agy تنها ارائه‌دهندهٔ مدل؛ حذف بخش ارائه‌دهنده‌ها و افزودن زنجیرهٔ انتخاب مدل

- حذف کامل providerهای mock و openai_compatible: دامنه، آداپوترها، رجیستری، کانفیگ، seed و رابطٔ مدیریت
- بخش «ارائه‌دهنده‌ها» از کنسول حذف شد؛ ورود اکانت‌ها فقط از صفحهٔ اکانت‌های agy
- صفحهٔ مدل‌ها بازطراحی شد: ارائه‌دهنده → اکانت → مدل (کاتالوگ لحظه‌ای و زندهٔ همان اکانت از Google) → افورت استدلال؛ بدون هیچ ورود دستی نام مدل
- مدل‌ها ستون reasoning_effort گرفتند؛ آداپتور آن را به thinkingConfig (بودجهٔ ۱۰۲۴ تا ۳۲۷۶۸) ترجمه می‌کند و فقط به مدل‌های استدلالی می‌فرستد
- مهاجرت c2d3e4f5b6a7: افزودن reasoning_effort و حذف ردیف‌های providerهای قدیمی
- تست‌ها: ۲۴۶ تست بک‌اند و ۵۱ تست فرانت سبز؛ ruff و mypy و tsc و eslint تمیز

.env.example

@@ -27,15 +27,14 @@ JWT_SECRET=change-me
 ENCRYPTION_KEY=change-me-32-bytes
 
 # --- AI models (hidden from customers) -------------------------------------
-# mock | openai_compatible | agy
+# agy (Google Antigravity) is the only model provider. The setting is kept for
-AI_DEFAULT_PROVIDER=mock
+# compatibility with older .env files; any value is ignored (with a warning).
-OPENAI_COMPATIBLE_BASE_URL=
+AI_DEFAULT_PROVIDER=agy
-OPENAI_COMPATIBLE_API_KEY=
 
 # --- agy (Google Antigravity) ----------------------------------------------
-# Only used when AI_DEFAULT_PROVIDER=agy. There are no API keys: accounts are
+# The only model provider. There are no API keys: accounts are Google
-# Google identities logged in via OAuth from the admin console
+# identities logged in via OAuth from the admin console
-# (/admin/accounts → افزودن اکانت agy). Any number of accounts can be added —
+# (/admin/agy-accounts → افزودن اکانت agy). Any number of accounts can be added —
 # the rotation pool spreads load across them and benches a rate-limited one
 # until its quota window resets. AGY_CLIENT_ID/SECRET default to the public
 # Antigravity CLI application; override them to use your own Google OAuth

README.md

@@ -21,8 +21,9 @@ provider. این‌ها فقط در **Admin Console** و فقط برای مال
 
 پلتفرم سه لایهٔ نقش دارد:
 
-1. **مالک سامانه (platform owner)** — providerها، مدل‌ها، routing policy، تعرفه‌ها،
+1. **مالک سامانه (platform owner)** — اکانت‌های agy، مدل‌ها، routing policy،
-   sandbox templateها، مشاهدهٔ کل مصرف.
+   تعرفه‌ها، sandbox templateها، مشاهدهٔ کل مصرف. agy تنها ارائه‌دهندهٔ مدل
+   سامانه است و بخش جداگانهٔ «ارائه‌دهنده‌ها» وجود ندارد.
 2. **مدیر سازمان مشتری** — ساخت سازمان، دعوت عضو، پروژه، فعال‌سازی کارمند، مصرف.
 3. **عضو سازمان** — گفتگو با مدیر/کارمند، ساخت task، آپلود فایل، دیدن خروجی.
 
@@ -217,7 +218,6 @@ web/
 
 | قابلیت | env | مقادیر |
 |---|---|---|
-| مدل | `AI_DEFAULT_PROVIDER` | `mock` \| `openai_compatible` \| `agy` |
 | Git | `GIT_PROVIDER` | `mock` \| `gitea` |
 | دامنهٔ تست | `DOMAIN_ROUTER` | `mock` \| `caddy` |
 | Sandbox | `SANDBOX_PROVIDER` | `mock` \| `daytona` \| `e2b` |
@@ -226,24 +226,25 @@ web/
 اگر credential واقعی نداریم، **قابلیت حذف نمی‌شود**: مدل دیتابیس، interface، UI و
 adapter mock کامل ساخته می‌شود و adapter واقعی پشت همین flag آماده می‌ماند.
 
-### ارائه‌دهندهٔ agy (Google Antigravity)
+### ارائه‌دهندهٔ agy (Google Antigravity) — تنها ارائه‌دهندهٔ مدل
 
 agy کلید API ندارد؛ **اکانت**‌ها هویت Google هستند و از طریق OAuth وارد
-می‌شوند. برای فعال‌سازی:
+می‌شوند. کاتالوگ مدل‌ها در seed ساخته می‌شود و همان اول فعال است. برای شروع:
 
-1. `AI_DEFAULT_PROVIDER=agy` را در `.env` بگذارید و `pnpm db:seed` را دوباره
+1. با حساب مالک سامانه وارد کنسول مدیریت شوید و از
-   اجرا کنید (کتالوگ مدل‌های agy فعال و کاتالوگ mock غیرفعال می‌شود).
+   `/admin/agy-accounts` → «افزودن اکانت agy» شروع کنید: لینک احراز هویت گوگل
-2. با حساب مالک سامانه وارد کنسول مدیریت شوید و از
-   `/admin/accounts` → «افزودن اکانت agy» شروع کنید: لینک احراز هویت گوگل
    باز می‌شود، بعد از تأیید، URL بازگشت را از نوار آدرس مرورگر کپی و Paste
    کنید (صفحهٔ مقصد عمداً بارگذاری نمی‌شود؛ پارامترها همان‌جا هستند).
-3. هر تعداد اکانت Google که بخواهید اضافه کنید. مدل‌ها را در
+2. در `/admin/models` زنجیرهٔ افزودن مدل کاملاً انتخابی است: ارائه‌دهنده
-   `/admin/models` در صورت نیاز فعال کنید.
+   (agy) → اکانت → مدل (کاتالوگ لحظه‌ای و زندهٔ همان اکانت) → افورت استدلال
+   (فقط برای مدل‌های استدلالی). هیچ نام مدلی دستی تایپ نمی‌شود و مشتری نام مدل
+   را هرگز نمی‌بیند.
 
-پول چرخشی، بار را بین اکانت‌ها پخش می‌کند و اکانتی که ۴۲۹ گرفته تا پایان
+هر تعداد اکانت Google که بخواهید اضافه کنید. پول چرخشی، بار را بین اکانت‌ها
-پنجرهٔ سهمیهٔ آن کنار گذاشته می‌شود (هم در حافظه و هم در دیتابیس، تا workerها
+پخش می‌کند و اکانتی که ۴۲۹ گرفته تا پایان پنجرهٔ سهمیهٔ آن کنار گذاشته
-هم همان تصویر را ببینند). refresh tokenها فقط encrypted at rest ذخیره
+می‌شود (هم در حافظه و هم در دیتابیس، تا workerها همان تصویر را ببینند).
-می‌شوند و هیچ‌وقت از API برنمی‌گردند.
+refresh tokenها فقط encrypted at rest ذخیره می‌شوند و هیچ‌وقت از API
+برنمی‌گردند.
 
 ---
 

apps/api/alembic/versions/c2d3e4f5b6a7_agy_only_and_model_effort.py

@@ -0,0 +1,62 @@
+"""agy becomes the only model provider; models gain a reasoning effort
+
+Two changes locked in together:
+
+1. ``ai_models`` gains ``reasoning_effort`` (the agy thinking-budget selector
+   the admin console sets per model). NOT NULL with an empty-string default so
+   existing rows backfill to "provider default" without a rewrite.
+2. Every provider row that is not ``agy`` — and the models and credentials that
+   hang off it — is deleted. agy is the only provider the platform talks to;
+   a leftover ``mock``/``openai_compatible`` provider would keep dead rows
+   selectable by ``pick_for_mode`` and confuse the console.
+
+The seed reconciles this on every run, so a database that was already re-seeded
+finds nothing left to do here.
+
+Revision ID: c2d3e4f5b6a7
+Revises: b1c2d3e4f5a6
+Create Date: 2026-10-13 09:05:00.000000
+"""
+
+from __future__ import annotations
+
+from collections.abc import Sequence
+
+import sqlalchemy as sa
+from alembic import op
+
+
+revision: str = "c2d3e4f5b6a7"
+down_revision: str | None = "b1c2d3e4f5a6"
+branch_labels: str | Sequence[str] | None = None
+depends_on: str | Sequence[str] | None = None
+
+
+def upgrade() -> None:
+    op.add_column(
+        "ai_models",
+        sa.Column(
+            "reasoning_effort",
+            sa.String(length=16),
+            nullable=False,
+            server_default="",
+        ),
+    )
+    # Children first: the FKs carry ON DELETE CASCADE, but spelling the deletes
+    # out keeps the intent readable and works on every backend/driver combo
+    # Alembic supports (including SQLite without PRAGMA introspection).
+    op.execute(
+        "DELETE FROM ai_models WHERE provider_id IN "
+        "(SELECT id FROM ai_providers WHERE type <> 'agy')"
+    )
+    op.execute(
+        "DELETE FROM ai_model_credentials WHERE provider_id IN "
+        "(SELECT id FROM ai_providers WHERE type <> 'agy')"
+    )
+    op.execute("DELETE FROM ai_providers WHERE type <> 'agy'")
+
+
+def downgrade() -> None:
+    # The deleted rows are unrecoverable by design: their content was mock
+    # fixtures and API keys the operator would have had to paste again.
+    op.drop_column("ai_models", "reasoning_effort")

apps/api/app/application/ports/model_provider.py

@@ -37,8 +37,7 @@ class ModelHealth:
 
