamam. {count} ses modeli görünüyor.",
+    "Connection works. {count} models visible.":
+        "Bağlantı tamam. {count} model görünüyor.",
     "Clean the transcript with a model": "Transkripti bir modelle temizle",
-    "OpenRouter is the quickest and the only one that needs nothing installed. "
-    "Claude Code and Codex clean up on the subscription you already have, "
-    "without a second key, and take a few seconds longer because each one opens "
-    "a session to do it.":
-        "En hızlısı OpenRouter'dır ve kurulu bir program istemeyen tek seçenektir. "
-        "Claude Code ile Codex, temizliği hâlihazırda ödediğin abonelik üzerinden "
-        "yapar, ikinci bir anahtar istemez; her biri bunun için bir oturum açtığından "
-        "birkaç saniye daha uzun sürer.",
     "{binary} is not on your PATH, so cleanup would fail and the raw transcript "
     "would be pasted. Install it, or pick another one above.":
         "{binary} PATH'te değil; temizleme başarısız olur ve ham transkript "
         "yapıştırılır. Kur ya da yukarıdan başka birini seç.",
     "Thinking": "Düşünme",
     "Model's own default": "Modelin kendi varsayılanı",
+    "Antigravity's own default": "Antigravity'nin kendi varsayılanı",
     "Off": "Kapalı",
     "Minimal": "En az",
     "Low": "Düşük",
@@ -502,18 +513,6 @@ TR = {
 
     # --- settings: the agent ------------------------------------------------
     "Agent": "Ajan",
-    "This shortcut records the same way dictation does, but the transcript is "
-    "not what gets pasted. It goes to an agent as a command, and what comes "
-    "back is pasted instead: the answer to a question, or a sentence saying "
-    "what was done. Claude Code and Codex run as the session you would have "
-    "opened yourself, with your skills, your connected services and your "
-    "account.":
-        "Bu kısayol dikte ile aynı şekilde kaydeder, ama yapıştırılan şey "
-        "transkript değildir. Transkript bir ajana komut olarak gider ve yerine "
-        "oradan döneni yapıştırılır: bir sorunun cevabı ya da ne yapıldığını "
-        "söyleyen bir cümle. Claude Code ve Codex, kendi açacağın oturumun "
-        "aynısı olarak çalışır: skill'lerinle, bağlı servislerinle ve kendi "
-        "hesabınla.",
     "How it runs": "Nasıl çalışıyor",
     "Runs on": "Şunun üstünde çalışır",
     "More thinking is slower, and you are standing in front of the screen while "
@@ -540,6 +539,24 @@ TR = {
         "Yukarıdaki çalışma dizini ve izinler burada bir şey ifade etmez.",
     "Needs no program installed, only the OpenRouter key.":
         "Kurulu bir programa değil, yalnızca OpenRouter anahtarına ihtiyaç duyar.",
+    "A plain question and a plain answer, over the OpenCode Go key you already "
+    "have. It runs no commands, opens no files and reaches none of your "
+    "services, so it can tell you what the capital of Peru is but not what is "
+    "in your calendar. Working directory and permissions above mean nothing "
+    "here.":
+        "Elindeki OpenCode Go anahtarı üzerinden düz bir soru ve düz bir cevap. "
+        "Komut çalıştırmaz, dosya açmaz, servislerinin hiçbirine erişmez; yani "
+        "Peru'nun başkentini söyler ama takviminde ne olduğunu söyleyemez. "
+        "Yukarıdaki çalışma dizini ve izinler burada bir şey ifade etmez.",
+    "Needs no program installed, only an OpenCode Go key.":
+        "Kurulu bir programa değil, yalnızca bir OpenCode Go anahtarına ihtiyaç duyar.",
+    "Antigravity has neither a permission mode nor a sandbox to hand it, so "
+    "what it may do without asking is whatever its own allow-rules say. The "
+    "Permissions and Sandbox boxes above belong to the other two; the working "
+    "directory still applies.":
+        "Antigravity'ye verilebilecek bir izin kipi ya da sandbox yok; sormadan "
+        "ne yapabileceğini kendi allow-rule'ları belirler. Yukarıdaki İzinler ve "
+        "Sandbox kutuları diğer ikisine ait; çalışma dizini burada da geçerli.",
     "{binary} is not on your PATH, so this cannot run yet. Install it, or pick "
     "another one above.":
         "{binary} PATH'te değil, dolayısıyla bu henüz çalışamaz. Kur ya da "
@@ -598,7 +615,7 @@ TR = {
     "configuration already says.":
         "Her komutla birlikte ajana söylenir, kendi yapılandırmanın zaten "
         "söylediklerinin üstüne eklenir.",
-    "  ·  asked Claude: {question}": "  ·  Claude'a soruldu: {question}",
+    "  ·  asked {who}: {question}": "  ·  {who} soruldu: {question}",
 
     # --- meetings: tray and pipeline ---------------------------------------
     "Record a meeting": "Toplantı kaydet",
@@ -668,12 +685,6 @@ TR = {
 
     # --- settings: meeting --------------------------------------------------
     "Minutes": "Tutanaklar",
-    "A meeting is recorded from two devices at once: your microphone and "
-    "whatever comes out of your speakers. Nothing has to guess who was "
-    "speaking, because the two never share a channel.":
-        "Toplantı iki aygıttan aynı anda kaydedilir: mikrofonun ve hoparlöründen "
-        "çıkan ses. Kimin konuştuğunun tahmin edilmesi gerekmez, çünkü ikisi hiç "
-        "aynı kanala girmez.",
     "Sound": "Ses",
     "Same as dictation": "Diktedekiyle aynı",
     "Current output": "Geçerli çıkış",
@@ -776,6 +787,12 @@ TR = {
     "Ready: {name}.": "Hazır: {name}.",
     "Nothing downloaded yet.": "Henüz bir şey indirilmedi.",
     "{name} has not been downloaded yet.": "{name} henüz indirilmedi.",
+    "{name} is here, but the program above is not. Download it first.":
+        "{name} burada, ama yukarıdaki program değil. Önce onu indirin.",
+    "{name} is not on this machine and this publisher does not offer it. "
+    "Choose another model, or another publisher.":
+        "{name} bu makinede yok ve bu yayıncı da sunmuyor. Başka bir model, "
+        "ya da başka bir yayıncı seçin.",
     "downloaded": "indirildi",
     "not downloaded": "indirilmedi",
     "Delete model": "Modeli sil",
@@ -801,16 +818,6 @@ TR = {
         "Düşünmeye eğitilmiş bir model, aksi söylenmedikçe düşünür; bir virgül "
         "için 300 token akıl yürütmek 300 token'lık bekleyiştir. Temizleme için "
         "doğrusu Kapalı.",
-    "OpenRouter is the quickest and the only one that needs nothing "
-    "installed. llama.cpp runs here, on a model downloaded below. Claude Code "
-    "and Codex clean up on the subscription you already have, without a "
-    "second key, and take a few seconds longer because each one opens a "
-    "session to do it.":
-        "OpenRouter en hızlısıdır ve kurulum istemeyen tek seçenektir. "
-        "llama.cpp burada, aşağıda indirilen bir modelle çalışır. Claude Code "
-        "ve Codex, ikinci bir anahtar olmadan zaten sahip olduğun abonelikle "
-        "temizler; her biri bunun için bir oturum açtığından birkaç saniye "
-        "daha sürer.",
     "whisper.cpp reaches the card through CUDA, ROCm or Vulkan when the build "
     "it is running was made with one. A build without any of them runs on the "
     "processor whatever this says.":

dikte/overlay.py

@@ -42,9 +42,11 @@ class Overlay(QWidget):
     of covering it, which is what lets a dictation and a command to the agent be
     under way at the same time and still both be visible."""
 
-    def __init__(self, corner="bottom-left", below=None, dismissable=False):
+    def __init__(self, corner="bottom-left", below=None, dismissable=False,
+                 screen_name=""):
         super().__init__(None)
         self.corner = corner
+        self.screen_name = screen_name
         self.below = below
         # A job that can run for ten minutes should not have to be watched for
         # ten minutes. Clicking such an indicator puts the progress away; the
@@ -251,8 +253,15 @@ class Overlay(QWidget):
         self.resize(width, HEIGHT)
 
     def _reposition(self):
-        # On a multi-monitor setup, show up where the user actually is.
+        # The screen the settings name, or, when none is named or it is not
-        screen = QApplication.screenAt(QCursor.pos()) or QApplication.primaryScreen()
+        # plugged in right now, where the user actually is. Names are connector
+        # names on X11 and model names on macOS, where two identical monitors
+        # can share one; the first then wins.
+        screen = next(
+            (item for item in QApplication.screens() if item.name() == self.screen_name),
+            None,
+        )
+        screen = screen or QApplication.screenAt(QCursor.pos()) or QApplication.primaryScreen()
         area = screen.availableGeometry()
         left = "left" in self.corner
         top = "top" in self.corner

dikte/settings_ui.py

@@ -6,7 +6,7 @@ import shutil
 import sys
 import threading
 
-from PyQt6.QtCore import QEvent, QObject, QRect, Qt, QUrl, pyqtSignal
+from PyQt6.QtCore import QEvent, QObject, QRect, Qt, QTimer, QUrl, pyqtSignal
 from PyQt6.QtGui import QDesktopServices, QGuiApplication, QKeySequence, QShortcut
 from PyQt6.QtWidgets import (
     QAbstractItemView, QAbstractSpinBox, QCheckBox, QComboBox, QDialog,
@@ -60,12 +60,24 @@ CLEANUP_MODELS = [
     "google/gemini-2.5-flash-lite", "anthropic/claude-haiku-4.5",
     "openai/gpt-5-mini", "meta-llama/llama-3.3-70b-instruct",
 ]
-# In the order they answer in. A request to OpenRouter is over in a second, a
+GEMINI_MODELS = [
-# model here takes a little longer and costs nothing, and the two CLIs the agent
+    "gemini-3.5-flash-lite", "gemini-3.1-flash-lite",
-# can run on open a whole session to do the smaller job.
+    "gemini-2.5-flash-lite", "gemini-3.5-flash", "gemini-2.5-flash",
+]
+# agy's model ids carry the reasoning effort in their suffix, which is why one
+# model appears here at more than one level. The same list seeds two boxes:
+# cleanup, which wants the bottom rung, and the agent, which sometimes does not.
+AGY_MODELS = [
+    "gemini-3.7-flash-low", "gemini-3.7-flash-medium", "gemini-3.7-flash-high",
+    "gemini-3.5-flash-low", "gemini-3.1-pro-low",
+]
+# In the order they answer in. The hosted requests are over in a second, a
+# model here takes a little longer and costs nothing, and the three CLIs the
+# agent can run on open a whole session to do the smaller job.
 CLEANUP_PROVIDERS = [
-    ("OpenRouter", "openrouter"), ("This machine (llama.cpp)", "local"),
+    ("OpenRouter", "openrouter"), ("Google AI Studio", "gemini"),
-    ("Claude Code", "claude"), ("Codex", "codex"),
+    ("OpenCode Go", "opencode"), ("This machine (llama.cpp)", "local"),
+    ("Claude Code", "claude"), ("Codex", "codex"), ("Antigravity", "agy"),
 ]
 # Cleaning up a sentence is the lightest thing either of them will ever be
 # asked, so the small model comes first.
@@ -77,7 +89,8 @@ MEETING_MODELS = [
     "anthropic/claude-sonnet-5", "openai/gpt-5.4", "x-ai/grok-4.5",
 ]
 ASSISTANT_PROVIDERS = [
-    ("Claude Code", "claude"), ("Codex", "codex"), ("OpenRouter", "openrouter"),
+    ("Claude Code", "claude"), ("Codex", "codex"), ("Antigravity", "agy"),
+    ("OpenRouter", "openrouter"), ("OpenCode Go", "opencode"),
 ]
 # Aliases resolve to the newest model of that name, so they age better than an
 # id does; a full id can be typed in when a particular one is wanted.
@@ -92,6 +105,13 @@ ASSISTANT_OR_MODELS = [
     "google/gemini-3.5-flash", "anthropic/claude-sonnet-5", "openai/gpt-5.4",
     "x-ai/grok-4.5", "google/gemini-3.1-pro-preview",
 ]
+# A starting set of the models OpenCode Go serves over /chat/completions; the
+# Fetch button asks the endpoint itself for the full catalog of the day.
+OPENCODE_MODELS = [
+    "deepseek-v4-flash", "deepseek-v4-pro", "glm-5.3", "glm-5.2", "glm-5.1",
+    "kimi-k3", "kimi-k2.7-code", "kimi-k2.6", "longcat-2.0",
+    "mimo-v2.5", "mimo-v2.5-pro", "hy3",
+]
 # What Claude Code may do without being able to ask. It cannot ask: there is no
 # window to answer in, so a mode that would have prompted denies instead.
 PERMISSION_MODES = [
@@ -227,6 +247,13 @@ class LocalModelBox(QGroupBox):
         self._pending = False
         self._stop = False
         self._wanted = ""              # the model to select once a list arrives
+        self._chosen_in = ""           # the publisher the selected model is from
+        # Typing or arrowing through the publisher box changes its text a
+        # character at a time, and each of those would otherwise be a request.
+        self._later = QTimer(self)
+        self._later.setSingleShot(True)
+        self._later.setInterval(400)
+        self._later.timeout.connect(self._later_fetch)
 
         form = QFormLayout(self)
 
