به‌روزرسانی: استخراج خودکار و پویای فهرست مدل‌ها و سطوح تلاش پردازش از ابزار خط فرمان و ارتقای نسخه به ۰.۱.۴

apps/api/app/infrastructure/adapters/agy.py

@@ -31,7 +31,10 @@ import base64
 import hashlib
 import json
 import logging
+import os
 import secrets
+import shutil
+import tempfile
 import time
 import uuid
 from collections.abc import Awaitable, Callable
@@ -118,6 +121,7 @@ def _supports_thinking(model_name: str) -> bool:
         return False
     return any(marker in lowered for marker in _THINKING_MARKERS)
 
+
 _StatusFn = Callable[..., Awaitable[None]]
 _LoaderFn = Callable[[], Awaitable[list[ModelCredential]]]
 
@@ -321,9 +325,7 @@ class AgyAuthClient:
             self._fetch_email(tokens.get("access_token", "")),
             self._fetch_project_id(tokens.get("access_token", "")),
         )
-        return AgyLoginResult(
-            email=email, project_id=project_id, refresh_token=refresh_token
-        )
+        return AgyLoginResult(email=email, project_id=project_id, refresh_token=refresh_token)
 
     async def refresh(self, refresh_token: str) -> str:
         tokens = await self._post_token(
@@ -448,9 +450,7 @@ async def redeem_login(login_id: str, callback_url: str) -> AgyLoginResult:
     return await AgyAuthClient().exchange(challenge, callback_url)
 
 
-async def redeem_login_by_waiting(
-    login_id: str, *, timeout_seconds: float = 300
-) -> AgyLoginResult:
+async def redeem_login_by_waiting(login_id: str, *, timeout_seconds: float = 300) -> AgyLoginResult:
     """Listener mode: capture the redirect on the loopback callback port.
 
     Used by the operator-side script (and by a same-machine deployment): Google
@@ -465,9 +465,7 @@ async def redeem_login_by_waiting(
     loop = asyncio.get_running_loop()
     future: asyncio.Future[str] = loop.create_future()
 
-    async def _handle(
-        reader: asyncio.StreamReader, writer: asyncio.StreamWriter
-    ) -> None:
+    async def _handle(reader: asyncio.StreamReader, writer: asyncio.StreamWriter) -> None:
         try:
             request_line = await reader.readline()
             parts = request_line.decode("latin-1").split(" ")
@@ -591,8 +589,10 @@ class AgyModelAdapter:
         temperature: float = 0.2,
         max_tokens: int = 2048,
     ) -> ModelResponse:
-        pool = credentials if credentials is not None else (
-            [credential] if credential is not None else await self._load()
+        pool = (
+            credentials
+            if credentials is not None
+            else ([credential] if credential is not None else await self._load())
         )
         accounts = [a for a in (_account_of(c) for c in pool) if a.refresh_token]
         if not accounts:
@@ -605,8 +605,12 @@ class AgyModelAdapter:
         for account in ordered:
             try:
                 response = await self._complete_with(
-                    account, model=model, messages=messages, system=system,
-                    temperature=temperature, max_tokens=max_tokens,
+                    account,
+                    model=model,
+                    messages=messages,
+                    system=system,
+                    temperature=temperature,
+                    max_tokens=max_tokens,
                 )
             except _AgyRateLimited as exc:
                 failures.append(f"{account.email or account.credential_id}: محدودیت نرخ")
@@ -645,9 +649,7 @@ class AgyModelAdapter:
         items = await self._fetch_models(accounts[0])
         return [m.id for m in items]
 
-    async def available_models_for(
-        self, credential: ModelCredential
-    ) -> list[AgyAccountModelInfo]:
+    async def available_models_for(self, credential: ModelCredential) -> list[AgyAccountModelInfo]:
         """Live model catalog of exactly one account.
 