 @runtime_checkable
 class ModelProviderPort(Protocol):
-    """Driven port. Implemented by ``MockModelProviderAdapter`` and
+    """Driven port. Implemented by ``AgyModelAdapter`` (the only provider)."""
-    ``OpenAICompatibleModelAdapter``."""
 
     async def complete(
         self,

apps/api/app/application/services/workforce_service.py

@@ -714,8 +714,8 @@ class WorkforceService:
     ) -> str:
         """Run one completion and record its usage. The single model entry point."""
         model = await self._resolve_model(employee)
-        # Pool-aware: rotation providers (agy) fan out across every usable
+        # Pool-aware: the agy adapter fans out across every usable account, so
-        # account; key-based providers simply get their one active credential.
+        # one account's rate limit never becomes everyone's outage.
         pool = await self.credentials.list_active_for_provider(model.provider_id)
         credential = pool[0] if pool else None
 

apps/api/app/core/config.py

@@ -60,9 +60,10 @@ class Settings(BaseSettings):
     encryption_key: str = ""
 
     # --- AI models (hidden from customers) -----------------------------------
-    ai_default_provider: str = "mock"
+    #: agy (Google Antigravity) is the only model provider the platform talks
-    openai_compatible_base_url: str = ""
+    #: to. The setting is kept so an old ``AI_DEFAULT_PROVIDER`` value in a
-    openai_compatible_api_key: str = ""
+    #: ``.env`` keeps booting: the registry warns and uses agy regardless.
+    ai_default_provider: str = "agy"
 
     # --- agy (Google Antigravity) --------------------------------------------
     #: OAuth client of the public Antigravity CLI application. Overridable so a

apps/api/app/core/redact.py

@@ -205,7 +205,6 @@ def _configured_secret_values() -> tuple[str, ...]:
     candidates = (
         settings.jwt_secret,
         settings.encryption_key,
-        settings.openai_compatible_api_key,
         settings.gitea_admin_token,
         settings.daytona_api_key,
         settings.e2b_api_key,

apps/api/app/domain/ai_control/models.py

@@ -14,13 +14,10 @@ from app.domain.base import new_id, utcnow
 
 
 class ProviderType(StrEnum):
-    MOCK = "mock"
-    OPENAI_COMPATIBLE = "openai_compatible"
-    ANTHROPIC = "anthropic"
-    GEMINI = "gemini"
-    OPENROUTER = "openrouter"
     #: Google Antigravity (``agy``) — OAuth accounts billed through a Google AI
-    #: subscription. Credentials of this type are *accounts*, not API keys.
+    #: subscription. It is the *only* provider the platform talks to; the
+    #: previous ``mock`` / ``openai_compatible`` kinds were removed with the
+    #: single-provider decision, and the migration drops their rows.
     AGY = "agy"
 
 
@@ -34,6 +31,11 @@ class CredentialStatus(StrEnum):
     ERROR = "error"
 
 
+#: agy reasoning efforts, cheapest first. The admin console offers exactly
+#: these; the adapter maps each to a thinking budget (see adapters/agy.py).
+REASONING_EFFORTS: tuple[str, ...] = ("minimal", "low", "medium", "high", "xhigh")
+
+
 class RoutingMode(StrEnum):
     FAST = "fast"
     CHEAP = "cheap"
@@ -65,7 +67,7 @@ class RateUnit(StrEnum):
 @dataclass(slots=True, kw_only=True)
 class Provider:
     name: str
-    type: ProviderType = ProviderType.MOCK
+    type: ProviderType = ProviderType.AGY
     base_url: str = ""
     id: str = field(default_factory=new_id)
     is_active: bool = True
@@ -74,12 +76,10 @@ class Provider:
 
 @dataclass(slots=True, kw_only=True)
 class ModelCredential:
-    """One credential for a provider.
+    """One credential for the agy provider — i.e. one Google *account*.
 
-    For key-based providers ``encrypted_api_key`` holds the encrypted API key and
+    ``encrypted_api_key`` holds the encrypted JSON blob
-    the agy fields stay empty. For the agy provider the credential *is an
+    ``{"refresh_token": ..., "project_id": ..., "email": ...}``, while ``email`` /
-    account*: ``encrypted_api_key`` holds the encrypted JSON blob
-    ``{"refresh_token": ..., "project_id": ..., "email": ...}`` while ``email`` /
     ``project_id`` duplicate the identity columns so the admin console can show
     and rotate accounts without touching the ciphertext.
     """
@@ -107,6 +107,10 @@ class AiModel:
     capabilities: list[str] = field(default_factory=list)
     input_price_per_1m: float = 0.0
     output_price_per_1m: float = 0.0
+    #: agy reasoning effort: ``minimal | low | medium | high | xhigh`` (empty =
+    #: provider default). Only thinking-capable models receive it; the adapter
+    #: maps it to a ``thinkingConfig`` budget.
+    reasoning_effort: str = ""
     id: str = field(default_factory=new_id)
     is_active: bool = True
     created_at: datetime = field(default_factory=utcnow)

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

@@ -91,6 +91,33 @@ SAFETY_SETTINGS: list[dict[str, str]] = [
 #: manual paste must fit inside it comfortably.
 CHALLENGE_TTL_SECONDS = 600
 
+#: agy ``reasoning_effort`` → thinking budget (tokens), mirroring the Antigravity
+#: client. An unknown value falls back to ``medium`` server-side, and the UI
+#: only ever offers these five.
+EFFORT_THINKING_BUDGETS: dict[str, int] = {
+    "minimal": 1024,
+    "low": 4096,
+    "medium": 8192,
+    "high": 16384,
+    "xhigh": 32768,
+}
+
+#: Where thinking is honoured. A request must never carry ``thinkingConfig`` to
+#: a model that does not support it — that is a hard 400 from the backend.
+_THINKING_MARKERS: tuple[str, ...] = (
+    "thinking",
+    "gemini-3",
+    "gemini-2.5-pro",
+    "gemini-2.5-flash",
+)
+
+
+def _supports_thinking(model_name: str) -> bool:
+    lowered = model_name.lower()
+    if "image" in lowered:
+        return False
+    return any(marker in lowered for marker in _THINKING_MARKERS)
+
 _StatusFn = Callable[..., Awaitable[None]]
 _LoaderFn = Callable[[], Awaitable[list[ModelCredential]]]
 
@@ -603,7 +630,25 @@ class AgyModelAdapter:
         accounts = [a for a in (_account_of(c) for c in await self._load()) if a.refresh_token]
         if not accounts:
             return []
-        token = await self._access_token(accounts[0])
+        return await self._fetch_models(accounts[0])
+
+    async def available_models_for(self, credential: ModelCredential) -> list[str]:
+        """Live model catalog of exactly one account.
+
+        Called per account selection in the admin console, so the list is
+        fetched fresh from Google every time (the access token is cached, the
+        ``fetchAvailableModels`` round-trip is not) and reflects what *this*
+        subscription is entitled to.
+        """
+        account = _account_of(credential)
+        if not account.refresh_token:
+            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[str]:
+        token = await self._access_token(account)
         for base in BASE_URLS:
             try:
                 async with httpx.AsyncClient(timeout=30.0) as client:
@@ -713,7 +758,8 @@ 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 = {
             "Authorization": f"Bearer {token}",
@@ -775,6 +821,7 @@ def _build_generate_payload(
     system: str,
     temperature: float,
     max_tokens: int,
+    reasoning_effort: str = "",
 ) -> dict:
     """Translate the port's OpenAI-shaped call into the Antigravity envelope."""
     system_parts: list[str] = [system] if system else []
@@ -786,13 +833,27 @@ def _build_generate_payload(
         role = "model" if message.role == "assistant" else "user"
         contents.append({"role": role, "parts": [{"text": message.content}]})
 
+    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)
+        generation["thinkingConfig"] = {
+            "thinkingBudget": budget,
+            "includeThoughts": True,
+        }
+        if max_tokens <= budget:
+            # The budget must fit inside the answer window or the thinking
+            # model has nowhere to emit its final answer.
+            generation["maxOutputTokens"] = budget + 8000
+
     request: dict = {
         "contents": contents,
         "sessionId": _random_session_id(),
-        "generationConfig": {
+        "generationConfig": generation,
-            "temperature": temperature,
-            "maxOutputTokens": max_tokens,
-        },
         "toolConfig": {"functionCallingConfig": {"mode": "VALIDATED"}},
         "safetySettings": SAFETY_SETTINGS,
     }

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

@@ -1,13 +1,13 @@
-"""Mock adapters — every port has one so the stack runs with zero credentials.
+"""Mock adapters for the non-AI ports, so the stack runs with zero credentials.
 