@@ -308,6 +335,8 @@ class LocalModelBox(QGroupBox):
         self._wanted = model
         self._pending = True
         self._show_program()
+        self._chosen_in = repo or (ggml.SUGGESTED_LLM[0] if self._repos is not None
+                                   else "")
         if self._repos is not None:
             self.repo.blockSignals(True)
             self.repo.clear()
@@ -347,11 +376,18 @@ class LocalModelBox(QGroupBox):
 
     def _fill_repos(self, current):
         def work():
-            self._listed.emit([("repos", ggml.llm_repos())], "")
+            self._listed.emit([("repos", ggml.llm_repos(), "")], "")
 
         threading.Thread(target=work, daemon=True).start()
 
     def _repo_changed(self):
+        if not self._downloading:
+            self._later.start()
+
+    def _later_fetch(self):
+        # A download that started inside the wait was not there to be seen when
+        # the timer went off, and rebuilding the rows underneath one is exactly
+        # what the guard above is for.
         if not self._downloading:
             self._fetch_models(self.repository())
 
@@ -361,18 +397,29 @@ class LocalModelBox(QGroupBox):
         def work():
             try:
                 found = self._models(repo) if self._repos is not None else self._models()
-                self._listed.emit([("models", found)], "")
+                self._listed.emit([("models", found, repo)], "")
             except ggml.LocalError as exc:
-                self._listed.emit([], str(exc))
+                self._listed.emit([("models", [], repo)], str(exc))
 
         threading.Thread(target=work, daemon=True).start()
 
     def _on_listed(self, payload, error):
+        kind, found, repo = payload[0] if payload else ("repos", [], "")
+        # A publisher changed while its predecessor's list was still on the way
+        # would otherwise be answered with the wrong models, whichever request
+        # happened to come back last.
+        if kind == "models" and repo != self.repository():
+            return
         if error:
+            # The list is the publisher's, so a failed one leaves the box no
+            # longer showing this publisher's models: emptying it is what keeps
+            # the two boxes saying the same thing. The message goes on after,
+            # because filling the box writes a status of its own.
+            if kind == "models":
+                self._fill_models([])
+            self._refresh_buttons()
             self.status.setText(error)
-            self._refresh_buttons()
             return
-        kind, found = payload[0]
         if kind == "repos":
             current = self.repo.currentText()
             self.repo.blockSignals(True)
@@ -386,7 +433,12 @@ class LocalModelBox(QGroupBox):
 
     def _fill_models(self, items):
         """One row per model, saying what it weighs and whether it is here."""
-        wanted = self._wanted or self.selected()
+        # The selection is only worth carrying over within the publisher it was
+        # made in. Carried across one, a model this repository does not publish
+        # would be added back as "not downloaded" and selected again, and
+        # changing the publisher would leave the model box looking untouched.
+        same = self._repos is None or self.repository() == self._chosen_in
+        wanted = self._wanted or (self.selected() if same else "")
         here = [name for name in (self._model_path(i.name).name for i in items)]
         self.model.blockSignals(True)
         self.model.clear()
@@ -411,6 +463,7 @@ class LocalModelBox(QGroupBox):
         self.model.blockSignals(False)
         self._fit_popup(self.model)
         self._wanted = ""
+        self._chosen_in = self.repository()
         self._model_changed()
 
     def _on_disk(self):
@@ -439,6 +492,9 @@ class LocalModelBox(QGroupBox):
         self._show_program()
         if error:
             self.program_label.setText(error)
+        # The model line says whether the program is here, so installing one
+        # changes what it should read.
+        self._refresh_buttons()
         self.changed.emit()
 
     def _current_item(self):
@@ -525,15 +581,30 @@ class LocalModelBox(QGroupBox):
     def _refresh_buttons(self):
         name = self.selected()
         here = bool(name) and ggml.have_model(self._model_path(name))
+        # A row carries what it takes to fetch it. The ones that do not are the
+        # models found on this disk and the one the settings name but the list
+        # does not offer: there is nothing to press Download for on those, and
+        # a button that can only do nothing is worse than one that is out.
+        item = self._current_item()
         self.delete_button.setEnabled(here and not self._downloading)
         self.download_button.setText(t("Stop") if self._downloading else t("Download"))
-        self.download_button.setEnabled(self._downloading or (bool(name) and not here))
+        self.download_button.setEnabled(self._downloading or (item is not None
+                                                              and not here))
         if self._downloading:
             return
         if not name:
             self.status.setText(t("Nothing downloaded yet."))
+        elif here and not ggml.program_path(self.program):
+            # The model alone runs nothing, and "Ready" over a missing program
+            # reads as though it does.
+            self.status.setText(t("{name} is here, but the program above is "
+                                  "not. Download it first.", name=name))
         elif here:
             self.status.setText(t("Ready: {name}.", name=name))
+        elif item is None:
+            self.status.setText(t("{name} is not on this machine and this "
+                                  "publisher does not offer it. Choose another "
+                                  "model, or another publisher.", name=name))
         else:
             self.status.setText(t("{name} has not been downloaded yet.", name=name))
 
@@ -558,8 +629,13 @@ class SettingsWindow(QDialog):
         return self._sources
 
     _models_loaded = pyqtSignal(list, str)
+    _gemini_models_loaded = pyqtSignal(list, str)
+    _opencode_models_loaded = pyqtSignal(list, str)
     _transcribe_models_loaded = pyqtSignal(list, str)
     _codex_models_loaded = pyqtSignal(list)
+    _agy_models_loaded = pyqtSignal(list)
+    # Which hosted provider's list arrived on its own at open, and the list.
+    _hosted_models_loaded = pyqtSignal(str, list)
     # Which key was tested, whether it worked, and what to write under it.
     _test_done = pyqtSignal(str, bool, str)
     # The release that was found, or None, and what went wrong instead.
@@ -594,6 +670,7 @@ class SettingsWindow(QDialog):
 
         tabs = self.tabs = QTabWidget(self)
         tabs.addTab(self._scrolled(self._general_tab()), t("General"))
+        tabs.addTab(self._scrolled(self._display_tab()), t("Display"))
         self.api_tab_index = tabs.addTab(
             self._scrolled(self._api_tab()), t("API and models"))
         tabs.addTab(self._scrolled(self._prompt_tab()), t("Cleanup rules"))
@@ -617,8 +694,12 @@ class SettingsWindow(QDialog):
         self._size_to_screen(680, 640)
 
         self._models_loaded.connect(self._on_models_loaded)
+        self._gemini_models_loaded.connect(self._on_gemini_models_loaded)
         self._transcribe_models_loaded.connect(self._on_transcribe_models_loaded)
         self._codex_models_loaded.connect(self._on_codex_models_loaded)
+        self._opencode_models_loaded.connect(self._on_opencode_models_loaded)
+        self._agy_models_loaded.connect(self._on_agy_models_loaded)
+        self._hosted_models_loaded.connect(self._on_hosted_models_loaded)
         self._test_done.connect(self._on_test_done)
         self._update_checked.connect(self._on_update_checked)
         self.transcriber.progress.connect(self._on_file_progress)
@@ -630,6 +711,8 @@ class SettingsWindow(QDialog):
             self.meetings.failed.connect(self._on_minutes_failed)
         self._load()
         self._load_codex_models()
+        self._load_agy_models()
+        self._load_hosted_models()
         # Connected after the load, so that filling the boxes in is not taken
         # for the user ticking them.
         self.file_timestamps.toggled.connect(self._remember_file_choices)
@@ -714,11 +797,6 @@ class SettingsWindow(QDialog):
         self.restore_clipboard = QCheckBox(t("Restore the previous clipboard after pasting"))
         form.addRow("", self.restore_clipboard)
 
-        self.corner = QComboBox()
-        for value in CORNERS:
-            self.corner.addItem(t(value), value)
-        form.addRow(t("Indicator corner"), self.corner)
-
         self.max_seconds = QSpinBox()
         self.max_seconds.setRange(10, 3600)
         self.max_seconds.setSuffix(t(" s"))
@@ -769,6 +847,31 @@ class SettingsWindow(QDialog):
                                   self.update_now))
         return page
 
+    def _display_tab(self):
+        page = QWidget()
+        form = QFormLayout(page)
+
+        self.indicator_screen = QComboBox()
+        self.indicator_screen.addItem(t("Follow the mouse pointer"), "")
+        for screen in QGuiApplication.screens():
+            # The native resolution, so that a scaled 4K screen reads
+            # 3840 × 2160 and not the 1920 × 1080 Qt sees through the scale.
+            area = screen.geometry()
+            ratio = screen.devicePixelRatio()
+            self.indicator_screen.addItem(
+                t("{name} ({width} × {height})", name=screen.name(),
+                  width=round(area.width() * ratio),
+                  height=round(area.height() * ratio)),
+                screen.name(),
+            )
+        form.addRow(t("Indicator screen"), self.indicator_screen)
+
+        self.corner = QComboBox()
+        for value in CORNERS:
+            self.corner.addItem(t(value), value)
+        form.addRow(t("Indicator corner"), self.corner)
+        return page
+
     def _api_tab(self):
         page = QWidget()
         outer = QVBoxLayout(page)
@@ -786,6 +889,12 @@ class SettingsWindow(QDialog):
         self.openrouter_key = self._key_row(
             keys_form, "openrouter", t("sk-or-… (falls back to OPENROUTER_API_KEY)"),
             self._test_openrouter)
+        self.gemini_key = self._key_row(
+            keys_form, "gemini", t("(falls back to GEMINI_API_KEY)"),
+            self._test_gemini, service="Google AI Studio")
+        self.opencode_key = self._key_row(
+            keys_form, "opencode", t("(falls back to OPENCODE_API_KEY)"),
+            self._test_opencode, service="OpenCode Go")
         outer.addWidget(keys)
 
         stt = QGroupBox(t("Speech to text"))
@@ -857,13 +966,6 @@ class SettingsWindow(QDialog):
         self.cleanup_provider = QComboBox()
         for label, value in CLEANUP_PROVIDERS:
             self.cleanup_provider.addItem(t(label), value)
-        self.cleanup_provider.setToolTip(t(
-            "OpenRouter is the quickest and the only one that needs nothing "
-            "installed. llama.cpp runs here, on a model downloaded below. Claude "
-            "Code and Codex clean up on the subscription you already have, "
-            "without a second key, and take a few seconds longer because each "
-            "one opens a session to do it."
-        ))
         self.cleanup_provider.currentIndexChanged.connect(self._cleanup_provider_changed)
         orr_form.addRow(t("Runs on"), self.cleanup_provider)
 
@@ -875,6 +977,15 @@ class SettingsWindow(QDialog):
         self.cleanup_model_row = self._row(self.cleanup_model, self.refresh_models)
         orr_form.addRow(t("Model"), self.cleanup_model_row)
 
+        self.cleanup_gemini_model = QComboBox()
+        self.cleanup_gemini_model.setEditable(True)
+        self.cleanup_gemini_model.addItems(GEMINI_MODELS)
+        self.refresh_gemini_models = QPushButton(t("Fetch model list"))
+        self.refresh_gemini_models.clicked.connect(self._load_gemini_models)
+        self.cleanup_gemini_model_row = self._row(
+            self.cleanup_gemini_model, self.refresh_gemini_models)
+        orr_form.addRow(t("Model"), self.cleanup_gemini_model_row)
+
         # One row per provider rather than one box that means a different thing
         # in each: an OpenRouter id and a Claude alias do not belong in the same
         # field, and only the row of whoever is chosen is on screen.
@@ -890,6 +1001,21 @@ class SettingsWindow(QDialog):
         self.cleanup_codex_model.setToolTip(_typed_model_note("Codex"))
         orr_form.addRow(t("Model"), self.cleanup_codex_model)
 
+        self.cleanup_opencode_model = QComboBox()
+        self.cleanup_opencode_model.setEditable(True)
+        self.cleanup_opencode_model.addItems(OPENCODE_MODELS)
+        self.cleanup_opencode_model.setToolTip(_typed_model_note("OpenCode Go"))
+        self.refresh_opencode_models = QPushButton(t("Fetch model list"))
+        self.refresh_opencode_models.clicked.connect(self._load_opencode_models)
+        self.cleanup_opencode_model_row = self._row(self.cleanup_opencode_model,
+                                                    self.refresh_opencode_models)
+        orr_form.addRow(t("Model"), self.cleanup_opencode_model_row)
+
+        self.cleanup_agy_model = QComboBox()
+        self.cleanup_agy_model.setEditable(True)
+        self.cleanup_agy_model.addItems([t("Antigravity's own default")] + AGY_MODELS)
+        orr_form.addRow(t("Model"), self.cleanup_agy_model)
+
         self.cleanup_reasoning = QComboBox()
         for label, value in REASONING_LEVELS:
             self.cleanup_reasoning.addItem(t(label), value)
@@ -969,17 +1095,6 @@ class SettingsWindow(QDialog):
     def _assistant_tab(self):
         page = QWidget()
         layout = QVBoxLayout(page)
-        intro = QLabel(t(
-            "This shortcut records the same way dictation does, but the "
-            "transcript is not what gets pasted. It goes to an agent as a "
-            "command, and what comes back is pasted instead: the answer to a "
-            "question, or a sentence saying what was done. Claude Code and "
-            "Codex run as the session you would have opened yourself, with your "
-            "skills, your connected services and your account."
-        ))
-        intro.setWordWrap(True)
-        layout.addWidget(intro)
-
         self.assistant_found = QLabel("")
         self.assistant_found.setWordWrap(True)
         layout.addWidget(self.assistant_found)
@@ -1013,7 +1128,7 @@ class SettingsWindow(QDialog):
         dir_note.setWordWrap(True)
         how_form.addRow(dir_note)
 