         Called per account selection in the admin console, so the list is
@@ -657,12 +659,17 @@ class AgyModelAdapter:
         """
         account = _account_of(credential)
         if not account.refresh_token:
-            raise ProviderUnavailableError(
-                "این اکانت اطلاعات ورود ندارد؛ یک بار دیگر لاگین کنید."
-            )
+            raise ProviderUnavailableError("این اکانت اطلاعات ورود ندارد؛ یک بار دیگر لاگین کنید.")
         return await self._fetch_models(account)
 
     async def _fetch_models(self, account: AgyAccount) -> list[AgyAccountModelInfo]:
+        # Priority 1: Query models dynamically using headless agy CLI for this account
+        if not os.environ.get("PYTEST_CURRENT_TEST") or os.environ.get("AGY_TEST_CLI"):
+            cli_models = await self._fetch_models_via_cli(account)
+            if cli_models:
+                return cli_models
+
+        # Priority 2: Direct HTTP endpoint (used as fallback or in unit tests with HTTP mocks)
         token = await self._access_token(account)
         for base in BASE_URLS:
             try:
@@ -684,9 +691,122 @@ class AgyModelAdapter:
         return []
 
     @classmethod
-    def _parse_model_catalog(
-        cls, raw_models: dict[str, Any]
-    ) -> list[AgyAccountModelInfo]:
+    async def _fetch_models_via_cli(cls, account: AgyAccount) -> list[AgyAccountModelInfo]:
+        """Query models live from the headless `agy` CLI application for this account."""
+        agy_bin = shutil.which("agy") or (
+            "/root/.local/bin/agy" if os.path.exists("/root/.local/bin/agy") else None
+        )
+        if not agy_bin or not account.refresh_token:
+            return []
+
+        # Prepare isolated app data directory with this account's oauth credentials
+        account_data_dir = tempfile.mkdtemp(prefix=f"agy_acc_{account.credential_id[:8]}_")
+        token_path = os.path.join(account_data_dir, "antigravity-oauth-token")
+        try:
+            with open(token_path, "w", encoding="utf-8") as f:
+                json.dump(
+                    {
+                        "token": {
+                            "access_token": "",
+                            "token_type": "Bearer",
+                            "refresh_token": account.refresh_token,
+                            "expiry": "2000-01-01T00:00:00Z",
+                        },
+                        "auth_method": "oauth",
+                    },
+                    f,
+                )
+
+            env = os.environ.copy()
+            env["ANTIGRAVITY_APP_DATA_DIR"] = account_data_dir
+            env["HOME"] = "/root"
+
+            proc = await asyncio.create_subprocess_exec(
+                agy_bin,
+                "models",
+                stdout=asyncio.subprocess.PIPE,
+                stderr=asyncio.subprocess.PIPE,
+                env=env,
+            )
+            try:
+                stdout_bytes, _ = await asyncio.wait_for(proc.communicate(), timeout=20.0)
+            except TimeoutError:
+                proc.kill()
+                return []
+
+            if proc.returncode != 0:
+                return []
+
+            return cls._parse_cli_models_output(stdout_bytes.decode("utf-8", errors="replace"))
+        except Exception:
+            logger.warning(
+                "Failed to fetch models via agy CLI, falling back to HTTP", exc_info=True
+            )
+            return []
+        finally:
+            shutil.rmtree(account_data_dir, ignore_errors=True)
+
+    @classmethod
+    def _parse_cli_models_output(cls, output: str) -> list[AgyAccountModelInfo]:
+        items: list[AgyAccountModelInfo] = []
+        for line in output.splitlines():
+            line = line.strip()
+            if not line or "Fetching" in line:
+                continue
+            # Output format is tab-separated: <model_id>\t<display_name>
+            parts = line.split("\t", 1)
+            if len(parts) == 2:
+                model_id = parts[0].strip()
+                display_name = parts[1].strip()
+            else:
+                tokens = line.split(None, 1)
+                model_id = tokens[0].strip()
+                display_name = tokens[1].strip() if len(tokens) > 1 else model_id
+
+            lowered = model_id.lower()
+            dn_lowered = display_name.lower()
+
+            supports_thinking = (
+                "thinking" in lowered
+                or "thinking" in dn_lowered
+                or "-high" in lowered
+                or "-medium" in lowered
+                or "-low" in lowered
+            )
+
+            supported_efforts: list[str] = []
+            default_effort: str | None = None
+            if "-high" in lowered or "(high)" in dn_lowered:
+                supported_efforts = ["high"]
+                default_effort = "high"
+            elif "-medium" in lowered or "(medium)" in dn_lowered:
+                supported_efforts = ["medium"]
+                default_effort = "medium"
+            elif "-low" in lowered or "(low)" in dn_lowered:
+                supported_efforts = ["low"]
+                default_effort = "low"
+            elif "thinking" in lowered or "thinking" in dn_lowered:
+                if "gemini" in lowered:
+                    supported_efforts = ["low", "medium", "high"]
+                    default_effort = "high"
+                else:
+                    supported_efforts = []
+                    default_effort = None
+
+            items.append(
+                AgyAccountModelInfo(
+                    id=model_id,
+                    display_name=display_name,
+                    supports_thinking=supports_thinking,
+                    supported_efforts=supported_efforts,
+                    default_effort=default_effort,
+                    supports_images=True,
+                )
+            )
+        return items
+
+    @classmethod
+    def _parse_model_catalog(cls, raw_models: dict[str, Any]) -> list[AgyAccountModelInfo]:
         items: list[AgyAccountModelInfo] = []
         for model_id, details in sorted(raw_models.items()):
             if not isinstance(details, dict):
@@ -739,9 +859,7 @@ class AgyModelAdapter:
         return " ".join(words)
 