 These are not toys: they enforce the same contracts (timeouts, output caps,
 validation) as the real adapters, so switching a ``*_PROVIDER`` env var does not
-change application behaviour.
+change application behaviour. The model port has no mock — ``agy`` is the only
+model provider (see ``adapters/agy.py``).
 """
 
 from __future__ import annotations
 
-import asyncio
 import hashlib
 import secrets
 import time
@@ -17,14 +17,11 @@ from typing import Any
 
 from app.application.ports import (
     BranchInfo,
-    ChatMessage,
     CheckoutSession,
     CommandResult,
     CommitInfo,
     GitCredentials,
     MergeResult,
-    ModelHealth,
-    ModelResponse,
     Notification,
     PaymentResult,
     PreviewDomainInfo,
@@ -38,7 +35,6 @@ from app.application.ports import (
     VectorHit,
     WorkspaceInfo,
 )
-from app.domain.ai_control.models import AiModel
 from app.domain.platform.rules import (
     assert_port_allocatable,
     build_host,
@@ -46,57 +42,6 @@ from app.domain.platform.rules import (
     validate_proxy_target,
 )
 
-# ---------------------------------------------------------------------------
-# Model
-# ---------------------------------------------------------------------------
-
-
-class MockModelProviderAdapter:
-    """Deterministic, offline stand-in for an LLM.
-
-    Returns a canned Persian acknowledgement that names the task so tests can
-    assert on it without a network call.
-    """
-
-    def __init__(self, latency_ms: int = 0) -> None:
-        self._latency_ms = latency_ms
-
-    async def complete(
-        self,
-        *,
-        model: AiModel,
-        credential=None,
-        credentials=None,  # accepted for port parity; the mock has no pool
-        messages: list[ChatMessage],
-        system: str = "",
-        temperature: float = 0.2,
-        max_tokens: int = 2048,
-    ) -> ModelResponse:
-        started = time.perf_counter()
-        if self._latency_ms:
-            await asyncio.sleep(self._latency_ms / 1000)
-        last_user = next((m.content for m in reversed(messages) if m.role == "user"), "")
-        prompt_tokens = sum(len(m.content.split()) for m in messages)
-        # The canned text must read like model output, not like a debug tag: the
-        # provider/model identity is platform-internal and must never surface in
-        # anything a customer can read (see ``test_run_payload_never_leaks_model_identity``).
-        content = (
-            "این پاسخ آزمایشی است. "
-            f"درخواست دریافتی: {last_user[:280] or '—'}"
-        )
-        return ModelResponse(
-            content=content,
-            model_internal_name=model.internal_name,
-            input_tokens=prompt_tokens + len(system.split()),
-            output_tokens=len(content.split()),
-            latency_ms=int((time.perf_counter() - started) * 1000),
-            raw={"mock": True},
-        )
-
-    async def health(self) -> ModelHealth:
-        return ModelHealth(ok=True, detail="mock provider")
-
-
 # ---------------------------------------------------------------------------
 # Git
 # ---------------------------------------------------------------------------

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

@@ -1,98 +0,0 @@
-"""Real model adapter for any OpenAI-compatible endpoint.
-
-Works with OpenAI, OpenRouter, vLLM, LiteLLM, Ollama's OpenAI shim, etc. The
-API key is read from the credential record (decrypted by the caller), never from
-the environment at request time.
-"""
-
-from __future__ import annotations
-
-import time
-
-import httpx
-
-from app.application.ports import ChatMessage, ModelHealth, ModelResponse
-from app.core.errors import ProviderUnavailableError
-from app.domain.ai_control.models import AiModel, ModelCredential
-
-
-class OpenAICompatibleModelAdapter:
-    def __init__(self, *, base_url: str, default_api_key: str = "", timeout: float = 120.0) -> None:
-        self.base_url = (base_url or "").rstrip("/")
-        self.default_api_key = default_api_key
-        self.timeout = timeout
-
-    def _resolve_key(self, credential: ModelCredential | None, decrypt) -> str:
-        if credential is not None and credential.encrypted_api_key:
-            try:
-                return decrypt(credential.encrypted_api_key)
-            except Exception:  # noqa: BLE001 - fall through to the env default
-                pass
-        return self.default_api_key
-
-    async def complete(
-        self,
-        *,
-        model: AiModel,
-        credential: ModelCredential | None = None,
-        credentials: list[ModelCredential] | None = None,  # no pool here
-        messages: list[ChatMessage],
-        system: str = "",
-        temperature: float = 0.2,
-        max_tokens: int = 2048,
-    ) -> ModelResponse:
-        if not self.base_url:
-            raise ProviderUnavailableError(
-                "OPENAI_COMPATIBLE_BASE_URL تنظیم نشده است."
-            )
-
-        payload_messages = ([{"role": "system", "content": system}] if system else []) + [
-            {"role": m.role, "content": m.content} for m in messages
-        ]
-        payload = {
-            "model": model.internal_name,
-            "messages": payload_messages,
-            "temperature": temperature,
-            "max_tokens": max_tokens,
-        }
-        headers = {"Content-Type": "application/json"}
-        api_key = self.default_api_key
-        if api_key:
-            headers["Authorization"] = f"Bearer {api_key}"
-
-        started = time.perf_counter()
-        try:
-            async with httpx.AsyncClient(timeout=self.timeout) as client:
-                response = await client.post(
-                    f"{self.base_url}/chat/completions", json=payload, headers=headers
-                )
-                response.raise_for_status()
-                data = response.json()
-        except httpx.HTTPStatusError as exc:
-            raise ProviderUnavailableError(
-                f"ارائه‌دهندهٔ مدل خطا برگرداند ({exc.response.status_code})."
-            ) from exc
-        except httpx.HTTPError as exc:
-            raise ProviderUnavailableError("اتصال به ارائه‌دهندهٔ مدل برقرار نشد.") from exc
-
-        choice = (data.get("choices") or [{}])[0]
-        content = (choice.get("message") or {}).get("content", "") or ""
-        usage = data.get("usage") or {}
-        return ModelResponse(
-            content=content,
-            model_internal_name=data.get("model", model.internal_name),
-            input_tokens=int(usage.get("prompt_tokens", 0) or 0),
-            output_tokens=int(usage.get("completion_tokens", 0) or 0),
-            latency_ms=int((time.perf_counter() - started) * 1000),
-            raw={"id": data.get("id", "")},
-        )
-
-    async def health(self) -> ModelHealth:
-        if not self.base_url:
-            return ModelHealth(ok=False, detail="base_url تنظیم نشده است.")
-        try:
-            async with httpx.AsyncClient(timeout=10.0) as client:
-                response = await client.get(f"{self.base_url}/models")
-            return ModelHealth(ok=response.status_code < 500, detail=str(response.status_code))
-        except httpx.HTTPError as exc:
-            return ModelHealth(ok=False, detail=str(exc))

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

@@ -2,7 +2,9 @@
 
 One place decides which concrete adapter backs each port, driven purely by the
 ``*_PROVIDER`` environment switches. Everything defaults to ``mock`` so a fresh
-clone runs with no credentials.
+clone runs with no credentials — *except* the model provider: ``agy`` (Google
+Antigravity) is the only one the platform talks to, and its accounts are added
+from the admin console.
 