-        # One scale for all three: how hard to think is one thing to want, and
+        # One scale for all four: how hard to think is one thing to want, and
         # each provider is handed the nearest rung it actually has.
         self.assistant_reasoning = QComboBox()
         for label, value in REASONING_LEVELS:
@@ -1036,7 +1151,7 @@ class SettingsWindow(QDialog):
         layout.addWidget(how)
 
         # One box per provider, only the chosen one on screen: they have nothing
-        # in common past the model, and three sets of half-relevant fields would
+        # in common past the model, and four sets of half-relevant fields would
         # be worse than none.
         self.claude_box = QGroupBox(t("Claude Code"))
         claude_form = QFormLayout(self.claude_box)
@@ -1087,6 +1202,41 @@ class SettingsWindow(QDialog):
         or_form.addRow(or_note)
         layout.addWidget(self.openrouter_box)
 
+        self.agy_box = QGroupBox(t("Antigravity"))
+        agy_form = QFormLayout(self.agy_box)
+        self.assistant_agy_model = QComboBox()
+        self.assistant_agy_model.setEditable(True)
+        self.assistant_agy_model.addItem(t("Antigravity's own default"), "")
+        for name in AGY_MODELS:
+            self.assistant_agy_model.addItem(name, name)
+        agy_form.addRow(t("Model"), self.assistant_agy_model)
+        agy_note = QLabel(t(
+            "Antigravity has neither a permission mode nor a sandbox to hand it, "
+            "so what it may do without asking is whatever its own allow-rules "
+            "say. The Permissions and Sandbox boxes above belong to the other "
+            "two; the working directory still applies."
+        ))
+        agy_note.setWordWrap(True)
+        agy_form.addRow(agy_note)
+        layout.addWidget(self.agy_box)
+
+        self.opencode_box = QGroupBox("OpenCode Go")
+        og_form = QFormLayout(self.opencode_box)
+        self.assistant_opencode_model = QComboBox()
+        self.assistant_opencode_model.setEditable(True)
+        self.assistant_opencode_model.addItems(OPENCODE_MODELS)
+        og_form.addRow(t("Model"), self.assistant_opencode_model)
+        og_note = QLabel(t(
+            "A plain question and a plain answer, over the OpenCode Go key you "
+            "already have. It runs no commands, opens no files and reaches none "
+            "of your services, so it can tell you what the capital of Peru is "
+            "but not what is in your calendar. Working directory and permissions "
+            "above mean nothing here."
+        ))
+        og_note.setWordWrap(True)
+        og_form.addRow(og_note)
+        layout.addWidget(self.opencode_box)
+
         thread = QGroupBox(t("The conversation"))
         thread_form = QFormLayout(thread)
         self.assistant_session_minutes = QSpinBox()
@@ -1142,14 +1292,6 @@ class SettingsWindow(QDialog):
     def _meeting_tab(self):
         page = QWidget()
         layout = QVBoxLayout(page)
-        intro = QLabel(t(
-            "A meeting is recorded from two devices at once: your microphone and "
-            "whatever comes out of your speakers. Nothing has to guess who was "
-            "speaking, because the two never share a channel."
-        ))
-        intro.setWordWrap(True)
-        layout.addWidget(intro)
-
         sources = QGroupBox(t("Sound"))
         sources_form = QFormLayout(sources)
         self.meeting_mic = QComboBox()
@@ -1558,7 +1700,7 @@ class SettingsWindow(QDialog):
             box.lineEdit().setPlaceholderText(placeholder)
         return box
 
-    def _key_row(self, form, provider, placeholder, tester):
+    def _key_row(self, form, provider, placeholder, tester, service=""):
         """A key field, its Test button and the line the answer lands on.
 
         The field and the pair the answer needs are filed under the provider's
@@ -1572,7 +1714,8 @@ class SettingsWindow(QDialog):
         button.clicked.connect(tester)
         answer = QLabel("")
         answer.setWordWrap(True)
-        form.addRow(cfg.TRANSCRIBERS[provider].service, self._row(field, button))
+        label = service or cfg.TRANSCRIBERS[provider].service
+        form.addRow(label, self._row(field, button))
         form.addRow("", answer)
         self._key_fields[provider] = field
         self._testers[provider] = (button, answer)
@@ -1634,6 +1777,10 @@ class SettingsWindow(QDialog):
         self.auto_paste.setChecked(conf["auto_paste"])
         self.paste_shortcut.setCurrentText(conf["paste_shortcut"])
         self.restore_clipboard.setChecked(conf["restore_clipboard"])
+        screen_name = conf["overlay_screen"]
+        if screen_name and self.indicator_screen.findData(screen_name) < 0:
+            self.indicator_screen.addItem(t("{name} (not connected)", name=screen_name), screen_name)
+        self._select_data(self.indicator_screen, screen_name)
         self._select_data(self.corner, conf["overlay_corner"])
         self.max_seconds.setValue(conf["max_seconds"])
         self.skip_silent.setChecked(conf["skip_silent"])
@@ -1646,6 +1793,8 @@ class SettingsWindow(QDialog):
         for name, who in cfg.TRANSCRIBERS.items():
             self._key_fields[name].setText(conf[who.key])
             self._models[name] = conf[who.model]
+        self.gemini_key.setText(conf["gemini_api_key"])
+        self.opencode_key.setText(conf["opencode_api_key"])
         self._shown_provider = ""
         self._select_data(self.transcribe_provider, conf["transcribe_provider"])
         self._provider_changed()  # selecting index 0 fires no signal
@@ -1656,10 +1805,17 @@ class SettingsWindow(QDialog):
 
         self.cleanup_enabled.setChecked(conf["cleanup_enabled"])
         self.cleanup_model.setCurrentText(conf["cleanup_model"])
+        self.cleanup_gemini_model.setCurrentText(
+            conf["cleanup_gemini_model"] or cfg.DEFAULTS["cleanup_gemini_model"]
+        )
         self.cleanup_claude_model.setCurrentText(conf["cleanup_claude_model"])
         self.cleanup_codex_model.setCurrentText(
             conf["cleanup_codex_model"] or t("Codex's own default")
         )
+        self.cleanup_agy_model.setCurrentText(
+            conf["cleanup_agy_model"] or t("Antigravity's own default")
+        )
+        self.cleanup_opencode_model.setCurrentText(conf["cleanup_opencode_model"])
         self._select_data(self.cleanup_provider, conf["cleanup_provider"])
         self._cleanup_provider_changed()  # selecting index 0 fires no signal
         self._select_data(self.cleanup_reasoning, conf["cleanup_reasoning"])
@@ -1690,6 +1846,8 @@ class SettingsWindow(QDialog):
         self.assistant_codex_model.setCurrentText(conf["assistant_codex_model"])
         self._select_data(self.assistant_codex_sandbox, conf["assistant_codex_sandbox"])
         self.assistant_openrouter_model.setCurrentText(conf["assistant_openrouter_model"])
+        self.assistant_agy_model.setCurrentText(conf["assistant_agy_model"])
+        self.assistant_opencode_model.setCurrentText(conf["assistant_opencode_model"])
         self._assistant_provider_changed()  # selecting index 0 fires no signal
         self._select_data(self.assistant_reasoning, conf["assistant_reasoning"])
         self.assistant_dir.setText(conf["assistant_dir"])
@@ -1741,6 +1899,7 @@ class SettingsWindow(QDialog):
         conf["auto_paste"] = self.auto_paste.isChecked()
         conf["paste_shortcut"] = self.paste_shortcut.currentText().strip()
         conf["restore_clipboard"] = self.restore_clipboard.isChecked()
+        conf["overlay_screen"] = self.indicator_screen.currentData() or ""
         conf["overlay_corner"] = self.corner.currentData() or "bottom-left"
         conf["max_seconds"] = self.max_seconds.value()
         conf["skip_silent"] = self.skip_silent.isChecked()
@@ -1756,6 +1915,8 @@ class SettingsWindow(QDialog):
         for name, who in cfg.TRANSCRIBERS.items():
             conf[who.key] = self._key_fields[name].text().strip()
             conf[who.model] = self._models[name].strip() or cfg.DEFAULTS[who.model]
+        conf["gemini_api_key"] = self.gemini_key.text().strip()
+        conf["opencode_api_key"] = self.opencode_key.text().strip()
         conf["local_model"] = self.local_whisper.selected()
         conf["local_gpu"] = self.local_gpu.isChecked()
         conf["local_preload"] = self.local_preload.isChecked()
@@ -1764,12 +1925,25 @@ class SettingsWindow(QDialog):
         conf["cleanup_enabled"] = self.cleanup_enabled.isChecked()
         conf["cleanup_provider"] = self.cleanup_provider.currentData() or "openrouter"
         conf["cleanup_model"] = self.cleanup_model.currentText().strip()
+        conf["cleanup_gemini_model"] = (
+            self.cleanup_gemini_model.currentText().strip()
+            or cfg.DEFAULTS["cleanup_gemini_model"]
+        )
         conf["cleanup_claude_model"] = (self.cleanup_claude_model.currentText().strip()
                                         or cfg.DEFAULTS["cleanup_claude_model"])
         codex_cleanup_model = self.cleanup_codex_model.currentText().strip()
         conf["cleanup_codex_model"] = (
             "" if codex_cleanup_model == t("Codex's own default") else codex_cleanup_model
         )
+        agy_cleanup_model = self.cleanup_agy_model.currentText().strip()
+        conf["cleanup_agy_model"] = (
+            "" if agy_cleanup_model == t("Antigravity's own default")
+            else agy_cleanup_model
+        )
+        conf["cleanup_opencode_model"] = (
+            self.cleanup_opencode_model.currentText().strip()
+            or cfg.DEFAULTS["cleanup_opencode_model"]
+        )
         conf["cleanup_reasoning"] = self.cleanup_reasoning.currentData() or ""
         conf["local_llm_model"] = self.local_llm.selected()
         conf["local_llm_repo"] = self.local_llm.repository()
@@ -1808,6 +1982,14 @@ class SettingsWindow(QDialog):
             self.assistant_openrouter_model.currentText().strip()
             or cfg.DEFAULTS["assistant_openrouter_model"]
         )
+        agy_model = self.assistant_agy_model.currentText().strip()
+        conf["assistant_agy_model"] = (
+            "" if agy_model == t("Antigravity's own default") else agy_model
+        )
+        conf["assistant_opencode_model"] = (
+            self.assistant_opencode_model.currentText().strip()
+            or cfg.DEFAULTS["assistant_opencode_model"]
+        )
         conf["assistant_reasoning"] = self.assistant_reasoning.currentData() or ""
         conf["assistant_dir"] = self.assistant_dir.text().strip()
         conf["assistant_timeout"] = self.assistant_timeout.value()
@@ -1971,6 +2153,30 @@ class SettingsWindow(QDialog):
             combo.setCurrentText(current)
         self.models_label.setText(t("{count} models loaded.", count=len(models)))
 
+    def _load_gemini_models(self):
+        self.refresh_gemini_models.setEnabled(False)
+        self.models_label.setText(t("Fetching model list…"))
+        key, base = self._typed_key("gemini")
+
+        def work():
+            try:
+                self._gemini_models_loaded.emit(api.gemini_models(key, base), "")
+            except api.ApiError as exc:
+                self._gemini_models_loaded.emit([], str(exc))
+
+        threading.Thread(target=work, daemon=True).start()
+
+    def _on_gemini_models_loaded(self, models, error):
+        self.refresh_gemini_models.setEnabled(True)
+        if error:
+            self.models_label.setText(t("Could not fetch the list: {error}", error=error))
+            return
+        current = self.cleanup_gemini_model.currentText()
+        self.cleanup_gemini_model.clear()
+        self.cleanup_gemini_model.addItems(models)
+        self.cleanup_gemini_model.setCurrentText(current)
+        self.models_label.setText(t("{count} models loaded.", count=len(models)))
+
     def _load_codex_models(self):
         """Ask Codex which models it offers, off the interface thread.
 