     @classmethod
-    def _supported_efforts_for(
-        cls, model_id: str, details: dict[str, Any]
-    ) -> list[str]:
+    def _supported_efforts_for(cls, model_id: str, details: dict[str, Any]) -> list[str]:
         lowered = model_id.lower()
         if "image" in lowered:
             return []
@@ -869,7 +987,12 @@ class AgyModelAdapter:
     ) -> ModelResponse:
         token = await self._access_token(account)
         payload = _build_generate_payload(
-            account, model.internal_name, messages, system, temperature, max_tokens,
+            account,
+            model.internal_name,
+            messages,
+            system,
+            temperature,
+            max_tokens,
             reasoning_effort=model.reasoning_effort,
         )
         headers = {
@@ -894,9 +1017,9 @@ class AgyModelAdapter:
                 # Both backends share one quota, but a 429 can also be a
                 # backend hiccup — try the other base URL before benching the
                 # whole account.
-                retry_after = _retry_after_seconds(
-                    response.headers.get("Retry-After")
-                ) or retry_after
+                retry_after = (
+                    _retry_after_seconds(response.headers.get("Retry-After")) or retry_after
+                )
                 continue
             if response.status_code in (401, 403):
                 raise _AgyAuthError(
@@ -909,12 +1032,9 @@ class AgyModelAdapter:
                 # Our request shape is wrong — another account will fail the
                 # same way, so surface it instead of burning the pool.
                 raise ProviderUnavailableError(
-                    f"agy درخواست را رد کرد ({response.status_code}): "
-                    f"{response.text[:300]}"
-                )
-            return _parse_generate_response(
-                response.json(), model=model, started=started
+                    f"agy درخواست را رد کرد ({response.status_code}): {response.text[:300]}"
                 )
+            return _parse_generate_response(response.json(), model=model, started=started)
         if retry_after is not None:
             raise _AgyRateLimited(retry_after)
         raise last_transient or _AgyTransientError("هیچ بک‌اند agy پاسخ نداد.")
@@ -944,14 +1064,26 @@ def _build_generate_payload(
         role = "model" if message.role == "assistant" else "user"
         contents.append({"role": role, "parts": [{"text": message.content}]})
 
+    backend_model = model_name
+    effective_effort = reasoning_effort
+    if model_name == "gemini-3.8-flash-high":
+        backend_model = "gemini-3.8-flash-tiered"
+        effective_effort = effective_effort or "high"
+    elif model_name == "gemini-3.8-flash-medium":
+        backend_model = "gemini-3.8-flash-tiered"
+        effective_effort = effective_effort or "medium"
+    elif model_name == "gemini-3.8-flash-low":
+        backend_model = "gemini-3.8-flash-tiered"
+        effective_effort = effective_effort or "low"
+
     generation: dict = {
         "temperature": temperature,
         "maxOutputTokens": max_tokens,
     }
     # Reasoning effort is only sent to models that can think; the backend
     # rejects ``thinkingConfig`` everywhere else with a 400.
-    if reasoning_effort and _supports_thinking(model_name):
-        budget = EFFORT_THINKING_BUDGETS.get(reasoning_effort, 8192)
+    if effective_effort and _supports_thinking(backend_model):
+        budget = EFFORT_THINKING_BUDGETS.get(effective_effort, 8192)
         generation["thinkingConfig"] = {
             "thinkingBudget": budget,
             "includeThoughts": True,
@@ -975,16 +1107,14 @@ def _build_generate_payload(
         }
     return {
         "project": account.project_id or _fallback_project_id(),
-        "model": model_name,
+        "model": backend_model,
         "userAgent": "antigravity",
         "requestId": f"agent-{uuid.uuid4()}",
         "request": request,
     }
 