 Adapters that hold process state (the mock Git forge, the mock port ledger) are
 cached as singletons so behaviour is consistent within a process.
@@ -10,25 +12,28 @@ cached as singletons so behaviour is consistent within a process.
 
 from __future__ import annotations
 
+import logging
 from functools import lru_cache
 
 from app.core.config import settings
 from app.core.errors import ProviderUnavailableError
 from app.infrastructure.adapters import mock
-from app.infrastructure.adapters.openai_compatible import OpenAICompatibleModelAdapter
+
+logger = logging.getLogger("agy.registry")
 
 
 @lru_cache
 def get_model_provider():
-    provider = settings.ai_default_provider.lower()
+    """The one model adapter: agy over its logged-in account pool."""
-    if provider == "mock":
+    configured = settings.ai_default_provider.strip().lower()
-        return mock.MockModelProviderAdapter()
+    if configured != "agy":
-    if provider == "openai_compatible":
+        # Tolerate a stale ``AI_DEFAULT_PROVIDER`` from before agy became the
-        return OpenAICompatibleModelAdapter(
+        # only provider: warn loudly, keep serving with agy either way.
-            base_url=settings.openai_compatible_base_url,
+        logger.warning(
-            default_api_key=settings.openai_compatible_api_key,
+            "AI_DEFAULT_PROVIDER=%r is obsolete — agy is the only supported "
+            "model provider; using it",
+            settings.ai_default_provider,
         )
-    if provider == "agy":
     from app.infrastructure.adapters.agy import (
         AgyModelAdapter,
         load_agy_accounts,
@@ -36,10 +41,6 @@ def get_model_provider():
     )
 
     return AgyModelAdapter(loader=load_agy_accounts, report=report_agy_state)
-    raise ProviderUnavailableError(
-        f"AI_DEFAULT_PROVIDER='{provider}' پشتیبانی نمی‌شود. "
-        "مقادیر مجاز: mock | openai_compatible | agy"
-    )
 
 
 @lru_cache

apps/api/app/infrastructure/db/models.py

@@ -115,7 +115,10 @@ class AiProviderRow(Base):
 
     id: Mapped[str] = _pk()
     name: Mapped[str] = mapped_column(String(120), unique=True, nullable=False)
-    type: Mapped[str] = mapped_column(String(32), default="mock", nullable=False)
+    #: agy is the only provider type; the column stays a string so old rows
+    #: (mock / openai_compatible, deleted by migration c2d3e4f5b6a7) cannot
+    #: break the enum mapping.
+    type: Mapped[str] = mapped_column(String(32), default="agy", nullable=False)
     base_url: Mapped[str] = mapped_column(String(512), default="", nullable=False)
     is_active: Mapped[bool] = mapped_column(Boolean, default=True, nullable=False)
     created_at: Mapped[datetime] = _created()
@@ -160,6 +163,8 @@ class AiModelRow(Base):
     capabilities: Mapped[list] = mapped_column(JSON, default=list, nullable=False)
     input_price_per_1m: Mapped[float] = mapped_column(Float, default=0.0, nullable=False)
     output_price_per_1m: Mapped[float] = mapped_column(Float, default=0.0, nullable=False)
+    #: agy reasoning effort ("" = provider default); see ``AiModel``.
+    reasoning_effort: Mapped[str] = mapped_column(String(16), default="", nullable=False)
     is_active: Mapped[bool] = mapped_column(Boolean, default=True, nullable=False)
     created_at: Mapped[datetime] = _created()
 

apps/api/app/infrastructure/repositories/ai_control.py

@@ -33,8 +33,14 @@ from app.infrastructure.db.models import (
 
 
 def _provider_to_domain(row: AiProviderRow) -> Provider:
+    # agy is the only provider kind; a legacy row (mock / openai_compatible,
+    # deleted by migration) still maps to the one enum member rather than
+    # raising, so a half-migrated database reads cleanly.
+    provider_type = (
+        ProviderType(row.type) if row.type == ProviderType.AGY.value else ProviderType.AGY
+    )
     return Provider(
-        id=row.id, name=row.name, type=ProviderType(row.type),
+        id=row.id, name=row.name, type=provider_type,
         base_url=row.base_url, is_active=row.is_active, created_at=row.created_at,
     )
 
@@ -57,6 +63,7 @@ def _model_to_domain(row: AiModelRow) -> AiModel:
         capabilities=list(row.capabilities or []),
         input_price_per_1m=row.input_price_per_1m,
         output_price_per_1m=row.output_price_per_1m,
+        reasoning_effort=row.reasoning_effort or "",
         is_active=row.is_active, created_at=row.created_at,
     )
 
@@ -267,6 +274,7 @@ class AiModelRepository:
             capabilities=list(model.capabilities),
             input_price_per_1m=model.input_price_per_1m,
             output_price_per_1m=model.output_price_per_1m,
+            reasoning_effort=model.reasoning_effort,
             is_active=model.is_active, created_at=model.created_at,
         )
         self.session.add(row)

apps/api/app/interfaces/http/routers/admin.py

@@ -1,11 +1,12 @@
-"""Platform owner console: providers, models, routing policies and rate cards.
+"""Platform owner console: agy models, routing policies and rate cards.
 
 Everything in this router is invisible to customers. Two rules are enforced
 here and tested:
 
 1. Only a platform owner may call any of it.
-2. A provider API key is **never** returned — not masked-and-then-leaked, not in
+2. Model identity stays inside the console: a customer never receives a raw
-   a nested object. Only ``masked_key`` leaves this module.
+   model or account name, and no credential material (a Google refresh token)
+   is ever returned — only the agy account metadata is, from the agy router.
 """
 
 from __future__ import annotations
@@ -17,15 +18,12 @@ from pydantic import BaseModel, ConfigDict, Field
 from sqlalchemy.ext.asyncio import AsyncSession
 
 from app.application.services.audit_service import record_audit
-from app.core.crypto import encrypt, mask
 from app.core.database import get_session
 from app.core.errors import ConflictError, NotFoundError, ValidationError
 from app.domain.ai_control.models import (
+    REASONING_EFFORTS,
     AiModel,
     Capability,
-    ModelCredential,
-    Provider,
-    ProviderType,
     RateCard,
     RateUnit,
     RoutingMode,
@@ -36,7 +34,6 @@ from app.domain.base import utcnow
 from app.domain.identity.models import User
 from app.infrastructure.repositories.ai_control import (
     AiModelRepository,
-    ModelCredentialRepository,
     ProviderRepository,
     RateCardRepository,
     RoutingPolicyRepository,
@@ -73,38 +70,6 @@ router = APIRouter(
 # ---------------------------------------------------------------------------
 
 
-class ProviderOut(BaseModel):
-    model_config = ConfigDict(from_attributes=True)
-
-    id: str
-    name: str
-    type: str
-    base_url: str
-    is_active: bool
-    created_at: datetime
-
-
-class CreateProviderRequest(BaseModel):
-    name: str = Field(min_length=2, max_length=120)
-    type: ProviderType = ProviderType.MOCK
-    base_url: str = Field(default="", max_length=512)
-
-
-class CredentialOut(BaseModel):
-    """Deliberately has no field capable of holding the plaintext key."""
-
-    id: str
-    provider_id: str
-    name: str
-    masked_key: str
-    is_active: bool
-
-
-class CreateCredentialRequest(BaseModel):
-    name: str = Field(min_length=1, max_length=120)
-    api_key: str = Field(min_length=4, max_length=512)
-
-
 class ModelOut(BaseModel):
     model_config = ConfigDict(from_attributes=True)
 
@@ -115,16 +80,19 @@ class ModelOut(BaseModel):
     capabilities: list[str]
     input_price_per_1m: float
     output_price_per_1m: float
+    reasoning_effort: str
     is_active: bool
 
 
 class CreateModelRequest(BaseModel):
-    provider_id: str
+    #: The provider is no longer a client choice — agy is the only one, resolved
+    #: server-side. The model id must come from the account's live catalog.
     internal_name: str = Field(min_length=1, max_length=160)
-    display_name_for_admin: str = Field(min_length=1, max_length=160)
+    display_name_for_admin: str = Field(default="", max_length=160)
     capabilities: list[Capability] = Field(default_factory=list)
     input_price_per_1m: float = Field(default=0.0, ge=0)
     output_price_per_1m: float = Field(default=0.0, ge=0)
+    reasoning_effort: str = Field(default="", max_length=16)
 
 
 class RoutingPolicyOut(BaseModel):
@@ -180,124 +148,29 @@ class AdminUsageRow(BaseModel):
     total_price: float
 