@@ -1998,6 +2204,112 @@ class SettingsWindow(QDialog):
                 combo.addItem(name, name)
             combo.setCurrentText(current)
 
+    def _load_opencode_models(self):
+        self.refresh_opencode_models.setEnabled(False)
+        self.models_label.setText(t("Fetching model list…"))
+        key, base = self._typed_key("opencode")
+
+        def work():
+            try:
+                self._opencode_models_loaded.emit(
+                    api.openai_models(key, base, "OpenCode Go"), "")
+            except api.ApiError as exc:
+                self._opencode_models_loaded.emit([], str(exc))
+
+        threading.Thread(target=work, daemon=True).start()
+
+    def _on_opencode_models_loaded(self, models, error):
+        self.refresh_opencode_models.setEnabled(True)
+        if error:
+            self.models_label.setText(t("Could not fetch the list: {error}", error=error))
+            return
+        self._fill_opencode_boxes(models)
+        self.models_label.setText(t("{count} models loaded.", count=len(models)))
+
+    def _fill_opencode_boxes(self, models):
+        # The agent runs on the same key and catalog, so its box is refilled
+        # from the same list.
+        for combo in (self.cleanup_opencode_model, self.assistant_opencode_model):
+            current = combo.currentText()
+            combo.clear()
+            combo.addItems(models)
+            combo.setCurrentText(current)
+
+    def _load_agy_models(self):
+        """Ask Antigravity which models it offers, off the interface thread.
+
+        The same arrangement as Codex, except agy answers over the network
+        rather than from a cache, so the couple of seconds it takes are spent
+        where nobody is waiting. Skipped when agy is not installed, which is
+        also when the built-in list stays on screen and nobody is running
+        Antigravity anyway.
+        """
+        if not shutil.which("agy"):
+            return
+
+        def work():
+            found = assistant.agy_models()
+            if found:
+                self._agy_models_loaded.emit(found)
+
+        threading.Thread(target=work, daemon=True).start()
+
+    def _on_agy_models_loaded(self, models):
+        for combo in (self.cleanup_agy_model, self.assistant_agy_model):
+            current = combo.currentText()
+            combo.clear()
+            combo.addItem(t("Antigravity's own default"), "")
+            for name in models:
+                combo.addItem(name, name)
+            combo.setCurrentText(current)
+
+    def _load_hosted_models(self):
+        """Fetch the hosted model lists at open, without being asked.
+
+        The Fetch buttons stay: they are the retry, and the place a failure is
+        worth explaining. Here nobody asked, so an error changes nothing on
+        screen and the built-in lists remain, and a provider whose key has not
+        been given yet is not called at all.
+        """
+        jobs = []
+        openrouter_key = self.conf.openrouter_key()
+        if openrouter_key:
+            jobs.append(("openrouter",
+                         lambda: api.openrouter_models(openrouter_key)))
+        gemini_key = self.conf.gemini_key()
+        gemini_base = self.conf["gemini_base_url"]
+        if gemini_key:
+            jobs.append(("gemini",
+                         lambda: api.gemini_models(gemini_key, gemini_base)))
+        opencode_key = self.conf.opencode_key()
+        opencode_base = self.conf["opencode_base_url"]
+        if opencode_key:
+            jobs.append(("opencode",
+                         lambda: api.openai_models(opencode_key, opencode_base,
+                                                   "OpenCode Go")))
+        for provider, fetch in jobs:
+            def work(provider=provider, fetch=fetch):
+                try:
+                    found = fetch()
+                except api.ApiError:
+                    return
+                if found:
+                    self._hosted_models_loaded.emit(provider, found)
+
+            threading.Thread(target=work, daemon=True).start()
+
+    def _on_hosted_models_loaded(self, provider, models):
+        if provider == "opencode":
+            self._fill_opencode_boxes(models)
+            return
+        combos = ((self.cleanup_model, self.meeting_model)
+                  if provider == "openrouter" else (self.cleanup_gemini_model,))
+        for combo in combos:
+            current = combo.currentText()
+            combo.clear()
+            combo.addItems(models)
+            combo.setCurrentText(current)
+
     def _test_openai(self):
         key, base = self._typed_key("openai")
         self._test_key("openai", lambda: t(
@@ -2016,11 +2328,30 @@ class SettingsWindow(QDialog):
         key, _ = self._typed_key("openrouter")
         self._test_key("openrouter", lambda: api.openrouter_key_status(key))
 
+    def _test_gemini(self):
+        key, base = self._typed_key("gemini")
+        self._test_key("gemini", lambda: t(
+            "Connection works. {count} models visible.",
+            count=len(api.gemini_models(key, base)),
+        ))
+
+    def _test_opencode(self):
+        key, base = self._typed_key("opencode")
+        self._test_key("opencode", lambda: t(
+            "Connection works. {count} models visible.",
+            count=len(api.openai_models(key, base, "OpenCode Go")),
+        ))
+
     def _typed_key(self, provider):
         """(key, base URL) for a provider, preferring what is in the field now."""
-        who = cfg.TRANSCRIBERS[provider]
+        if provider in cfg.TRANSCRIBERS:
+            who = cfg.TRANSCRIBERS[provider]
+            key_setting, url_setting = who.key, who.url
+        else:
+            key_setting = f"{provider}_api_key"
+            url_setting = f"{provider}_base_url"
         typed = self._key_fields[provider].text().strip()
-        return typed or self.conf.api_key(who.key), self.conf[who.url]
+        return typed or self.conf.api_key(key_setting), self.conf[url_setting]
 
     def _test_key(self, provider, ask):
         """Run `ask` off the interface thread and write its answer under the key.
@@ -2244,10 +2575,16 @@ class SettingsWindow(QDialog):
         provider = self.cleanup_provider.currentData() or "openrouter"
         self.cleanup_form.setRowVisible(self.cleanup_model_row,
                                         provider == "openrouter")
+        self.cleanup_form.setRowVisible(self.cleanup_gemini_model_row,
+                                        provider == "gemini")
         self.cleanup_form.setRowVisible(self.cleanup_claude_model,
                                         provider == "claude")
         self.cleanup_form.setRowVisible(self.cleanup_codex_model,
                                         provider == "codex")
+        self.cleanup_form.setRowVisible(self.cleanup_opencode_model_row,
+                                        provider == "opencode")
+        self.cleanup_form.setRowVisible(self.cleanup_agy_model,
+                                        provider == "agy")
         self.cleanup_form.setRowVisible(self.cleanup_reasoning,
                                         provider != "local")
         self.cleanup_form.setRowVisible(self.local_llm, provider == "local")
@@ -2256,6 +2593,10 @@ class SettingsWindow(QDialog):
         found = shutil.which(binary) if binary else ""
         if provider == "local":
             self.models_label.setText(t("Runs on this machine, on llama.cpp."))
+        elif provider == "gemini":
+            self.models_label.setText(t("Runs on Google AI Studio."))
+        elif provider == "opencode":
+            self.models_label.setText(t("Runs on OpenCode Go."))
         elif not binary:
             self.models_label.setText(t("Runs on OpenRouter."))
         elif found:
@@ -2272,6 +2613,8 @@ class SettingsWindow(QDialog):
         self.claude_box.setVisible(provider == "claude")
         self.codex_box.setVisible(provider == "codex")
         self.openrouter_box.setVisible(provider == "openrouter")
+        self.agy_box.setVisible(provider == "agy")
+        self.opencode_box.setVisible(provider == "opencode")
         self._refresh_assistant_status()
 
     def _refresh_assistant_status(self):
@@ -2279,9 +2622,14 @@ class SettingsWindow(QDialog):
         binary = assistant.executable(provider)
         found = shutil.which(binary) if binary else ""
         if not binary:
-            self.assistant_found.setText(
+            if provider == "opencode":
-                t("Needs no program installed, only the OpenRouter key.")
+                self.assistant_found.setText(
-            )
+                    t("Needs no program installed, only an OpenCode Go key.")
+                )
+            else:
+                self.assistant_found.setText(
+                    t("Needs no program installed, only the OpenRouter key.")
+                )
         elif found:
             self.assistant_found.setText(t("Found: {path}", path=found))
         else:
@@ -2400,7 +2748,11 @@ class SettingsWindow(QDialog):
                 # The text of an answer says nothing about what was asked, and
                 # out of that context half of them read like non sequiturs.
                 asked = (row.get("question") or row.get("raw") or "").replace("\n", " ")
-                header += t("  ·  asked Claude: {question}",
+                # Rows written before the provider was recorded are all Claude's,
+                # because it was the only one the history could name.
+                who = assistant.SERVICES.get(row.get("assistant"), "Claude")
+                header += t("  ·  asked {who}: {question}",
+                            who=i18n.name(who, "dative"),
                             question=asked[:60] + ("…" if len(asked) > 60 else ""))
             item = QListWidgetItem(f"{header}\n{preview}")
             item.setData(Qt.ItemDataRole.UserRole, row)

dikte/worker.py

@@ -181,7 +181,8 @@ class Pipeline(QObject):
                 "cleanup_error": warning,
                 "mode": "ask" if ask else "",
                 "question": question,
-                "assistant_model": conf["assistant_model"] if ask else "",
+                "assistant": assistant.provider(conf) if ask else "",
+                "assistant_model": assistant.model(conf) if ask else "",
                 "raw": raw,
                 "text": text,
             }

packaging/dikte.spec

@@ -56,6 +56,18 @@ analysis = Analysis(                            # noqa: F821
     noarchive=False,
 )
 
+# Qt's xcb platform plugin uses libxkbcommon in two halves: the core library
+# and libxkbcommon-x11, which allocates keymap objects and hands them to the
+# core half to use and free, so the two must come from the same build. The
+# build machine has only the core half installed, which had PyInstaller
+# bundling that one while the other kept coming from the user's system, and a
+# 22.04-era core freeing what a current x11 half allocated is the startup
+# crash of issue #57. Ship neither: any desktop that can show a window
+# carries both, from one build.
+if not (MACOS or WINDOWS):
+    analysis.binaries = [entry for entry in analysis.binaries
+                         if "libxkbcommon" not in entry[0]]
+
 archive = PYZ(analysis.pure)                    # noqa: F821
 
 executable = EXE(                               # noqa: F821

tests/test_api.py

@@ -119,9 +119,22 @@ class ExtractError(unittest.TestCase):
         body = json.dumps({"error": {"code": 42}})
         self.assertIn("42", api._extract_error(body))
 
+    def test_an_error_wrapped_in_an_array(self):
+        """Google's 503 arrives this way, and .get() on a list raises."""
+        body = json.dumps([{"error": {"code": 503,
+                                      "message": "The model is overloaded."}}])
+        self.assertEqual(api._extract_error(body), "The model is overloaded.")
+
     def test_a_body_that_is_not_json(self):
         self.assertEqual(api._extract_error("<html>502</html>"), "<html>502</html>")
 
+    def test_no_shape_at_all_still_comes_back_as_a_string(self):
+        """It runs while an ApiError is being raised: throwing here would
+        escape the `except ApiError` holding the raw transcript."""
+        for body in ("[]", "[1, 2]", '"a string"', "null", "17"):
+            with self.subTest(body=body):
+                self.assertIsInstance(api._extract_error(body), str)
+
     def test_a_wall_of_html_is_cut_short(self):
         self.assertEqual(len(api._extract_error("x" * 5000)), 300)
 
@@ -384,6 +397,37 @@ class Cleanup(DikteTest):
         self.assertEqual(sent_json(calls[0])["reasoning"],
                          {"effort": "high", "exclude": True})
 
+    def test_gemini_takes_openai_s_flat_field_rather_than_the_object(self):
+        _, calls = self.call(chat_reply("Hello."), reasoning="low",
+                             provider="gemini", service="Google AI Studio")
+        payload = sent_json(calls[0])
+        self.assertEqual(payload["reasoning_effort"], "low")
+        self.assertNotIn("reasoning", payload)
+
+    def test_off_is_asked_for_as_the_lowest_rung_google_actually_has(self):
+        """Sending "none" is a 400, and Flash left alone thinks."""
+        _, calls = self.call(chat_reply("Hello."), reasoning="none",
+                             provider="gemini", service="Google AI Studio")
+        self.assertEqual(sent_json(calls[0])["reasoning_effort"], "minimal")
+
+    def test_a_rung_google_does_not_have_lands_on_the_nearest_one(self):
+        for asked in ("xhigh", "max"):
+            with self.subTest(asked=asked):
+                _, calls = self.call(chat_reply("Hello."), reasoning=asked,
+                                     provider="gemini", service="Google AI Studio")
+                self.assertEqual(sent_json(calls[0])["reasoning_effort"], "high")
+
+    def test_gemini_left_on_the_model_s_own_default_is_told_nothing(self):
+        _, calls = self.call(chat_reply("Hello."), provider="gemini",
+                             service="Google AI Studio")
+        self.assertNotIn("reasoning_effort", sent_json(calls[0]))
+
+    def test_a_missing_gemini_key_says_google_ai_studio(self):
+        with self.assertRaises(api.ApiError) as caught:
+            api.cleanup("hello", "", "gemini-3.5-flash-lite", "prompt",
+                        provider="gemini", service="Google AI Studio")
+        self.assertIn("Google AI Studio", str(caught.exception))
+
     def test_a_local_base_url(self):
         _, calls = self.call(chat_reply("Hello."), base_url="http://localhost:1234/v1")
         self.assertEqual(calls[0].full_url, "http://localhost:1234/v1/chat/completions")
@@ -513,6 +557,40 @@ class ModelLists(DikteTest):
             api.openai_models("", api.GROQ_URL, "Groq")
         self.assertIn("Groq", str(caught.exception))
 