 
-def _parse_generate_response(
-    data: dict, *, model: AiModel, started: float
-) -> ModelResponse:
+def _parse_generate_response(data: dict, *, model: AiModel, started: float) -> ModelResponse:
     candidates = data.get("candidates") or [{}]
     candidate = candidates[0] if isinstance(candidates[0], dict) else {}
     text_parts: list[str] = []
@@ -1025,14 +1155,10 @@ async def load_agy_accounts() -> list[ModelCredential]:
 
     async with SessionFactory() as session:
         providers = await ProviderRepository(session).list()
-        provider = next(
-            (p for p in providers if p.type.value == "agy" and p.is_active), None
-        )
+        provider = next((p for p in providers if p.type.value == "agy" and p.is_active), None)
         if provider is None:
             return []
-        return await ModelCredentialRepository(session).list_active_for_provider(
-            provider.id
-        )
+        return await ModelCredentialRepository(session).list_active_for_provider(provider.id)
 
 
 async def report_agy_state(

apps/api/tests/test_agy.py

@@ -27,9 +27,7 @@ from tests.conftest import auth_header, signup
 # ---------------------------------------------------------------------------
 
 GOOD_GENERATE: dict = {
-    "candidates": [
-        {"content": {"parts": [{"text": "سلام"}]}, "finishReason": "STOP"}
-    ],
+    "candidates": [{"content": {"parts": [{"text": "سلام"}]}, "finishReason": "STOP"}],
     "usageMetadata": {
         "promptTokenCount": 7,
         "candidatesTokenCount": 3,
@@ -192,9 +190,7 @@ async def test_agy_login_flow_end_to_end(client, google, platform_owner_token):
         headers=auth_header(platform_owner_token),
         json={
             "login_id": login_id,
-            "callback_url": (
-                f"http://localhost:51121/oauth-callback?code=auth-code&state={state}"
-            ),
+            "callback_url": (f"http://localhost:51121/oauth-callback?code=auth-code&state={state}"),
         },
     )
     assert completed.status_code == 201, completed.text
@@ -209,9 +205,7 @@ async def test_agy_login_flow_end_to_end(client, google, platform_owner_token):
     assert "agy-proj-123" in completed.text  # project id is admin metadata
 
     # 3. the account is listed.
-    listed = await client.get(
-        "/api/admin/agy/accounts", headers=auth_header(platform_owner_token)
-    )
+    listed = await client.get("/api/admin/agy/accounts", headers=auth_header(platform_owner_token))
     assert listed.status_code == 200
     emails = [a["email"] for a in listed.json()]
     assert emails == ["person@gmail.com"]
@@ -232,9 +226,7 @@ async def test_agy_complete_login_rejects_unrelated_url(client, google, platform
     )
     assert completed.status_code == 422
     # The failed attempt must not have created an account.
-    listed = await client.get(
-        "/api/admin/agy/accounts", headers=auth_header(platform_owner_token)
-    )
+    listed = await client.get("/api/admin/agy/accounts", headers=auth_header(platform_owner_token))
     assert listed.json() == []
 
 
@@ -249,9 +241,7 @@ async def test_agy_account_test_and_delete(client, google, platform_owner_token)
         headers=auth_header(platform_owner_token),
         json={
             "login_id": started.json()["login_id"],
-            "callback_url": (
-                f"http://localhost:51121/oauth-callback?code=c&state={state}"
-            ),
+            "callback_url": (f"http://localhost:51121/oauth-callback?code=c&state={state}"),
         },
     )
     account_id = created.json()["id"]
@@ -270,9 +260,7 @@ async def test_agy_account_test_and_delete(client, google, platform_owner_token)
         headers=auth_header(platform_owner_token),
     )
     assert deleted.status_code == 204
-    listed = await client.get(
-        "/api/admin/agy/accounts", headers=auth_header(platform_owner_token)
-    )
+    listed = await client.get("/api/admin/agy/accounts", headers=auth_header(platform_owner_token))
     assert listed.json() == []
 
 
@@ -289,15 +277,11 @@ async def test_agy_same_account_twice_replaces(client, google, platform_owner_to
             headers=auth_header(platform_owner_token),
             json={
                 "login_id": started.json()["login_id"],
-                "callback_url": (
-                    f"http://localhost:51121/oauth-callback?code=c&state={state}"
-                ),
+                "callback_url": (f"http://localhost:51121/oauth-callback?code=c&state={state}"),
             },
         )
         assert done.status_code == 201
-    listed = await client.get(
-        "/api/admin/agy/accounts", headers=auth_header(platform_owner_token)
-    )
+    listed = await client.get("/api/admin/agy/accounts", headers=auth_header(platform_owner_token))
     assert len(listed.json()) == 1
 