 
-# ---------------------------------------------------------------------------
-# Providers
-# ---------------------------------------------------------------------------
-
-
-@router.get("/providers", response_model=list[ProviderOut])
-async def list_providers(session: AsyncSession = Depends(get_session)) -> list[ProviderOut]:
-    providers = await ProviderRepository(session).list()
-    return [
-        ProviderOut(
-            id=p.id, name=p.name, type=p.type.value, base_url=p.base_url,
-            is_active=p.is_active, created_at=p.created_at,
-        )
-        for p in providers
-    ]
-
-
-@router.post("/providers", response_model=ProviderOut, status_code=status.HTTP_201_CREATED)
-async def create_provider(
-    payload: CreateProviderRequest,
-    request: Request,
-    owner: User = Depends(require_platform_owner),
-    session: AsyncSession = Depends(get_session),
-) -> ProviderOut:
-    repo = ProviderRepository(session)
-    if await repo.get_by_name(payload.name) is not None:
-        raise ConflictError("ارائه‌دهنده‌ای با این نام وجود دارد.")
-    provider = await repo.create(
-        Provider(name=payload.name, type=payload.type, base_url=payload.base_url)
-    )
-    ip, ua = client_meta(request)
-    await record_audit(
-        session, action=AuditAction.PROVIDER_CHANGE, actor_user_id=owner.id,
-        target_type="provider", target_id=provider.id,
-        metadata={"name": provider.name, "type": provider.type.value},
-        ip=ip, user_agent=ua,
-    )
-    return ProviderOut(
-        id=provider.id, name=provider.name, type=provider.type.value,
-        base_url=provider.base_url, is_active=provider.is_active,
-        created_at=provider.created_at,
-    )
-
-
-@router.post(
-    "/providers/{provider_id}/credentials",
-    response_model=CredentialOut,
-    status_code=status.HTTP_201_CREATED,
-)
-async def create_credential(
-    provider_id: str,
-    payload: CreateCredentialRequest,
-    request: Request,
-    owner: User = Depends(require_platform_owner),
-    session: AsyncSession = Depends(get_session),
-) -> CredentialOut:
-    provider = await ProviderRepository(session).get(provider_id)
-    if provider is None:
-        raise NotFoundError("ارائه‌دهنده یافت نشد.")
-
-    credential = await ModelCredentialRepository(session).create(
-        ModelCredential(
-            provider_id=provider_id,
-            name=payload.name,
-            # Encrypted at rest; the plaintext exists only inside this request.
-            encrypted_api_key=encrypt(payload.api_key),
-        )
-    )
-    ip, ua = client_meta(request)
-    await record_audit(
-        session, action=AuditAction.PROVIDER_CHANGE, actor_user_id=owner.id,
-        target_type="credential", target_id=credential.id,
-        metadata={"provider_id": provider_id, "name": credential.name},
-        ip=ip, user_agent=ua,
-    )
-    return CredentialOut(
-        id=credential.id, provider_id=credential.provider_id, name=credential.name,
-        masked_key=mask(payload.api_key), is_active=credential.is_active,
-    )
-
-
-@router.get("/providers/{provider_id}/credentials", response_model=list[CredentialOut])
-async def list_credentials(
-    provider_id: str, session: AsyncSession = Depends(get_session)
-) -> list[CredentialOut]:
-    credentials = await ModelCredentialRepository(session).list_for_provider(provider_id)
-    return [
-        CredentialOut(
-            id=c.id, provider_id=c.provider_id, name=c.name,
-            # We cannot recover the plaintext to mask it — and we must not try.
-            masked_key="••••••",
-            is_active=c.is_active,
-        )
-        for c in credentials
-    ]
-
-
 # ---------------------------------------------------------------------------
 # Models
 # ---------------------------------------------------------------------------
 
 
+def _model_out(model: AiModel) -> ModelOut:
+    return ModelOut(
+        id=model.id, provider_id=model.provider_id, internal_name=model.internal_name,
+        display_name_for_admin=model.display_name_for_admin,
+        capabilities=list(model.capabilities),
+        input_price_per_1m=model.input_price_per_1m,
+        output_price_per_1m=model.output_price_per_1m,
+        reasoning_effort=model.reasoning_effort, is_active=model.is_active,
+    )
+
+
 @router.get("/models", response_model=list[ModelOut])
 async def list_models(
     provider_id: str | None = Query(default=None),
     session: AsyncSession = Depends(get_session),
 ) -> list[ModelOut]:
     models = await AiModelRepository(session).list(provider_id=provider_id)
-    return [
+    return [_model_out(m) for m in models]
-        ModelOut(
-            id=m.id, provider_id=m.provider_id, internal_name=m.internal_name,
-            display_name_for_admin=m.display_name_for_admin,
-            capabilities=list(m.capabilities),
-            input_price_per_1m=m.input_price_per_1m,
-            output_price_per_1m=m.output_price_per_1m, is_active=m.is_active,
-        )
-        for m in models
-    ]
 
 
 @router.post("/models", response_model=ModelOut, status_code=status.HTTP_201_CREATED)
@@ -307,33 +180,39 @@ async def create_model(
     owner: User = Depends(require_platform_owner),
     session: AsyncSession = Depends(get_session),
 ) -> ModelOut:
-    if await ProviderRepository(session).get(payload.provider_id) is None:
+    from app.interfaces.http.routers.agy import ensure_agy_provider
-        raise NotFoundError("ارائه‌دهنده یافت نشد.")
 
-    model = await AiModelRepository(session).create(
+    effort = payload.reasoning_effort.strip().lower()
+    if effort and effort not in REASONING_EFFORTS:
+        raise ValidationError(
+            "افورت نامعتبر است. مقادیر مجاز: " + " | ".join(REASONING_EFFORTS)
+        )
+
+    provider = await ensure_agy_provider(ProviderRepository(session))
+    models = AiModelRepository(session)
+    existing = await models.list(provider_id=provider.id)
+    if any(m.internal_name == payload.internal_name for m in existing):
+        raise ConflictError("مدلی با این نام داخلی از قبل ثبت شده است.")
+
+    model = await models.create(
         AiModel(
-            provider_id=payload.provider_id,
+            provider_id=provider.id,
             internal_name=payload.internal_name,
-            display_name_for_admin=payload.display_name_for_admin,
+            display_name_for_admin=payload.display_name_for_admin or payload.internal_name,
             capabilities=[str(c) for c in payload.capabilities],
             input_price_per_1m=payload.input_price_per_1m,
             output_price_per_1m=payload.output_price_per_1m,
+            reasoning_effort=effort,
         )
     )
     ip, ua = client_meta(request)
     await record_audit(
         session, action=AuditAction.MODEL_CHANGE, actor_user_id=owner.id,
         target_type="model", target_id=model.id,
-        metadata={"internal_name": model.internal_name},
+        metadata={"internal_name": model.internal_name, "reasoning_effort": effort},
         ip=ip, user_agent=ua,
     )
-    return ModelOut(
+    return _model_out(model)
-        id=model.id, provider_id=model.provider_id, internal_name=model.internal_name,
-        display_name_for_admin=model.display_name_for_admin,
-        capabilities=list(model.capabilities),
-        input_price_per_1m=model.input_price_per_1m,
-        output_price_per_1m=model.output_price_per_1m, is_active=model.is_active,
-    )
 
 
 # ---------------------------------------------------------------------------

apps/api/app/interfaces/http/routers/agy.py

@@ -110,7 +110,7 @@ class AgyTestOut(BaseModel):
 # ---------------------------------------------------------------------------
 
 
-async def _ensure_agy_provider(repo: ProviderRepository):
+async def ensure_agy_provider(repo: ProviderRepository):
     provider = await repo.get_by_name(AGY_PROVIDER_NAME)
     if provider is not None:
         return provider
@@ -176,7 +176,7 @@ async def complete_agy_login(
 
     providers = ProviderRepository(session)
     credentials = ModelCredentialRepository(session)
-    provider = await _ensure_agy_provider(providers)
+    provider = await ensure_agy_provider(providers)
 
     envelope = json.dumps(
         {
@@ -331,13 +331,29 @@ async def test_agy_account(
     )
 
 
-@router.get("/models", response_model=list[str])
+@router.get("/accounts/{credential_id}/models", response_model=list[str])
-async def list_agy_models() -> list[str]:
+async def list_account_models(
-    """Model ids the logged-in agy subscription exposes (registry helper)."""
+    credential_id: str,
-    from app.infrastructure.adapters.registry import get_model_provider
+    session: AsyncSession = Depends(get_session),
+) -> list[str]:
+    """Live model catalog of one account — fetched from Google on every call.
 
-    provider = get_model_provider()
+    The admin console calls this when an account is selected, so the model
-    lister = getattr(provider, "available_models", None)
+    list always reflects that subscription's current entitlement instead of a
-    if lister is None:
+    cached snapshot.
-        return []
+    """
-    return await lister()
+    from app.core.errors import ValidationError
+    from app.infrastructure.adapters.agy import AgyModelAdapter
+
+    repo = ModelCredentialRepository(session)
+    credential = await repo.get(credential_id)
+    if credential is None:
+        raise NotFoundError("اکانت agy یافت نشد.")
+
+    adapter = AgyModelAdapter()  # stateless: one forced account
+    try:
+        return await adapter.available_models_for(credential)
+    except ValidationError:
+        raise
+    except Exception as exc:  # noqa: BLE001 - the console wants the reason
+        raise ValidationError(str(exc)) from exc

apps/api/app/seed.py

@@ -2,9 +2,9 @@
 
 Run with ``pnpm db:seed``. Creates a platform owner, a demo user, a sample
 organization that both of them own, the system agent templates, routing
-policies, rate cards and a mock provider with two models. The agy (Google
+policies, rate cards and the agy (Google Antigravity) provider with its model
-Antigravity) provider and its model catalog are seeded too; agy models become
+catalog. agy is the only model provider the platform talks to; accounts are
-routable only when ``AI_DEFAULT_PROVIDER=agy``.
+added from the admin console.
 