+    def test_gemini_keeps_only_the_models_that_answer_a_chat_request(self):
+        with fake_urlopen({"data": [{"id": "gemini-3.5-flash"},
+                                    {"id": "text-embedding-004"},
+                                    {"id": "imagen-4.0"},
+                                    {"id": "gemini-2.5-flash-lite"}]}) as calls:
+            models = api.gemini_models("AIza-test")
+        self.assertEqual(calls[0].full_url,
+                         "https://generativelanguage.googleapis.com/v1beta/openai/models")
+        self.assertEqual(models, ["gemini-2.5-flash-lite", "gemini-3.5-flash"])
+
+    def test_the_long_form_of_an_id_is_shortened_to_what_a_request_wants(self):
+        with fake_urlopen({"data": [{"id": "models/gemini-3.5-flash-lite"}]}):
+            self.assertEqual(api.gemini_models("AIza-test"),
+                             ["gemini-3.5-flash-lite"])
+
+    def test_a_gemini_id_that_is_not_a_chat_model_is_left_out(self):
+        """Google names its pictures and its voices `gemini` too."""
+        with fake_urlopen({"data": [{"id": "gemini-3.5-flash"},
+                                    {"id": "gemini-embedding-001"},
+                                    {"id": "gemini-2.5-flash-image"},
+                                    {"id": "gemini-2.5-flash-preview-tts"},
+                                    {"id": "gemini-2.5-native-audio"}]}):
+            self.assertEqual(api.gemini_models("AIza-test"), ["gemini-3.5-flash"])
+
+    def test_gemini_sends_the_key_as_a_bearer_token(self):
+        with fake_urlopen({"data": []}) as calls:
+            api.gemini_models("AIza-test")
+        self.assertEqual(calls[0].get_header("Authorization"), "Bearer AIza-test")
+
+    def test_a_missing_gemini_key_says_google_ai_studio(self):
+        with self.assertRaises(api.ApiError) as caught:
+            api.gemini_models("")
+        self.assertIn("Google AI Studio", str(caught.exception))
+
 
 if __name__ == "__main__":
     unittest.main()

tests/test_assistant.py

@@ -97,15 +97,38 @@ class Provider(DikteTest):
     def test_what_each_one_runs(self):
         self.assertEqual(assistant.executable("claude"), "claude")
         self.assertEqual(assistant.executable("codex"), "codex")
+        self.assertEqual(assistant.executable("agy"), "agy")
         self.assertEqual(assistant.executable("openrouter"), "")
+        self.assertEqual(assistant.executable("opencode"), "")
+
+    def test_the_model_recorded_is_the_one_that_answered(self):
+        """The history used to write Claude's setting whoever had answered."""
+        self.assertEqual(assistant.model(self.config()), "sonnet")
+        self.assertEqual(
+            assistant.model(self.config(assistant_provider="codex")), "codex")
+        self.assertEqual(
+            assistant.model(self.config(assistant_provider="agy")), "agy")
+        self.assertEqual(
+            assistant.model(self.config(assistant_provider="agy",
+                                        assistant_agy_model="gemini-3.1-pro-low")),
+            "gemini-3.1-pro-low")
+        self.assertEqual(
+            assistant.model(self.config(assistant_provider="openrouter")),
+            "google/gemini-3.5-flash")
 
     def test_what_each_one_is_called(self):
         self.assertEqual(assistant.display_name(self.config()), "Claude")
-        self.assertEqual(
+        for name, called in (("codex", "Codex"), ("agy", "Antigravity"),
-            assistant.display_name(self.config(assistant_provider="codex")), "Codex")
+                             ("openrouter", "OpenRouter"),
-        self.assertEqual(
+                             ("opencode", "OpenCode Go")):
-            assistant.display_name(self.config(assistant_provider="openrouter")),
+            with self.subTest(name=name):
-            "OpenRouter")
+                self.assertEqual(
+                    assistant.display_name(self.config(assistant_provider=name)),
+                    called)
+
+    def test_every_provider_has_a_name_to_be_called_by(self):
+        """_conclude writes its errors in it, so a gap here is a bare id."""
+        self.assertEqual(set(assistant.SERVICES), set(assistant.PROVIDERS))
 
 
 class Effort(unittest.TestCase):
@@ -113,21 +136,26 @@ class Effort(unittest.TestCase):
 
     def test_the_scales_cover_the_same_settings(self):
         self.assertEqual(set(assistant.CLAUDE_EFFORT), set(assistant.CODEX_EFFORT))
+        self.assertEqual(set(assistant.CLAUDE_EFFORT), set(assistant.AGY_EFFORT))
 
-    def test_codex_has_no_rung_above_high(self):
+    def test_neither_codex_nor_agy_has_a_rung_above_high(self):
-        self.assertEqual(assistant.CODEX_EFFORT["xhigh"], "high")
+        for scale in (assistant.CODEX_EFFORT, assistant.AGY_EFFORT):
-        self.assertEqual(assistant.CODEX_EFFORT["max"], "high")
+            self.assertEqual(scale["xhigh"], "high")
+            self.assertEqual(scale["max"], "high")
 
-    def test_neither_one_asks_for_a_rung_below_low(self):
+    def test_none_of_them_asks_for_a_rung_below_low(self):
         # Claude has none; Codex has one, but calls it "minimal" on the older
-        # models and "none" on the newer ones, and refuses the wrong word.
+        # models and "none" on the newer ones, and refuses the wrong word; agy
-        for scale in (assistant.CLAUDE_EFFORT, assistant.CODEX_EFFORT):
+        # has three rungs and no word for off at all.
+        for scale in (assistant.CLAUDE_EFFORT, assistant.CODEX_EFFORT,
+                      assistant.AGY_EFFORT):
             self.assertEqual(scale["none"], "low")
             self.assertEqual(scale["minimal"], "low")
 
     def test_an_empty_setting_asks_for_nothing(self):
-        self.assertEqual(assistant.CLAUDE_EFFORT.get("", ""), "")
+        for scale in (assistant.CLAUDE_EFFORT, assistant.CODEX_EFFORT,
-        self.assertEqual(assistant.CODEX_EFFORT.get("", ""), "")
+                      assistant.AGY_EFFORT):
+            self.assertEqual(scale.get("", ""), "")
 
 
 class Session(DikteTest):
@@ -276,25 +304,32 @@ class Conclude(DikteTest):
 
     def test_an_answer_and_its_session(self):
         answer, warning = assistant._conclude(
-            self.found(answer="done", session="abc"), 0, "", "", "Claude")
+            self.found(answer="done", session="abc"), 0, "", "", "claude")
         self.assertEqual(answer, "done")
         self.assertEqual(warning, "")
         self.assertEqual(assistant.read_session("claude", 1800), "abc")
 
-    def test_codex_stores_under_its_own_name(self):
+    def test_each_one_stores_under_its_own_name(self):
-        assistant._conclude(self.found(answer="done", session="t-1"), 0, "",
+        for name, session in (("codex", "t-1"), ("agy", "c-9")):
-                            "", "Codex")
+            with self.subTest(name=name):
-        self.assertEqual(assistant.read_session("codex", 1800), "t-1")
+                assistant._conclude(self.found(answer="done", session=session),
+                                    0, "", "", name)
+                self.assertEqual(assistant.read_session(name, 1800), session)
+
+    def test_the_error_is_written_in_the_provider_s_own_name(self):
+        with self.assertRaises(assistant.AssistantError) as caught:
+            assistant._conclude(self.found(), 1, "", "", "agy")
+        self.assertIn("Antigravity", str(caught.exception))
 
     def test_a_non_zero_exit_with_nothing_to_show_for_it(self):
         with self.assertRaises(assistant.AssistantError) as caught:
-            assistant._conclude(self.found(), 1, "it all went wrong\n", "", "Claude")
+            assistant._conclude(self.found(), 1, "it all went wrong\n", "", "claude")
         self.assertIn("it all went wrong", str(caught.exception))
 
     def test_a_session_that_is_gone_is_raised_apart(self):
         with self.assertRaises(assistant._SessionGone):
             assistant._conclude(self.found(), 1, "session abc not found",
-                                "abc", "Claude")
+                                "abc", "claude")
 
     def test_the_recovery_no_longer_hangs_on_the_words_the_cli_chose(self):
         # The complaint used to be matched by substring, which a CLI update or
@@ -321,11 +356,11 @@ class Conclude(DikteTest):
     def test_a_session_that_is_gone_only_matters_when_one_was_resumed(self):
         with self.assertRaises(assistant.AssistantError):
             assistant._conclude(self.found(), 1, "session abc not found",
-                                "", "Claude")
+                                "", "claude")
 
     def test_an_answer_survives_a_non_zero_exit(self):
         answer, _ = assistant._conclude(self.found(answer="done"), 1, "noise",
-                                        "", "Claude")
+                                        "", "claude")
         self.assertEqual(answer, "done")
 
     def test_an_answer_on_a_resumed_session_is_kept_rather_than_retried(self):
@@ -336,12 +371,12 @@ class Conclude(DikteTest):
     def test_a_reported_failure_with_no_answer(self):
         with self.assertRaises(assistant.AssistantError) as caught:
             assistant._conclude(self.found(failure="the model refused"), 0, "",
-                                "", "Claude")
+                                "", "claude")
         self.assertIn("refused", str(caught.exception))
 
     def test_a_run_that_said_nothing_at_all(self):
         with self.assertRaises(assistant.AssistantError) as caught:
-            assistant._conclude(self.found(), 0, "", "", "Codex")
+            assistant._conclude(self.found(), 0, "", "", "codex")
         self.assertIn("Codex", str(caught.exception))
 
 
@@ -565,6 +600,109 @@ class AskCodex(DikteTest):
         self.assertIn("quota", str(caught.exception))
 
 
+class AskAgy(DikteTest):
+    """agy's stream is shaped nothing like the other two: the key is `event`,
+    the answer arrives whole in `result.response`, and the conversation to
+    resume is named in the first line rather than the last."""
+
+    def run_ask(self, conf=None, events=None, session=""):
+        conf = conf or self.config(assistant_provider="agy")
+        proc = FakeCli(events or [
+            {"event": "init", "conversation_id": "c-9", "init": {"cwd": "/home"}},
+            {"event": "result",
+             "result": {"conversation_id": "c-9", "status": "SUCCESS",
+                        "response": "  done  "}},
+        ])
+        stages = []
+        with only_these_tools("agy"), \
+                mock.patch.object(subprocess, "Popen", return_value=proc) as popen:
+            result = assistant._ask_agy("book it", conf, session,
+                                        stages.append, None)
+        return result, popen.call_args.args[0], stages
+
+    def test_the_answer_comes_back_stripped(self):
+        (answer, warning), _, _ = self.run_ask()
+        self.assertEqual(answer, "done")
+        self.assertEqual(warning, "")
+
+    def test_the_instruction_is_kept_apart_from_the_command(self):
+        """agy takes no system prompt, so the two must not read as one."""
+        conf = self.config(assistant_provider="agy")
+        _, cmd, _ = self.run_ask(conf)
+        body = cmd[cmd.index("-p") + 1]
+        self.assertTrue(body.startswith(conf.assistant_prompt()))
+        self.assertIn("\n\n---\n\n", body)
+        self.assertTrue(body.endswith("book it"))
+
+    def test_a_first_command_starts_a_project_of_its_own(self):
+        """Without it agy works in whichever project it was last in."""
+        _, cmd, _ = self.run_ask()
+        self.assertIn("--new-project", cmd)
+        self.assertNotIn("--conversation", cmd)
+
+    def test_a_second_command_carries_the_conversation_rather_than_starting_one(self):
+        _, cmd, _ = self.run_ask(session="c-9")
+        self.assertEqual(cmd[cmd.index("--conversation") + 1], "c-9")
+        self.assertNotIn("--new-project", cmd)
+
+    def test_the_conversation_is_kept_under_agy_s_own_name(self):
+        self.run_ask()
+        self.assertEqual(assistant.read_session("agy", 1800), "c-9")
+
+    def test_it_is_not_left_to_give_up_before_the_caller_does(self):
+        conf = self.config(assistant_provider="agy", assistant_timeout=90)
+        _, cmd, _ = self.run_ask(conf)
+        self.assertEqual(cmd[cmd.index("--print-timeout") + 1], "90s")
+
+    def test_no_model_named_means_whatever_agy_is_set_to(self):
+        _, cmd, _ = self.run_ask()
+        self.assertNotIn("--model", cmd)
+
+    def test_a_model_of_your_own(self):
+        _, cmd, _ = self.run_ask(
+            self.config(assistant_provider="agy",
+                        assistant_agy_model="gemini-3.1-pro-low"))
+        self.assertEqual(cmd[cmd.index("--model") + 1], "gemini-3.1-pro-low")
+
+    def test_a_tool_is_named_in_the_corner_as_it_starts(self):
+        _, _, stages = self.run_ask(events=[
+            {"event": "step_update",
+             "step_update": {"step_type": "tool", "state": "ACTIVE",
+                             "tool_name": "run_command"}},
+            {"event": "step_update",
+             "step_update": {"step_type": "tool", "state": "DONE",
+                             "tool_name": "run_command"}},
+            {"event": "result",
+             "result": {"status": "SUCCESS", "response": "done"}},
+        ])
+        self.assertEqual(stages, ["Running a command…"])
+
+    def test_the_two_dozen_browser_tools_are_one_line_between_them(self):
+        _, _, stages = self.run_ask(events=[
+            {"event": "step_update",
+             "step_update": {"step_type": "tool", "state": "ACTIVE",
+                             "tool_name": "browser_click_element"}},
+            {"event": "result",
+             "result": {"status": "SUCCESS", "response": "done"}},
+        ])
+        self.assertEqual(stages, ["Working in the browser…"])
+
+    def test_a_turn_that_did_not_succeed_is_a_failure_rather_than_an_answer(self):
+        with self.assertRaises(assistant.AssistantError) as caught:
+            self.run_ask(events=[
+                {"event": "result",
+                 "result": {"status": "ERROR", "response": "the model refused"}},
+            ])
+        self.assertIn("refused", str(caught.exception))
+
+    def test_a_failure_with_nothing_to_say_is_still_named(self):
+        with self.assertRaises(assistant.AssistantError) as caught:
+            self.run_ask(events=[
+                {"event": "result", "result": {"status": "ERROR"}},
+            ])
+        self.assertIn("Antigravity", str(caught.exception))
+
+
 class AskOpenRouter(DikteTest):
     def test_a_question_and_an_answer(self):
         conf = self.config(assistant_provider="openrouter",
@@ -600,6 +738,41 @@ class AskOpenRouter(DikteTest):
             assistant.ask("when is it", conf)
 