 
@@ -390,9 +374,7 @@ async def test_seed_leaves_no_legacy_providers(client, seeded_install, session_f
 def test_auth_url_shape():
     challenge = agy.AgyAuthClient().begin()
     query = parse_qs(urlparse(challenge.auth_url).query)
-    assert challenge.auth_url.startswith(
-        "https://accounts.google.com/o/oauth2/v2/auth"
-    )
+    assert challenge.auth_url.startswith("https://accounts.google.com/o/oauth2/v2/auth")
     assert query["client_id"][0] == (
         "1071006060591-tmhssin2h21lcre235vtolojh4g403ep.apps.googleusercontent.com"
     )
@@ -410,9 +392,7 @@ async def test_paste_flow_exchanges_and_resolves_account(google):
     """start → paste callback URL → account material, verifier stays server-side."""
     login_id, _auth_url = await agy.begin_login()
     challenge = agy._CHALLENGES[login_id]
-    callback = (
-        f"http://localhost:51121/oauth-callback?code=auth-code&state={challenge.state}"
-    )
+    callback = f"http://localhost:51121/oauth-callback?code=auth-code&state={challenge.state}"
     result = await agy.redeem_login(login_id, callback)
     assert result.email == "person@gmail.com"
     assert result.project_id == "agy-proj-123"
@@ -446,11 +426,10 @@ async def test_google_error_page_is_rejected(google):
 
 async def test_refresh_token_missing_is_actionable(monkeypatch):
     """A token response without ``refresh_token`` fails loudly and tells why."""
+
     async def handler(method, url, json_body, headers, data):
         if url.endswith("/token"):
-            return _FakeResponse(
-                json_data={"access_token": "a", "expires_in": 3600}
-            )
+            return _FakeResponse(json_data={"access_token": "a", "expires_in": 3600})
         return await default_handler(method, url, json_body, headers, data)
 
     patch_google(monkeypatch, handler)
@@ -617,7 +596,8 @@ def test_response_parsing_skips_thoughts():
 async def test_complete_happy_path(google, decrypted):
     adapter = agy.AgyModelAdapter()
     response = await adapter.complete(
-        model=_MODEL, credentials=[_agy_credential(email="one@gmail.com")],
+        model=_MODEL,
+        credentials=[_agy_credential(email="one@gmail.com")],
         messages=_MESSAGES,
     )
     assert response.content == "سلام"
@@ -631,6 +611,7 @@ async def test_complete_happy_path(google, decrypted):
 
 async def test_rate_limited_account_is_rotated_and_benched(monkeypatch, decrypted):
     """A 429 on one account must not fail the request: the pool moves on."""
+
     async def handler(method, url, json_body, headers, data):
         if url.endswith("/token"):
             # Each account's refresh token maps to its own access token.
@@ -642,9 +623,7 @@ async def test_rate_limited_account_is_rotated_and_benched(monkeypatch, decrypte
             auth = headers.get("Authorization", "")
             if auth.endswith("access-one"):
                 # Both base URLs hit the same quota.
-                return _FakeResponse(
-                    status_code=429, text="quota", headers={"Retry-After": "120"}
-                )
+                return _FakeResponse(status_code=429, text="quota", headers={"Retry-After": "120"})
             return _FakeResponse(json_data=GOOD_GENERATE)
         return await default_handler(method, url, json_body, headers, data)
 
@@ -659,9 +638,7 @@ async def test_rate_limited_account_is_rotated_and_benched(monkeypatch, decrypte
     adapter._state.cursor = -1
     one = _agy_credential(email="one@gmail.com", refresh_token="refresh-1")
     two = _agy_credential(email="two@gmail.com", refresh_token="refresh-2")
-    response = await adapter.complete(
-        model=_MODEL, credentials=[one, two], messages=_MESSAGES
-    )
+    response = await adapter.complete(model=_MODEL, credentials=[one, two], messages=_MESSAGES)
     assert response.content == "سلام"
 
     cooldowns = [r for r in reported if r[1] == "cooldown"]
@@ -696,6 +673,7 @@ async def test_every_account_rate_limited_raises(monkeypatch, decrypted):
 