 Safe to run repeatedly: every step checks for existing rows first.
 
@@ -20,7 +20,6 @@ import logging
 
 from sqlalchemy import select
 
-from app.core.config import settings
 from app.core.database import Base, SessionFactory, engine
 from app.core.security import hash_password
 from app.domain.ai_control.models import (
@@ -64,14 +63,17 @@ DEMO_ORG_NAME = "سازمان نمونه"
 DEMO_ORG_SLUG = "sample-org"
 
 #: The agy (Google Antigravity) model catalog. ``internal_name`` is the model id
-#: the Antigravity backend expects; the display name is admin-only. Seeded for
+#: the Antigravity backend expects; the display name is admin-only. The 4th
-#: every deployment, activated for routing only when agy is the default
+#: element is the reasoning effort (empty = provider default) — the console can
-#: provider — see the seed step that consumes this table.
+#: change it per model, and the adapter only sends it to thinking-capable
-AGY_MODELS: list[tuple[str, str, list[Capability]]] = [
+#: models. Seeded for every deployment and reconciled on *every* run, so a
+#: database created by an older version converges without being rebuilt.
+AGY_MODELS: list[tuple[str, str, list[Capability], str]] = [
     (
         "gemini-3-flash",
         "Gemini 3 Flash",
         [Capability.TEXT, Capability.CODE, Capability.LONG_CONTEXT],
+        "",
     ),
     (
         "gemini-3-pro-high",
@@ -80,23 +82,27 @@ AGY_MODELS: list[tuple[str, str, list[Capability]]] = [
             Capability.TEXT, Capability.REASONING, Capability.CODE,
             Capability.VISION, Capability.LONG_CONTEXT,
         ],
+        "high",
     ),
     (
         "claude-sonnet-4-5",
         "Claude Sonnet 4.5",
         [Capability.TEXT, Capability.CODE, Capability.TOOLS],
+        "",
     ),
     (
         "claude-sonnet-4-5-thinking",
         "Claude Sonnet 4.5 (Thinking)",
         [Capability.TEXT, Capability.CODE, Capability.REASONING],
+        "high",
     ),
     (
         "claude-opus-4-5-thinking",
         "Claude Opus 4.5 (Thinking)",
         [Capability.TEXT, Capability.REASONING, Capability.CODE, Capability.LONG_CONTEXT],
+        "high",
     ),
-    ("gpt-oss-120b-medium", "GPT-OSS 120B Medium", [Capability.TEXT, Capability.CODE]),
+    ("gpt-oss-120b-medium", "GPT-OSS 120B Medium", [Capability.TEXT, Capability.CODE], ""),
 ]
 
 AGENT_TEMPLATES: list[tuple[str, RoleKey, str, list[str]]] = [
@@ -270,51 +276,11 @@ async def seed() -> None:
         # as either one lands on a populated dashboard rather than an empty one.
         await _ensure_org(session, members=[owner, demo])
 
-        if not (await session.execute(select(AiProviderRow))).scalars().first():
+        # The agy catalog (Google Antigravity). agy is the only model provider,
-            provider = Provider(name="mock", type=ProviderType.MOCK, base_url="")
+        # so the provider row and its models are created on a first run and
-            session.add(
+        # reconciled on every later one: display names, capabilities, efforts
-                AiProviderRow(
+        # and activation follow this table, and a model removed here stays in
-                    id=provider.id, name=provider.name, type=provider.type.value,
+        # the database (neither table nor row deletion is safe for a seed).
-                    base_url=provider.base_url, is_active=True,
-                )
-            )
-            await session.flush()
-            for internal, display, caps in (
-                ("mock-fast", "Mock Fast", [Capability.TEXT]),
-                (
-                    "mock-reasoning",
-                    "Mock Reasoning",
-                    [Capability.TEXT, Capability.REASONING, Capability.CODE],
-                ),
-            ):
-                model = AiModel(
-                    provider_id=provider.id,
-                    internal_name=internal,
-                    display_name_for_admin=display,
-                    capabilities=[str(c) for c in caps],
-                    input_price_per_1m=0.3,
-                    output_price_per_1m=1.2,
-                )
-                session.add(
-                    AiModelRow(
-                        id=model.id, provider_id=model.provider_id,
-                        internal_name=model.internal_name,
-                        display_name_for_admin=model.display_name_for_admin,
-                        capabilities=model.capabilities,
-                        input_price_per_1m=model.input_price_per_1m,
-                        output_price_per_1m=model.output_price_per_1m,
-                        is_active=True,
-                    )
-                )
-            logger.info("created mock provider + 2 models")
-
-        # The agy catalog (Google Antigravity). It is seeded for every
-        # deployment — the admin console needs the provider and its models to
-        # exist before any account is logged in. Routing activation is
-        # reconciled on *every* seed run below, not only at insert time, so
-        # flipping AI_DEFAULT_PROVIDER and re-seeding reshapes routing without
-        # a fresh database.
-        agy_is_default = settings.ai_default_provider.strip().lower() == "agy"
         agy_provider_row = (
             await session.execute(
                 select(AiProviderRow).where(AiProviderRow.name == "agy")
@@ -328,7 +294,7 @@ async def seed() -> None:
             )
             session.add(agy_provider_row)
             await session.flush()
-        for internal, display, caps in AGY_MODELS:
+        for internal, display, caps, effort in AGY_MODELS:
             existing = (
                 await session.execute(
                     select(AiModelRow).where(
@@ -338,6 +304,12 @@ async def seed() -> None:
                 )
             ).scalars().first()
             if existing is not None:
+                # Reconcile: an older database has these rows with different
+                # metadata (efforts did not exist, capabilities moved).
+                existing.display_name_for_admin = display
+                existing.capabilities = [str(c) for c in caps]
+                existing.reasoning_effort = effort
+                existing.is_active = True
                 continue
             model = AiModel(
                 provider_id=agy_provider_row.id,
@@ -349,6 +321,7 @@ async def seed() -> None:
                 # customer-facing price on the rate card.
                 input_price_per_1m=0.0,
                 output_price_per_1m=0.0,
+                reasoning_effort=effort,
             )
             session.add(
                 AiModelRow(
@@ -358,39 +331,12 @@ async def seed() -> None:
                     capabilities=model.capabilities,
                     input_price_per_1m=model.input_price_per_1m,
                     output_price_per_1m=model.output_price_per_1m,
-                    is_active=False,
+                    reasoning_effort=model.reasoning_effort,
+                    is_active=True,
                 )
             )
-        # Reconcile routing with the configured default on every run: agy models
-        # are routable only when agy is the default, and the mock catalog must
-        # not win when it is not — a model without accounts fails at call time.
-        agy_model_rows = (
-            await session.execute(
-                select(AiModelRow).where(
-                    AiModelRow.provider_id == agy_provider_row.id
-                )
-            )
-        ).scalars()
-        for row in agy_model_rows:
-            row.is_active = agy_is_default
-        if agy_is_default:
-            mock_provider_row = (
-                await session.execute(
-                    select(AiProviderRow).where(AiProviderRow.name == "mock")
-                )
-            ).scalars().first()
-            if mock_provider_row is not None:
-                mock_model_rows = (
-                    await session.execute(
-                        select(AiModelRow).where(
-                            AiModelRow.provider_id == mock_provider_row.id
-                        )
-                    )
-                ).scalars()
-                for row in mock_model_rows:
-                    row.is_active = False
         logger.info(
-                "agy is the default provider; agy catalog activated "
+            "agy is the only model provider; catalog active "
             "(accounts are added from the admin console)"
         )
 

apps/api/tests/conftest.py

@@ -1,7 +1,10 @@
 """Shared test fixtures.
 