 
+class AskOpenCode(DikteTest):
+    def test_a_question_and_an_answer(self):
+        conf = self.config(assistant_provider="opencode",
+                           opencode_api_key="opencode-test-key")
+        with fake_urlopen({"choices": [{"message": {"content": "on Thursday"}}]}):
+            answer, warning = assistant.ask("when is it", conf)
+        self.assertEqual(answer, "on Thursday")
+        self.assertEqual(warning, "")
+
+    def test_the_conversation_is_ours_to_keep(self):
+        conf = self.config(assistant_provider="opencode",
+                           opencode_api_key="opencode-test-key")
+        with fake_urlopen({"choices": [{"message": {"content": "on Thursday"}}]}):
+            assistant.ask("when is it", conf)
+        stored = assistant.read_messages("opencode", 1800)
+        self.assertEqual([row["content"] for row in stored],
+                         ["when is it", "on Thursday"])
+
+    def test_the_model_and_endpoint_are_opencode_s_own(self):
+        conf = self.config(assistant_provider="opencode",
+                           opencode_api_key="opencode-test-key",
+                           assistant_opencode_model="glm-5.3")
+        with fake_urlopen({"choices": [{"message": {"content": "on Thursday"}}]}) as calls:
+            assistant.ask("when is it", conf)
+        sent = json.loads(calls[0].data.decode("utf-8"))
+        self.assertEqual(sent["model"], "glm-5.3")
+        self.assertIn("https://opencode.ai/zen/go/v1/chat/completions",
+                      calls[0].full_url)
+
+    def test_an_api_failure_reads_as_an_assistant_failure(self):
+        conf = self.config(assistant_provider="opencode")
+        with self.assertRaises(assistant.AssistantError):
+            assistant.ask("when is it", conf)
+
+
 class Ask(DikteTest):
     def test_a_cli_that_is_not_installed_says_where_to_change_it(self):
         with only_these_tools(), \
@@ -710,5 +883,45 @@ class CodexModels(DikteTest):
             self.assertEqual(assistant.codex_models(), [])
 
 
+class AgyModels(DikteTest):
+    """The model list read off `agy models`: one id, a tab, a display name."""
+
+    LISTING = ("gemini-4-flash-high\tGemini 4 Flash (High)\n"
+               "gemini-4-flash-low\tGemini 4 Flash (Low)\n"
+               "a line with no tab is not a model\n"
+               "\ta tab with no id in front of it is not one either\n")
+
+    def models(self, reply, code=0):
+        with only_these_tools("agy"), \
+                mock.patch.object(subprocess, "run",
+                                  return_value=FakeCompleted(
+                                      returncode=code, stdout=reply)) as run:
+            found = assistant.agy_models()
+        self.run_call = run
+        return found
+
+    def test_the_listing_arrives_in_agy_s_own_order(self):
+        found = self.models(self.LISTING)
+        self.assertEqual(found, ["gemini-4-flash-high", "gemini-4-flash-low"])
+        self.assertEqual(self.run_call.call_args.args[0], ["agy", "models"])
+
+    def test_an_agy_that_is_not_installed_is_not_run(self):
+        with only_these_tools(), \
+                mock.patch.object(subprocess, "run") as run:
+            self.assertEqual(assistant.agy_models(), [])
+        run.assert_not_called()
+
+    def test_a_call_that_failed_answers_with_nothing(self):
+        self.assertEqual(self.models("error: not logged in", code=1), [])
+        self.assertEqual(self.models(""), [])
+
+    def test_an_agy_that_hangs_is_given_up_on(self):
+        with only_these_tools("agy"), \
+                mock.patch.object(subprocess, "run",
+                                  side_effect=subprocess.TimeoutExpired(
+                                      ["agy"], 30)):
+            self.assertEqual(assistant.agy_models(), [])
+
+
 if __name__ == "__main__":
     unittest.main()

tests/test_cleanup.py

@@ -57,7 +57,10 @@ class Provider(DikteTest):
     def test_what_each_one_runs(self):
         self.assertEqual(cleanup.executable("claude"), "claude")
         self.assertEqual(cleanup.executable("codex"), "codex")
+        self.assertEqual(cleanup.executable("agy"), "agy")
         self.assertEqual(cleanup.executable("openrouter"), "")
+        self.assertEqual(cleanup.executable("gemini"), "")
+        self.assertEqual(cleanup.executable("opencode"), "")
 
     def test_the_model_named_in_the_history_is_the_one_that_did_it(self):
         self.assertEqual(cleanup.model(self.config(cleanup_model="some/model")),
@@ -73,6 +76,19 @@ class Provider(DikteTest):
         self.assertEqual(
             cleanup.model(self.config(cleanup_provider="codex",
                                       cleanup_codex_model="gpt-5.4")), "gpt-5.4")
+        self.assertEqual(
+            cleanup.model(self.config(cleanup_provider="gemini")),
+            "gemini-3.5-flash-lite")
+        # Antigravity is left on its own default the way Codex is.
+        self.assertEqual(
+            cleanup.model(self.config(cleanup_provider="agy")), "agy")
+        self.assertEqual(
+            cleanup.model(self.config(cleanup_provider="agy",
+                                      cleanup_agy_model="gemini-3.7-flash-low")),
+            "gemini-3.7-flash-low")
+        self.assertEqual(
+            cleanup.model(self.config(cleanup_provider="opencode",
+                                      cleanup_opencode_model="glm-5.3")), "glm-5.3")
 
 
 class OpenRouter(DikteTest):
@@ -94,6 +110,72 @@ class OpenRouter(DikteTest):
         self.assertEqual(calls, [])
 
 
+class OpenCode(DikteTest):
+    def test_it_is_one_request_with_the_settings_as_they_were(self):
+        conf = self.config(cleanup_provider="opencode",
+                           opencode_api_key="opencode-test-key",
+                           cleanup_opencode_model="some/model",
+                           cleanup_reasoning="low")
+        with mock.patch.object(api, "cleanup", return_value="Done.") as call:
+            self.assertEqual(cleanup.run("uh, done", conf, "the rules"), "Done.")
+        text, key, model, prompt = call.call_args.args
+        self.assertEqual((text, key, model, prompt),
+                         ("uh, done", "opencode-test-key", "some/model", "the rules"))
+        self.assertEqual(call.call_args.kwargs["reasoning"], "low")
+        self.assertEqual(call.call_args.kwargs["provider"], "opencode")
+        self.assertEqual(call.call_args.kwargs["service"], "OpenCode Go")
+        self.assertEqual(call.call_args.kwargs["base_url"],
+                         "https://opencode.ai/zen/go/v1")
+
+    def test_no_cli_is_started_for_it(self):
+        conf = self.config(cleanup_provider="opencode",
+                           opencode_api_key="opencode-test-key")
+        patcher, calls = fake_cli(stdout="never")
+        with patcher, mock.patch.object(api, "cleanup", return_value="Done."):
+            cleanup.run("uh, done", conf, "the rules")
+        self.assertEqual(calls, [])
+
+
+class GoogleAiStudio(DikteTest):
+    """Cleanup over Google's OpenAI-compatible endpoint: one request, no CLI."""
+
+    def setUp(self):
+        super().setUp()
+        self.conf = self.config(cleanup_provider="gemini",
+                                gemini_api_key="AIza-test")
+
+    def test_it_goes_to_google_with_the_settings_as_they_were(self):
+        self.conf["cleanup_reasoning"] = "none"
+        with fake_urlopen(chat_reply("Done.")) as calls:
+            self.assertEqual(cleanup.run("uh, done", self.conf, "the rules"),
+                             "Done.")
+        self.assertEqual(
+            calls[0].full_url,
+            "https://generativelanguage.googleapis.com/v1beta/openai/chat/completions")
+        payload = sent_json(calls[0])
+        self.assertEqual(payload["model"], "gemini-3.5-flash-lite")
+        self.assertEqual(payload["reasoning_effort"], "minimal")
+        self.assertIn("uh, done", payload["messages"][1]["content"])
+
+    def test_the_key_travels_as_a_bearer_token(self):
+        with fake_urlopen(chat_reply("Done.")) as calls:
+            cleanup.run("uh, done", self.conf, "the rules")
+        self.assertEqual(calls[0].get_header("Authorization"), "Bearer AIza-test")
+
+    def test_a_missing_key_names_google_rather_than_openrouter(self):
+        self.conf["gemini_api_key"] = ""
+        with mock.patch.dict(os.environ, {}, clear=True), \
+                self.assertRaises(api.ApiError) as caught:
+            cleanup.run("uh, done", self.conf, "the rules")
+        self.assertIn("Google AI Studio", str(caught.exception))
+
+    def test_no_cli_is_started_for_it(self):
+        patcher, calls = fake_cli(stdout="never")
+        with patcher, fake_urlopen(chat_reply("Done.")):
+            cleanup.run("uh, done", self.conf, "the rules")
+        self.assertEqual(calls, [])
+
+
 class ClaudeCode(DikteTest):
     def setUp(self):
         super().setUp()
@@ -221,6 +303,63 @@ class Codex(DikteTest):
             self.run_cleanup(stdout="tokens used 400", last_message="")
 
 
+class Antigravity(DikteTest):
+    def setUp(self):
+        super().setUp()
+        self.conf = self.config(cleanup_provider="agy")
+        self.patch_attr(cleanup.shutil, "which", lambda name: f"/usr/bin/{name}")
+
+    def run_cleanup(self, text="uh, book it", **kwargs):
+        patcher, calls = fake_cli(**kwargs)
+        with patcher:
+            answer = cleanup.run(text, self.conf, "the rules")
+        return answer, calls[0]
+
+    def test_the_rules_ride_in_front_of_the_transcript(self):
+        answer, cmd = self.run_cleanup(stdout="Book it.\n")
+        self.assertEqual(answer, "Book it.")
+        self.assertEqual(cmd[0], "agy")
+        self.assertEqual(cmd[cmd.index("-p") + 1],
+                         "the rules\n\n---\n\n<transcript>\nuh, book it\n</transcript>")
+
+    def test_it_starts_somewhere_of_its_own_and_takes_no_slash_commands(self):
+        """Without --new-project agy works in whichever project it was last in."""
+        _, cmd = self.run_cleanup(stdout="Book it.")
+        self.assertIn("--new-project", cmd)
+        self.assertIn("--disable-slash-commands", cmd)
+        self.assertEqual(cmd[cmd.index("--output-format") + 1], "text")
+
+    def test_it_is_not_left_to_give_up_before_the_caller_does(self):
+        _, cmd = self.run_cleanup(stdout="Book it.")
+        self.assertEqual(cmd[cmd.index("--print-timeout") + 1], "180s")
+
+    def test_the_model_is_left_alone_until_one_is_typed_in(self):
+        _, cmd = self.run_cleanup(stdout="Book it.")
+        self.assertNotIn("--model", cmd)
+        self.conf["cleanup_agy_model"] = "gemini-3.7-flash-low"
+        _, cmd = self.run_cleanup(stdout="Book it.")
+        self.assertEqual(cmd[cmd.index("--model") + 1], "gemini-3.7-flash-low")
+
+    def test_the_thinking_setting_lands_on_the_nearest_rung_agy_has(self):
+        self.conf["cleanup_reasoning"] = "max"
+        _, cmd = self.run_cleanup(stdout="Book it.")
+        self.assertEqual(cmd[cmd.index("--effort") + 1], "high")
+
+    def test_no_thinking_setting_means_no_flag(self):
+        _, cmd = self.run_cleanup(stdout="Book it.")
+        self.assertNotIn("--effort", cmd)
+
+    def test_an_answer_of_nothing_is_a_failure_rather_than_an_empty_paste(self):
+        with self.assertRaises(cleanup.CleanupError):
+            self.run_cleanup(stdout="   ")
+
+    def test_a_program_that_is_not_installed_says_so_before_running_anything(self):
+        self.patch_attr(cleanup.shutil, "which", lambda name: "")
+        with self.assertRaises(cleanup.CleanupError) as caught:
+            self.run_cleanup(stdout="Book it.")
+        self.assertIn("agy", str(caught.exception))
+
+
 if __name__ == "__main__":
     unittest.main()
 

tests/test_cli.py

@@ -9,12 +9,14 @@ socket is faked, and everything that runs locally runs for real.
 import contextlib
 import io
 import json
+import sys
 import unittest
 import webbrowser
 from typing import ClassVar
 from unittest import mock
 
 from dikte import audio
+from dikte import cleanup
 from dikte import cli
 from dikte import config as cfg
 from dikte import ggml
@@ -423,6 +425,16 @@ class Providers(DikteTest):
         self.assertIn("groq", out)
         self.assertIn("Groq", out)
 