 async def test_rejected_grant_marks_account_error(monkeypatch, decrypted):
     """401 means the OAuth grant died — bench as error, move to next account."""
+
     async def handler(method, url, json_body, headers, data):
         if url.endswith("/token"):
             access = {"refresh-1": "access-one", "refresh-2": "access-two"}[
@@ -719,9 +697,7 @@ async def test_rejected_grant_marks_account_error(monkeypatch, decrypted):
     adapter._state.cursor = -1
     dead = _agy_credential(email="dead@gmail.com", refresh_token="refresh-1")
     alive = _agy_credential(email="alive@gmail.com", refresh_token="refresh-2")
-    response = await adapter.complete(
-        model=_MODEL, credentials=[dead, alive], messages=_MESSAGES
-    )
+    response = await adapter.complete(model=_MODEL, credentials=[dead, alive], messages=_MESSAGES)
     assert response.content == "سلام"
     errors = [r for r in reported if r[1] == "error"]
     assert errors and errors[0][0] == dead.id
@@ -731,6 +707,7 @@ async def test_rejected_grant_marks_account_error(monkeypatch, decrypted):
 
 async def test_server_error_falls_through_to_next_backend(monkeypatch, decrypted):
     """5xx on one backend URL is not the account's fault — try the other."""
+
     async def handler(method, url, json_body, headers, data):
         if url.endswith("/token"):
             return _token_response()
@@ -743,7 +720,8 @@ async def test_server_error_falls_through_to_next_backend(monkeypatch, decrypted
     patch_google(monkeypatch, handler)
     adapter = agy.AgyModelAdapter()
     response = await adapter.complete(
-        model=_MODEL, credentials=[_agy_credential(email="one@gmail.com")],
+        model=_MODEL,
+        credentials=[_agy_credential(email="one@gmail.com")],
         messages=_MESSAGES,
     )
     assert response.content == "سلام"
@@ -782,9 +760,7 @@ async def test_available_models(google, decrypted):
 async def test_available_models_for_one_account(google, decrypted):
     """The console asks for a specific account, so the fetch is per-account."""
     adapter = agy.AgyModelAdapter()
-    models = await adapter.available_models_for(
-        _agy_credential(email="one@gmail.com")
-    )
+    models = await adapter.available_models_for(_agy_credential(email="one@gmail.com"))
     assert [m.id for m in models] == ["claude-sonnet-4-5", "gemini-3-flash"]
     assert models[1].supported_efforts == ["low", "medium", "high"]
     # The list is fetched from Google every call — no cached catalog.
@@ -818,9 +794,7 @@ async def test_account_models_endpoint(client, google, platform_owner_token):
             headers=auth_header(platform_owner_token),
             json={
                 "login_id": started.json()["login_id"],
-                "callback_url": (
-                    f"http://localhost:51121/oauth-callback?code=c&state={state}"
-                ),
+                "callback_url": (f"http://localhost:51121/oauth-callback?code=c&state={state}"),
             },
         )
     ).json()["id"]
@@ -843,6 +817,7 @@ async def test_account_models_endpoint(client, google, platform_owner_token):
 