-Tests run against an in-memory SQLite database and the ``mock`` adapters, so the
+Tests run against an in-memory SQLite database with the real agy-shaped adapter
-suite needs no Postgres, no Redis and no credentials.
+wiring and a fake model provider, so the suite needs no Postgres, no Redis, no
+Google and no accounts. agy is the only model provider the platform talks to
+(see ``app/infrastructure/adapters/registry.py``); the fake keeps completions
+deterministic and offline.
 """
 
 from __future__ import annotations
@@ -96,6 +99,76 @@ async def client(engine, session_factory):
     app.dependency_overrides.clear()
 
 
+# ---------------------------------------------------------------------------
+# Fake model provider — the only sanctioned stand-in for agy in tests
+# ---------------------------------------------------------------------------
+
+
+class _FakeModelProvider:
+    """Offline stand-in for the agy adapter.
+
+    The canned text must read like model output, not like a debug tag: the
+    provider/model identity is platform-internal and must never surface in
+    anything a customer can read (see the e2e smoke test).
+    """
+
+    async def complete(
+        self,
+        *,
+        model,
+        messages,
+        system="",
+        temperature=0.2,
+        max_tokens=2048,
+        credential=None,
+        credentials=None,
+    ):
+        import asyncio
+        import time
+
+        from app.application.ports import ModelResponse
+
+        started = time.perf_counter()
+        last_user = next((m.content for m in reversed(messages) if m.role == "user"), "")
+        prompt_tokens = sum(len(m.content.split()) for m in messages)
+        # A tiny sleep so latency bookkeeping runs a real code path.
+        await asyncio.sleep(0)
+        content = (
+            "این پاسخ آزمایشی است. "
+            f"درخواست دریافتی: {last_user[:280] or '—'}"
+        )
+        return ModelResponse(
+            content=content,
+            model_internal_name=model.internal_name,
+            input_tokens=prompt_tokens + len(system.split()),
+            output_tokens=len(content.split()),
+            latency_ms=int((time.perf_counter() - started) * 1000),
+            raw={},
+        )
+
+    async def health(self):
+        from app.application.ports import ModelHealth
+
+        return ModelHealth(ok=True, detail="fake agy pool")
+
+
+@pytest.fixture(autouse=True)
+def _fake_model_provider(monkeypatch):
+    """Swap the real model adapter out of every request path.
+
+    ``container`` and the health router bind ``get_model_provider`` at import
+    time, so both import sites are patched; without this a completion test
+    would open the developer's real database and reach the real Google API.
+    """
+    from app.application import container as di_container
+    from app.interfaces.http.routers import health as health_router
+
+    provider = _FakeModelProvider()
+    monkeypatch.setattr(di_container, "get_model_provider", lambda: provider)
+    monkeypatch.setattr(health_router, "get_model_provider", lambda: provider)
+    return provider
+
+
 async def signup(
     client: AsyncClient, email: str, password: str = "password123", name: str = ""
 ) -> dict:
@@ -149,10 +222,31 @@ SYSTEM_ROUTING_POLICIES: list[tuple[str, str, bool, str]] = [
     ("استدلال عمیق", "reasoning", False, "برای تحلیل پیچیده."),
 ]
 
-#: The mock models ``app/seed.py`` installs: (internal_name, display, capabilities).
+#: The agy models ``app/seed.py`` installs: (internal_name, display,
-SYSTEM_MODELS: list[tuple[str, str, list[str]]] = [
+#: capabilities, reasoning_effort). agy is the only provider, so this is the
-    ("mock-fast", "Mock Fast", ["text"]),
+#: whole catalogue.
-    ("mock-reasoning", "Mock Reasoning", ["text", "reasoning", "code"]),
+SYSTEM_MODELS: list[tuple[str, str, list[str], str]] = [
+    ("gemini-3-flash", "Gemini 3 Flash", ["text", "code", "long_context"], ""),
+    (
+        "gemini-3-pro-high",
+        "Gemini 3 Pro",
+        ["text", "reasoning", "code", "vision", "long_context"],
+        "high",
+    ),
+    ("claude-sonnet-4-5", "Claude Sonnet 4.5", ["text", "code", "tools"], ""),
+    (
+        "claude-sonnet-4-5-thinking",
+        "Claude Sonnet 4.5 (Thinking)",
+        ["text", "code", "reasoning"],
+        "high",
+    ),
+    (
+        "claude-opus-4-5-thinking",
+        "Claude Opus 4.5 (Thinking)",
+        ["text", "reasoning", "code", "long_context"],
+        "high",
+    ),
+    ("gpt-oss-120b-medium", "GPT-OSS 120B Medium", ["text", "code"], ""),
 ]
 
 #: The rate cards ``app/seed.py`` installs: (unit, customer_price, internal_cost).
@@ -174,11 +268,11 @@ SYSTEM_SANDBOX_TEMPLATES: list[tuple[str, str, str]] = [
 async def seeded(session_factory):
     """Install the platform-level seed rows a fresh install would have.
 
-    Agent templates, routing policies, the mock provider/models, the rate cards
+    Agent templates, routing policies, the agy provider with its model catalog,
-    and the sandbox templates are all *platform* data (no organization owns
+    the rate cards and the sandbox templates are all *platform* data (no
-    them), so without this an employee cannot be created, a run cannot resolve a
+    organization owns them), so without this an employee cannot be created, a
-    model to route to, usage cannot be priced and no sandbox can be started —
+    run cannot resolve a model to route to, usage cannot be priced and no
-    none of which the real seeded install lacks.
+    sandbox can be started — none of which the real seeded install lacks.
     """
     from app.infrastructure.db.models import (
         AgentTemplateRow,
@@ -214,9 +308,11 @@ async def seeded(session_factory):
         # surfaces as a ``ProviderUnavailableError`` and a *failed* run.
         provider_id = new_id()
         session.add(
-            AiProviderRow(id=provider_id, name="mock", type="mock", base_url="", is_active=True)
+            AiProviderRow(
+                id=provider_id, name="agy", type="agy", base_url="", is_active=True
             )
-        for internal, display, capabilities in SYSTEM_MODELS:
+        )
+        for internal, display, capabilities, effort in SYSTEM_MODELS:
             session.add(
                 AiModelRow(
                     provider_id=provider_id,
@@ -225,6 +321,7 @@ async def seeded(session_factory):
                     capabilities=capabilities,
                     input_price_per_1m=0.3,
                     output_price_per_1m=1.2,
+                    reasoning_effort=effort,
                     is_active=True,
                 )
             )

apps/api/tests/test_admin_control_plane.py

@@ -3,7 +3,9 @@
 Two properties matter more than any feature here:
 
 1. A normal customer can never reach these endpoints.
-2. A provider API key can never come back out of the API.
+2. Model identity and account material stay inside the console — a customer
+   never sees a raw model, and nothing that could act as a credential (a Google
+   refresh token, an API key) ever comes back out.
 """
 
 from __future__ import annotations
@@ -12,146 +14,92 @@ from httpx import AsyncClient
 
 from tests.conftest import auth_header, signup
 
-SECRET_KEY = "sk-super-secret-value-do-not-leak-1234567890"
 
-
+async def test_normal_user_cannot_list_models(client: AsyncClient):
-async def test_normal_user_cannot_list_providers(client: AsyncClient):
     token = (await signup(client, "peon@example.com"))["access_token"]
-    response = await client.get("/api/admin/providers", headers=auth_header(token))
+    response = await client.get("/api/admin/models", headers=auth_header(token))
     assert response.status_code == 403
     assert response.json()["error"]["code"] == "permission_denied"
 
 
-async def test_normal_user_cannot_create_provider(client: AsyncClient):
+async def test_normal_user_cannot_create_model(client: AsyncClient):
     token = (await signup(client, "peon2@example.com"))["access_token"]
     response = await client.post(
-        "/api/admin/providers",
+        "/api/admin/models",
-        json={"name": "evil", "type": "mock", "base_url": ""},
+        json={"internal_name": "evil-model"},
         headers=auth_header(token),
     )
     assert response.status_code == 403
 
 
 async def test_admin_endpoints_require_authentication(client: AsyncClient):
-    response = await client.get("/api/admin/providers")
+    response = await client.get("/api/admin/models")
     assert response.status_code == 401
 
 
-async def test_platform_owner_can_create_and_list_provider(
+async def test_model_creation_resolves_the_agy_provider_itself(
     client: AsyncClient, platform_owner_token: str
 ):
+    """The provider is not a client choice: agy is the only one."""
     created = await client.post(
-        "/api/admin/providers",
+        "/api/admin/models",
-        json={"name": "openrouter", "type": "openrouter", "base_url": "https://openrouter.ai/api/v1"},
+        json={
+            "internal_name": "gemini-3-pro-high",
+            "capabilities": ["text", "reasoning"],
+            "input_price_per_1m": 0.3,
+            "output_price_per_1m": 1.2,
+            "reasoning_effort": "high",
+        },
         headers=auth_header(platform_owner_token),
     )
     assert created.status_code == 201, created.text
-    assert created.json()["name"] == "openrouter"
+    body = created.json()
+    assert body["internal_name"] == "gemini-3-pro-high"
+    # The display name falls back to the model id; nothing is typed twice.
+    assert body["display_name_for_admin"] == "gemini-3-pro-high"
+    assert body["reasoning_effort"] == "high"
+    assert body["provider_id"], "the agy provider must be resolved server-side"
 
-    listing = await client.get("/api/admin/providers", headers=auth_header(platform_owner_token))
+    listing = await client.get("/api/admin/models", headers=auth_header(platform_owner_token))
     assert listing.status_code == 200
-    assert [p["name"] for p in listing.json()] == ["openrouter"]
+    assert [m["internal_name"] for m in listing.json()] == ["gemini-3-pro-high"]
 