+    def test_opencode_is_a_choice_and_reports_under_its_own_name(self):
+        parser = cli.build_parser()
+        self.assertEqual(
+            parser.parse_args(["test-key", "opencode"]).which, "opencode")
+        self.write_config({"opencode_api_key": "opencode-test"})
+        with fake_urlopen({"data": [{"id": "deepseek-v4-flash"}]}):
+            code, out, _ = self.run_cmd(cli.cmd_test_key, which="opencode")
+        self.assertEqual(code, 0)
+        self.assertIn("opencode: connection works, 1 models visible", out)
+
 
 class Updates(DikteTest):
     """`dikte update` looks, says what it found, and installs nothing."""
@@ -503,6 +515,35 @@ class Doctor(DikteTest):
         self.assertIn("OpenRouter key, cleaning up on some/model",
                       self.run_doctor(as_json=False, cleanup_model="some/model"))
 
+    def test_cleanup_on_opencode_is_a_question_about_its_own_key(self):
+        reply = self.run_doctor(cleanup_provider="opencode",
+                                cleanup_opencode_model="glm-5.3")
+        self.assertEqual(reply["cleanup"]["provider"], "opencode")
+        self.assertEqual(reply["cleanup"]["model"], "glm-5.3")
+        self.assertIn("OpenCode Go key, cleaning up on glm-5.3",
+                      self.run_doctor(as_json=False, cleanup_provider="opencode",
+                                      cleanup_opencode_model="glm-5.3"))
+
+    def test_it_survives_every_provider_cleanup_can_be_set_to(self):
+        """It used to raise KeyError on the local model, whose executable is ""."""
+        for name in cleanup.PROVIDERS:
+            with self.subTest(provider=name):
+                reply = self.run_doctor(cleanup_provider=name)
+                self.assertEqual(reply["cleanup"]["provider"], name)
+                self.run_doctor(as_json=False, cleanup_provider=name)
+
+    def test_a_provider_with_no_key_to_check_says_so_rather_than_no(self):
+        """A CLI needs none, so `false` there would read as one gone missing."""
+        self.assertIsNone(self.run_doctor(cleanup_provider="claude")["cleanup"]["key"])
+        self.assertIsNone(self.run_doctor(cleanup_provider="local")["cleanup"]["key"])
+        self.assertIs(self.run_doctor(cleanup_provider="gemini")["cleanup"]["key"],
+                      False)
+
+    def test_cleanup_on_google_is_a_question_about_its_own_key(self):
+        line = self.run_doctor(as_json=False, cleanup_provider="gemini",
+                               cleanup_gemini_model="gemini-2.5-flash")
+        self.assertIn("Google AI Studio key, cleaning up on gemini-2.5-flash", line)
+
     def test_it_asks_after_the_programs_this_desktop_actually_uses(self):
         """A missing ydotool on a Mac is a red mark with nothing behind it."""
         with mock.patch.object(cli.paste, "desktop", return_value=paste.MACOS):
@@ -628,7 +669,11 @@ class WithoutAnInstance(DikteTest):
     def run_verb(self, argv):
         # launch_gui replaces this process with the application, so it never
         # comes back in real use and must not be allowed to here.
+        # `ask` with no text reads what was piped in, and the runner's own
+        # stdin is not that: under pytest it is an object that refuses to be
+        # read at all.
         with mock.patch.object(ipc, "send", return_value=None), \
+                mock.patch.object(sys, "stdin", io.StringIO()), \
                 mock.patch.object(cli, "launch_gui") as launch, \
                 captured() as (out, err):
             code = cli.run(argv)

tests/test_config.py

@@ -187,6 +187,10 @@ class Keys(DikteTest):
     def test_every_provider_falls_back_to_the_variable_of_its_own_name(self):
         with mock.patch.dict(os.environ, {"GROQ_API_KEY": "gsk-env"}):
             self.assertEqual(cfg.Config().groq_key(), "gsk-env")
+        with mock.patch.dict(os.environ, {"GEMINI_API_KEY": "AIza-env"}):
+            self.assertEqual(cfg.Config().gemini_key(), "AIza-env")
+        with mock.patch.dict(os.environ, {"OPENCODE_API_KEY": "opencode-env"}):
+            self.assertEqual(cfg.Config().opencode_key(), "opencode-env")
 
 
 class TranscribeTarget(DikteTest):
@@ -574,6 +578,19 @@ class Defaults(unittest.TestCase):
     def test_the_keys_ship_empty(self):
         self.assertEqual(cfg.DEFAULTS["openai_api_key"], "")
         self.assertEqual(cfg.DEFAULTS["openrouter_api_key"], "")
+        self.assertEqual(cfg.DEFAULTS["gemini_api_key"], "")
+        self.assertEqual(cfg.DEFAULTS["opencode_api_key"], "")
+
+    def test_google_ai_studio_is_a_cleanup_provider_and_not_a_transcriber(self):
+        """Its compatible endpoint has no /audio/transcriptions behind it."""
+        self.assertNotIn("gemini", cfg.TRANSCRIBERS)
+        self.assertIn("gemini", cleanup.PROVIDERS)
+
+    def test_opencode_ships_on_its_own_endpoint(self):
+        self.assertEqual(cfg.DEFAULTS["opencode_base_url"],
+                         "https://opencode.ai/zen/go/v1")
+        self.assertEqual(cfg.DEFAULTS["cleanup_opencode_model"], "deepseek-v4-flash")
+        self.assertEqual(cfg.DEFAULTS["assistant_opencode_model"], "deepseek-v4-flash")
 
     def test_every_language_specific_prompt_has_both_languages(self):
         for name in ("CLEANUP_PROMPT", "FILE_CLEANUP_PROMPT", "MEETING_PROMPT",

tests/test_ggml.py

@@ -238,6 +238,45 @@ class InstallProgram(Local):
                                "whisper-bin-ubuntu-x64.tar.gz")
         self.assertTrue(urls[1].endswith("whisper-bin-ubuntu-x64.tar.gz"))
 
+    def test_the_nightly_pointer_is_followed_to_where_the_builds_are(self):
+        """llama.cpp's latest release carries a tag name, not the binaries."""
+        self.patch_attr(ggml, "_arch", lambda: "x64")
+        self.patch_attr(ggml, "_has_vulkan", lambda: False)
+        marker = self.release(ggml.NIGHTLY_TAG)
+        nightly = dict(self.release("llama-b10809-bin-ubuntu-x64.tar.gz"),
+                       tag_name="b10809")
+
+        def opener(request, timeout=None):
+            url = request.full_url
+            if url.endswith("/releases/latest"):
+                return json_body(marker)
+            if url.endswith("/releases/tags/b10809"):
+                return json_body(nightly)
+            if url.endswith(ggml.NIGHTLY_TAG):
+                return body(b"b10809\n")
+            return body(self.archive)
+
+        with mock.patch("urllib.request.urlopen", side_effect=opener):
+            tag, found = ggml._pick_asset(ggml.LLAMA)
+        self.assertEqual(tag, "b10809")
+        self.assertEqual(found.name, "llama-b10809-bin-ubuntu-x64.tar.gz")
+
+    def test_without_a_pointer_the_newest_release_that_has_a_build_is_taken(self):
+        self.patch_attr(ggml, "_arch", lambda: "x64")
+        self.patch_attr(ggml, "_has_vulkan", lambda: False)
+        marker = self.release("source.zip")
+        listing = [dict(self.release("llama-b2-bin-win-cpu-x64.zip"), tag_name="b2"),
+                   dict(self.release("llama-b1-bin-ubuntu-x64.tar.gz"), tag_name="b1")]
+
+        def opener(request, timeout=None):
+            url = request.full_url
+            return json_body(listing if "per_page" in url else marker)
+
+        with mock.patch("urllib.request.urlopen", side_effect=opener):
+            tag, found = ggml._pick_asset(ggml.LLAMA)
+        self.assertEqual(tag, "b1")
+        self.assertEqual(found.name, "llama-b1-bin-ubuntu-x64.tar.gz")
+
     def test_a_release_with_nothing_for_this_machine_says_so(self):
         self.patch_attr(ggml, "_arch", lambda: "x64")
         with fake_urlopen(self.release("whisper-bin-Win32.zip")):

tests/test_paths.py

@@ -42,9 +42,12 @@ class Directories(unittest.TestCase):
 
     def test_a_mac_does_not_read_the_xdg_variables(self):
         """A Mac with them set from some other tool still stores in one place."""
-        with mock.patch.dict(os.environ, {"XDG_CONFIG_HOME": "/c"}):
+        # Something no temporary directory can be called: the home this runs
+        # under is a mkdtemp path, and a two-letter needle matched the "/c" in
+        # somebody's TMPDIR rather than the variable being read.
+        with mock.patch.dict(os.environ, {"XDG_CONFIG_HOME": "/xdg-elsewhere"}):
             config_dir, _ = paths.directories("darwin")
-        self.assertNotIn("/c", config_dir.as_posix())
+        self.assertNotIn("xdg-elsewhere", config_dir.as_posix())
 
     def test_windows_keeps_the_models_out_of_the_roaming_profile(self):
         """Settings roam with the account; several gigabytes must not."""

tests/test_ui.py

@@ -8,6 +8,7 @@ next time anybody presses Save. That is the failure this catches.
 
 import os
 import sys
+import time
 import unittest
 from typing import ClassVar
 from unittest import mock
@@ -21,6 +22,7 @@ from dikte import cleanup
 from dikte import config as cfg
 from dikte import ggml
 from dikte import hotkey
+from dikte import hub
 from dikte import ipc
 from dikte import overlay as overlay_module
 from dikte import paste
@@ -28,6 +30,11 @@ from dikte import settings_ui
 from dikte import update
 from tests.support import DikteTest, only_these_tools
 
+# The harness below replaces this method on the class so that opening a window
+# in a test never calls anybody; taken here, before any test runs, so the two
+# tests about what it does when called still have the real one.
+REAL_LOAD_HOSTED_MODELS = settings_ui.SettingsWindow._load_hosted_models
+
 # One application for the whole run; Qt allows no second one.
 _app = QApplication.instance() or QApplication([])
 
@@ -42,6 +49,7 @@ CHANGED = {
     "paste_shortcut": "ctrl+shift+v",
     "restore_clipboard": True,
     "overlay_corner": "top-right",
+    "overlay_screen": "DP-1",
     "max_seconds": 120,
     "skip_silent": False,
     "silence_db": -42.0,
@@ -50,6 +58,7 @@ CHANGED = {
     "openai_api_key": "sk-test-key",
     "groq_api_key": "gsk-test-key",
     "openrouter_api_key": "sk-or-test-key",
+    "gemini_api_key": "AIza-test-key",
     "transcribe_provider": "openrouter",
     "transcribe_model": "whisper-1",
     "groq_transcribe_model": "whisper-large-v3",
@@ -59,6 +68,8 @@ CHANGED = {
     "cleanup_model": "some/other-model",
     "cleanup_claude_model": "opus",
     "cleanup_codex_model": "gpt-5",
+    "cleanup_gemini_model": "gemini-2.5-flash",
+    "cleanup_agy_model": "gemini-3.1-pro-low",
     "cleanup_reasoning": "high",
     "local_model": "ggml-small.bin",
     "local_gpu": False,
@@ -78,6 +89,7 @@ CHANGED = {
     "assistant_codex_model": "gpt-5",
     "assistant_codex_sandbox": "read-only",
     "assistant_openrouter_model": "some/agent-model",
+    "assistant_agy_model": "gemini-3.1-pro-low",
     "assistant_reasoning": "high",
     "assistant_dir": "/tmp",
     "assistant_timeout": 600,
@@ -135,6 +147,10 @@ class Settings(DikteTest):
                                             "_load_transcribe_models"))
         self.enterContext(mock.patch.object(settings_ui.SettingsWindow,
                                             "_load_codex_models"))
+        self.enterContext(mock.patch.object(settings_ui.SettingsWindow,
+                                            "_load_agy_models"))
+        self.enterContext(mock.patch.object(settings_ui.SettingsWindow,
+                                            "_load_hosted_models"))
         # The local model boxes fetch their own list the moment they are shown,
         # from a thread, which is nobody's test failing but a real request.
         self.enterContext(mock.patch.object(settings_ui.LocalModelBox,
@@ -161,7 +177,7 @@ class Settings(DikteTest):
     def test_the_window_opens_with_every_tab_on_it(self):
         window = self.window(cfg.Config())
         tabs = window.findChildren(settings_ui.QTabWidget)[0]
-        self.assertEqual(tabs.count(), 9)
+        self.assertEqual(tabs.count(), 10)
         self.assertEqual(window.windowTitle(), "Dikte Settings")
 
     def test_no_tab_can_stretch_the_window_past_a_small_screen(self):
@@ -261,6 +277,7 @@ class Settings(DikteTest):
         """An OpenRouter id and a Claude alias are not the same field."""
         window = self.window(cfg.Config())
         boxes = {"openrouter": window.cleanup_model_row,
+                 "opencode": window.cleanup_opencode_model_row,
                  "claude": window.cleanup_claude_model,
                  "codex": window.cleanup_codex_model}
         for provider, box in boxes.items():
@@ -284,6 +301,90 @@ class Settings(DikteTest):
         self.assertEqual(window.cleanup_codex_model.currentText(),
                          "my-own-model")
 