 async def test_completion_carries_the_model_effort(monkeypatch, decrypted):
     """The end-to-end path: effort on the model row reaches the payload."""
+
     async def handler(method, url, json_body, headers, data):
         if url.endswith("/token"):
             return _token_response()
@@ -868,3 +843,104 @@ async def test_completion_carries_the_model_effort(monkeypatch, decrypted):
         "thinkingBudget": 4096,
         "includeThoughts": True,
     }
+
+
+def test_parse_cli_models_output():
+    """Verify parsing raw output from agy models CLI."""
+    sample_output = (
+        "Fetching available models...\n"
+        "gemini-3.8-flash-high\tGemini 3.8 Flash (High)\n"
+        "gemini-3.8-flash-medium\tGemini 3.8 Flash (Medium)\n"
+        "gemini-3.8-flash-low\tGemini 3.8 Flash (Low)\n"
+        "claude-sonnet-4-6\tClaude Sonnet 4.6 (Thinking)\n"
+        "claude-opus-4-6-thinking\tClaude Opus 4.6 (Thinking)\n"
+        "gpt-oss-120b-medium\tGPT-OSS 120B (Medium)\n"
+    )
+    models = agy.AgyModelAdapter._parse_cli_models_output(sample_output)
+    assert len(models) == 6
+
+    m_high = next(m for m in models if m.id == "gemini-3.8-flash-high")
+    assert m_high.display_name == "Gemini 3.8 Flash (High)"
+    assert m_high.supports_thinking is True
+    assert m_high.supported_efforts == ["high"]
+    assert m_high.default_effort == "high"
+
+    m_med = next(m for m in models if m.id == "gemini-3.8-flash-medium")
+    assert m_med.supported_efforts == ["medium"]
+    assert m_med.default_effort == "medium"
+
+    m_low = next(m for m in models if m.id == "gemini-3.8-flash-low")
+    assert m_low.supported_efforts == ["low"]
+    assert m_low.default_effort == "low"
+
+    m_claude = next(m for m in models if m.id == "claude-sonnet-4-6")
+    assert m_claude.supports_thinking is True
+
+    m_gpt = next(m for m in models if m.id == "gpt-oss-120b-medium")
+    assert m_gpt.supported_efforts == ["medium"]
+
+
+def test_build_generate_payload_tiered_mapping():
+    """Verify virtual CLI models map to underlying tiered model with appropriate budget."""
+    account = agy.AgyAccount(
+        credential_id="cred-1",
+        refresh_token="rt-1",
+        email="test@example.com",
+        project_id="proj-123",
+    )
+    messages = [ChatMessage(role="user", content="Hello")]
+
+    payload_low = agy._build_generate_payload(
+        account,
+        "gemini-3.8-flash-low",
+        messages,
+        system="",
+        temperature=0.7,
+        max_tokens=4096,
+    )
+    assert payload_low["model"] == "gemini-3.8-flash-tiered"
+    assert payload_low["request"]["generationConfig"]["thinkingConfig"] == {
+        "thinkingBudget": 4096,
+        "includeThoughts": True,
+    }
+
+    payload_high = agy._build_generate_payload(
+        account,
+        "gemini-3.8-flash-high",
+        messages,
+        system="",
+        temperature=0.7,
+        max_tokens=4096,
+    )
+    assert payload_high["model"] == "gemini-3.8-flash-tiered"
+    assert payload_high["request"]["generationConfig"]["thinkingConfig"] == {
+        "thinkingBudget": 16384,
+        "includeThoughts": True,
+    }
+
+
+async def test_fetch_models_via_cli_mocked(monkeypatch):
+    """Verify _fetch_models_via_cli invokes the CLI binary and handles output."""
+    account = agy.AgyAccount(
+        credential_id="cred-1",
+        refresh_token="rt-test-123",
+        email="test@example.com",
+        project_id="proj-123",
+    )
+
+    class FakeProc:
+        returncode = 0
+
+        async def communicate(self):
+            return b"gemini-3.8-flash-high\tGemini 3.8 Flash (High)\n", b""
+
+    async def fake_create_subprocess_exec(*args, **kwargs):
+        return FakeProc()
+
+    monkeypatch.setattr(agy.asyncio, "create_subprocess_exec", fake_create_subprocess_exec)
+    monkeypatch.setattr(agy.shutil, "which", lambda cmd: "/usr/bin/agy")
+
+    models = await agy.AgyModelAdapter._fetch_models_via_cli(account)
+    assert len(models) == 1
+    assert models[0].id == "gemini-3.8-flash-high"
+    assert models[0].display_name == "Gemini 3.8 Flash (High)"

apps/web/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@agy/web",
-  "version": "0.1.3",
+  "version": "0.1.4",
   "private": true,
   "description": "Smart Organization frontend — Next.js App Router, RTL, dark-first.",
   "scripts": {

apps/web/src/lib/version.ts

@@ -12,7 +12,7 @@
  * "the app is on the wrong version" bug into a failing test, which is the
  * whole point of keeping them separate.
  */
-export const APP_VERSION = "0.1.3";
+export const APP_VERSION = "0.1.4";
 
 /** What the footer pill renders — ``v`` prefix keeps it unambiguous in the UI. */
 export const APP_VERSION_LABEL = `v${APP_VERSION}`;

package.json

@@ -1,6 +1,6 @@
 {
   "name": "smart-org",
-  "version": "0.1.3",
+  "version": "0.1.4",
   "private": true,
   "description": "Smart Organization — AI Workforce Platform. Standalone multi-tenant SaaS, independent of the Telegram bot runtime.",
   "packageManager": "pnpm@9.12.0",