+    def test_opencode_answering_refills_both_of_its_boxes(self):
+        """The fetched catalog replaces the built-in list in the cleanup and
+        agent boxes alike, and neither loses what was picked."""
+        conf = self.config(cleanup_opencode_model="my-own-model")
+        window = self.window(conf)
+        window._on_opencode_models_loaded(["glm-9", "kimi-k9"], "")
+        for combo in (window.cleanup_opencode_model,
+                      window.assistant_opencode_model):
+            with self.subTest(combo=combo.objectName() or "combo"):
+                offered = [combo.itemText(i) for i in range(combo.count())]
+                self.assertEqual(offered, ["glm-9", "kimi-k9"])
+        self.assertEqual(window.cleanup_opencode_model.currentText(),
+                         "my-own-model")
+
+    def test_opencode_s_list_arriving_at_open_leaves_the_other_boxes_alone(self):
+        conf = self.config(cleanup_opencode_model="my-own-model",
+                           meeting_model="some/meeting-model")
+        window = self.window(conf)
+        before = [window.meeting_model.itemText(i)
+                  for i in range(window.meeting_model.count())]
+        window._on_hosted_models_loaded("opencode", ["glm-5.3", "kimi-k3"])
+        combo = window.cleanup_opencode_model
+        offered = [combo.itemText(i) for i in range(combo.count())]
+        self.assertEqual(offered, ["glm-5.3", "kimi-k3"])
+        self.assertEqual(combo.currentText(), "my-own-model")
+        self.assertEqual([window.meeting_model.itemText(i)
+                          for i in range(window.meeting_model.count())], before)
+
+    def test_opencode_cleanup_offers_a_fetch_button_of_its_own(self):
+        """The OpenRouter button leaves the screen with its box, so OpenCode Go
+        carries its own."""
+        window = self.window(cfg.Config())
+        window._select_data(window.cleanup_provider, "opencode")
+        self.assertFalse(window.cleanup_opencode_model_row.isHidden())
+        self.assertTrue(window.cleanup_model_row.isHidden())
+
+    def test_agy_answering_refills_both_of_its_boxes(self):
+        """The same arrangement as Codex: both boxes, nothing chosen is lost."""
+        conf = self.config(cleanup_agy_model="my-own-model")
+        window = self.window(conf)
+        window._on_agy_models_loaded(["gemini-4-flash-low", "gemini-4-pro-low"])
+        for combo in (window.cleanup_agy_model, window.assistant_agy_model):
+            with self.subTest(combo=combo.objectName() or "combo"):
+                offered = [combo.itemText(i) for i in range(combo.count())]
+                self.assertEqual(offered[1:],
+                                 ["gemini-4-flash-low", "gemini-4-pro-low"])
+        self.assertEqual(window.cleanup_agy_model.currentText(), "my-own-model")
+
+    def test_openrouter_s_list_arriving_at_open_refills_cleanup_and_meetings(self):
+        conf = self.config(cleanup_model="my/own-model")
+        window = self.window(conf)
+        window._on_hosted_models_loaded("openrouter", ["a/one", "b/two"])
+        for combo in (window.cleanup_model, window.meeting_model):
+            with self.subTest(combo=combo.objectName() or "combo"):
+                offered = [combo.itemText(i) for i in range(combo.count())]
+                self.assertEqual(offered, ["a/one", "b/two"])
+        self.assertEqual(window.cleanup_model.currentText(), "my/own-model")
+
+    def test_google_s_list_arriving_at_open_refills_its_own_box_only(self):
+        window = self.window(self.config(cleanup_gemini_model="gemini-x"))
+        before = window.cleanup_model.count()
+        window._on_hosted_models_loaded("gemini", ["gemini-4-flash"])
+        offered = [window.cleanup_gemini_model.itemText(i)
+                   for i in range(window.cleanup_gemini_model.count())]
+        self.assertEqual(offered, ["gemini-4-flash"])
+        self.assertEqual(window.cleanup_gemini_model.currentText(), "gemini-x")
+        self.assertEqual(window.cleanup_model.count(), before)
+
+    def test_no_key_no_call_home_at_open(self):
+        """Opening Settings is not consent to be talked about to two vendors."""
+        window = self.window(self.config())
+        with mock.patch.dict(os.environ, {}, clear=True), \
+                mock.patch.object(settings_ui.threading, "Thread") as thread:
+            REAL_LOAD_HOSTED_MODELS(window)
+        thread.assert_not_called()
+
+    def test_a_key_on_file_is_fetched_with_at_open(self):
+        window = self.window(self.config(openrouter_api_key="sk-or-x",
+                                         gemini_api_key="AIza-x",
+                                         opencode_api_key="opencode-x"))
+        with mock.patch.object(settings_ui.threading, "Thread") as thread:
+            REAL_LOAD_HOSTED_MODELS(window)
+        self.assertEqual(thread.call_count, 3)
+
     def test_the_update_line_names_the_version_that_is_running(self):
         window = self.window(cfg.Config())
         self.assertIn(settings_ui.__version__, window.update_status.text())
@@ -941,6 +1042,17 @@ class Overlay(DikteTest):
                 widget.show_recording()
                 widget._reposition()
 
+    def test_a_named_screen_is_used_instead_of_the_pointer_screen(self):
+        screen = mock.Mock()
+        screen.name.return_value = "DP-1"
+        screen.availableGeometry.return_value = settings_ui.QRect(1920, 0, 1920, 1080)
+        widget = self.overlay(screen_name="DP-1")
+        with mock.patch.object(QApplication, "screens", return_value=[screen]), \
+                mock.patch.object(QApplication, "screenAt") as screen_at:
+            widget._reposition()
+        screen_at.assert_not_called()
+        self.assertEqual(widget.pos(), QPoint(1948, 995))
+
     def test_a_warning_and_an_error_both_show(self):
         widget = self.overlay()
         widget.show_warning("cleanup failed")
@@ -1001,7 +1113,11 @@ class MeetingSources(DikteTest):
                 mock.patch.object(settings_ui.SettingsWindow,
                                   "_load_transcribe_models"), \
                 mock.patch.object(settings_ui.SettingsWindow,
-                                  "_load_codex_models"):
+                                  "_load_codex_models"), \
+                mock.patch.object(settings_ui.SettingsWindow,
+                                  "_load_agy_models"), \
+                mock.patch.object(settings_ui.SettingsWindow,
+                                  "_load_hosted_models"):
             window = settings_ui.SettingsWindow(cfg.Config())
         self.addCleanup(window.deleteLater)
         self.addCleanup(window.close)
@@ -1033,6 +1149,10 @@ class LocalModels(DikteTest):
         # And one with Codex on it would ask it for its model list.
         self.enterContext(mock.patch.object(settings_ui.SettingsWindow,
                                             "_load_codex_models"))
+        self.enterContext(mock.patch.object(settings_ui.SettingsWindow,
+                                            "_load_agy_models"))
+        self.enterContext(mock.patch.object(settings_ui.SettingsWindow,
+                                            "_load_hosted_models"))
 
     def window(self, conf):
         window = settings_ui.SettingsWindow(conf)
@@ -1096,6 +1216,75 @@ class LocalModels(DikteTest):
                      for row in range(box.repo.count()))
         self.assertGreaterEqual(view.minimumWidth(), widest)
 
+    @staticmethod
+    def _item(name, size=1 << 20):
+        return hub.Item(name, f"https://example.invalid/{name}", size, "")
+
+    def test_a_row_with_nothing_to_fetch_does_not_offer_a_download(self):
+        # The model the settings name is not in the list any more, so its row
+        # was rebuilt from the name alone and carries no file to fetch. The
+        # button stayed lit and the press did nothing at all.
+        box = self.window(self.config(local_llm_model="gone.gguf")).local_llm
+        box.load("gone.gguf", "ggml-org/SmolLM3-3B-GGUF")
+        self.assertEqual(box.selected(), "gone.gguf")
+        self.assertFalse(box.download_button.isEnabled())
+        self.assertIn("gone.gguf", box.status.text())
+        self.assertIn("publisher", box.status.text())
+
+    def test_a_model_without_its_program_does_not_say_it_is_ready(self):
+        # The model runs on the program above it, and "Ready" over a missing
+        # one is what had people asking why nothing transcribed.
+        box = self.window(cfg.Config()).local_whisper
+        path = ggml.whisper_model_path("ggml-small.bin")
+        path.parent.mkdir(parents=True, exist_ok=True)
+        path.write_bytes(b"not really a model")
+        box.load("ggml-small.bin")
+        self.assertFalse(ggml.program_path(ggml.WHISPER))
+        self.assertNotIn("Ready", box.status.text())
+        self.assertIn("program", box.status.text())
+
+    def test_changing_the_publisher_changes_the_model(self):
+        # The model chosen under the old publisher is not published by the new
+        # one. Carried over, it was added back as "not downloaded" and selected
+        # again, and the box looked as though the change had not taken.
+        box = self.window(self.config(local_llm_model="gemma-3-4b-it-Q4_K_M.gguf",
+                                      local_llm_repo="ggml-org/gemma-3-4b-it-GGUF")).local_llm
+        box.load("gemma-3-4b-it-Q4_K_M.gguf", "ggml-org/gemma-3-4b-it-GGUF")
+        box.repo.blockSignals(True)
+        box.repo.setCurrentText("ggml-org/SmolLM3-3B-GGUF")
+        box.repo.blockSignals(False)
+        box._on_listed([("models", [self._item("SmolLM3-Q4_K_M.gguf")],
+                         "ggml-org/SmolLM3-3B-GGUF")], "")
+        self.assertEqual(box.selected(), "SmolLM3-Q4_K_M.gguf")
+        self.assertEqual(box.model.count(), 1)
+
+    def test_a_list_for_a_publisher_that_is_no_longer_chosen_is_dropped(self):
+        # Every change starts its own request, and they do not come back in the
+        # order they went out.
+        box = self.window(cfg.Config()).local_llm
+        box.load("", "ggml-org/SmolLM3-3B-GGUF")
+        box.repo.blockSignals(True)
+        box.repo.setCurrentText("ggml-org/SmolLM3-3B-GGUF")
+        box.repo.blockSignals(False)
+        box._on_listed([("models", [self._item("SmolLM3-Q4_K_M.gguf")],
+                         "ggml-org/SmolLM3-3B-GGUF")], "")
+        box._on_listed([("models", [self._item("gemma-3-4b-it-Q4_K_M.gguf")],
+                         "ggml-org/gemma-3-4b-it-GGUF")], "")
+        self.assertEqual(box.selected(), "SmolLM3-Q4_K_M.gguf")
+
+    def test_the_publisher_box_is_not_asked_on_every_keystroke(self):
+        box = self.window(cfg.Config()).local_llm
+        with mock.patch.object(box, "_fetch_models") as fetch:
+            for text in ("g", "gg", "ggm", "ggml-org/SmolLM3-3B-GGUF"):
+                box.repo.setCurrentText(text)
+            fetch.assert_not_called()
+            box._later.setInterval(0)
+            box._later.start()
+            _app.processEvents()
+            time.sleep(0.05)
+            _app.processEvents()
+        self.assertEqual(fetch.call_count, 1)
+
     def test_only_the_chosen_transcriber_is_on_screen(self):
         window = self.window(self.config(transcribe_provider="openai"))
         self.assertTrue(window.stt_form.isRowVisible(window.transcribe_model_row))
@@ -1113,3 +1302,29 @@ class LocalModels(DikteTest):
         self.assertFalse(window.cleanup_form.isRowVisible(window.cleanup_model_row))
         # Its own thinking box, because the two default to opposite things.
         self.assertFalse(window.cleanup_form.isRowVisible(window.cleanup_reasoning))
+
+    def test_each_cleaner_brings_its_own_model_row_and_no_other(self):
+        window = self.window(cfg.Config())
+        rows = {"openrouter": window.cleanup_model_row,
+                "gemini": window.cleanup_gemini_model_row,
+                "claude": window.cleanup_claude_model,
+                "codex": window.cleanup_codex_model,
+                "agy": window.cleanup_agy_model}
+        for chosen, row in rows.items():
+            with self.subTest(provider=chosen):
+                window._select_data(window.cleanup_provider, chosen)
+                for name, other in rows.items():
+                    self.assertEqual(window.cleanup_form.isRowVisible(other),
+                                     name == chosen)
+
+    def test_each_agent_brings_its_own_box_and_no_other(self):
+        window = self.window(cfg.Config())
+        boxes = {"claude": window.claude_box, "codex": window.codex_box,
+                 "agy": window.agy_box, "openrouter": window.openrouter_box}
+        for chosen, box in boxes.items():
+            with self.subTest(provider=chosen):
+                window._select_data(window.assistant_provider, chosen)
+                for name, other in boxes.items():
+                    # isHidden rather than isVisible: the window itself is never
+                    # shown in a test, so nothing in it is ever visible.
+                    self.assertEqual(other.isHidden(), name != chosen)