 
-async def test_duplicate_provider_name_is_rejected(client: AsyncClient, platform_owner_token: str):
+async def test_duplicate_model_name_is_rejected(client: AsyncClient, platform_owner_token: str):
-    payload = {"name": "dupe", "type": "mock", "base_url": ""}
+    payload = {"internal_name": "claude-sonnet-4-5-thinking", "reasoning_effort": "high"}
     first = await client.post(
-        "/api/admin/providers", json=payload, headers=auth_header(platform_owner_token)
+        "/api/admin/models", json=payload, headers=auth_header(platform_owner_token)
     )
     assert first.status_code == 201
     second = await client.post(
-        "/api/admin/providers", json=payload, headers=auth_header(platform_owner_token)
+        "/api/admin/models", json=payload, headers=auth_header(platform_owner_token)
     )
     assert second.status_code == 409
 
 
-async def test_credential_creation_returns_only_a_masked_key(
+async def test_unknown_reasoning_effort_is_rejected(
     client: AsyncClient, platform_owner_token: str
 ):
-    provider = (
-        await client.post(
-            "/api/admin/providers",
-            json={"name": "p1", "type": "openai_compatible", "base_url": "https://x.test/v1"},
-            headers=auth_header(platform_owner_token),
-        )
-    ).json()
-
     response = await client.post(
-        f"/api/admin/providers/{provider['id']}/credentials",
+        "/api/admin/models",
-        json={"name": "primary", "api_key": SECRET_KEY},
+        json={"internal_name": "gemini-3-flash", "reasoning_effort": "ultra"},
         headers=auth_header(platform_owner_token),
     )
-    assert response.status_code == 201, response.text
+    assert response.status_code == 422
-    body = response.json()
-    assert SECRET_KEY not in str(body)
-    assert body["masked_key"].startswith("•")
-    assert body["masked_key"].endswith(SECRET_KEY[-4:])
 
 
-async def test_listing_credentials_never_exposes_the_key(
+async def test_agy_account_and_model_endpoints_are_owner_only(
     client: AsyncClient, platform_owner_token: str
 ):
-    provider = (
+    """The account catalog is owner-only and never carries token material."""
-        await client.post(
+    token = (await signup(client, "peon3@example.com"))["access_token"]
-            "/api/admin/providers",
-            json={"name": "p2", "type": "openai_compatible", "base_url": "https://y.test/v1"},
-            headers=auth_header(platform_owner_token),
-        )
-    ).json()
-    await client.post(
-        f"/api/admin/providers/{provider['id']}/credentials",
-        json={"name": "primary", "api_key": SECRET_KEY},
-        headers=auth_header(platform_owner_token),
-    )
-
     response = await client.get(
-        f"/api/admin/providers/{provider['id']}/credentials",
+        "/api/admin/agy/accounts/whatever/models", headers=auth_header(token)
-        headers=auth_header(platform_owner_token),
     )
-    assert response.status_code == 200
+    assert response.status_code == 403
-    assert SECRET_KEY not in response.text
-    assert response.json()[0]["masked_key"] == "••••••"
 
-
+    unauthenticated = await client.get("/api/admin/agy/accounts/whatever/models")
-async def test_model_creation_and_listing(client: AsyncClient, platform_owner_token: str):
+    assert unauthenticated.status_code == 401
-    provider = (
-        await client.post(
-            "/api/admin/providers",
-            json={"name": "p3", "type": "mock", "base_url": ""},
-            headers=auth_header(platform_owner_token),
-        )
-    ).json()
-
-    created = await client.post(
-        "/api/admin/models",
-        json={
-            "provider_id": provider["id"],
-            "internal_name": "internal-coder-v1",
-            "display_name_for_admin": "Coder V1",
-            "capabilities": ["text", "code"],
-            "input_price_per_1m": 0.3,
-            "output_price_per_1m": 1.2,
-        },
-        headers=auth_header(platform_owner_token),
-    )
-    assert created.status_code == 201, created.text
-
-    listing = await client.get("/api/admin/models", headers=auth_header(platform_owner_token))
-    assert listing.status_code == 200
-    assert listing.json()[0]["internal_name"] == "internal-coder-v1"
-
-
-async def test_model_requires_existing_provider(client: AsyncClient, platform_owner_token: str):
-    response = await client.post(
-        "/api/admin/models",
-        json={
-            "provider_id": "does-not-exist",
-            "internal_name": "x",
-            "display_name_for_admin": "X",
-        },
-        headers=auth_header(platform_owner_token),
-    )
-    assert response.status_code == 404
 
 
 async def test_routing_policy_creation_and_default_swap(

apps/api/tests/test_agy.py

@@ -344,41 +344,42 @@ async def test_seed_creates_agy_catalog(client, seeded_install, session_factory)
         "claude-opus-4-5-thinking",
         "gpt-oss-120b-medium",
     }
-    # Default deployment keeps routing on mock: agy models wait inactive.
+    # agy is the only provider: the catalog is routable from the first run.
-    assert all(active is False for _, active in models)
-
-
-async def test_seed_activates_agy_when_it_is_default(
-    monkeypatch, client, seeded_install, session_factory
-):
-    from app.core.config import settings
-
-    monkeypatch.setattr(settings, "ai_default_provider", "agy")
-    await seeded_install.seed()
-    models = await _agy_models(session_factory)
     assert all(active is True for _, active in models)
 
 
-async def test_seed_deactivates_mock_when_agy_is_default(
+async def test_seed_round_trips_efforts(client, seeded_install, session_factory):
-    monkeypatch, client, seeded_install, session_factory
+    """The seed's effort column is what the adapter later sends to Google."""
-):
     from sqlalchemy import select
 
-    from app.core.config import settings
     from app.infrastructure.db.models import AiModelRow, AiProviderRow
 
-    monkeypatch.setattr(settings, "ai_default_provider", "agy")
-    await seeded_install.seed()
-
     async with session_factory() as session:
         rows = (
             await session.execute(
                 select(AiModelRow)
                 .join(AiProviderRow, AiModelRow.provider_id == AiProviderRow.id)
-                .where(AiProviderRow.name == "mock")
+                .where(AiProviderRow.name == "agy")
             )
         ).scalars()
-        assert all(row.is_active is False for row in rows)
+        efforts = {row.internal_name: row.reasoning_effort for row in rows}
+    assert efforts["gemini-3-pro-high"] == "high"
+    assert efforts["claude-sonnet-4-5-thinking"] == "high"
+    assert efforts["gemini-3-flash"] == ""
+
+
+async def test_seed_leaves_no_legacy_providers(client, seeded_install, session_factory):
+    """agy is the only provider row a seeded install may carry."""
+    from sqlalchemy import select
+
+    from app.infrastructure.db.models import AiProviderRow
+
+    async with session_factory() as session:
+        rows = (await session.execute(select(AiProviderRow))).scalars()
+        types = {row.type for row in rows}
+        names = {row.name for row in rows}
+    assert types == {"agy"}
+    assert "mock" not in names
 
 
 # ---------------------------------------------------------------------------
@@ -519,6 +520,64 @@ def test_payload_translation(decrypted):
     assert all(s["threshold"] == "OFF" for s in payload["request"]["safetySettings"])
 
 
+def test_effort_becomes_a_thinking_config_on_a_thinking_model():
+    """Effort is translated, not forwarded: high → a 16384-token budget."""
+    account = agy.AgyAccount(
+        credential_id="c1", email="a@b.c", project_id="proj-1", refresh_token="refresh-1"
+    )
+    payload = agy._build_generate_payload(
+        account,
+        "claude-sonnet-4-5-thinking",
+        [ChatMessage(role="user", content="سلام")],
+        system="",
+        temperature=0.4,
+        max_tokens=512,
+        reasoning_effort="high",
+    )
+    generation = payload["request"]["generationConfig"]
+    assert generation["thinkingConfig"] == {
+        "thinkingBudget": 16384,
+        "includeThoughts": True,
+    }
+    # The budget must fit inside the answer window.
+    assert generation["maxOutputTokens"] == 16384 + 8000
+
+
+def test_effort_is_not_sent_to_a_non_thinking_model():
+    """``thinkingConfig`` on a model that cannot think is a hard 400."""
+    account = agy.AgyAccount(
+        credential_id="c1", email="a@b.c", project_id="proj-1", refresh_token="refresh-1"
+    )
+    payload = agy._build_generate_payload(
+        account,
+        "claude-sonnet-4-5",
+        [ChatMessage(role="user", content="سلام")],
+        system="",
+        temperature=0.4,
+        max_tokens=512,
+        reasoning_effort="high",
+    )
+    assert "thinkingConfig" not in payload["request"]["generationConfig"]
+    assert payload["request"]["generationConfig"]["maxOutputTokens"] == 512
+
+
+def test_unknown_effort_falls_back_to_medium():
+    account = agy.AgyAccount(
+        credential_id="c1", email="a@b.c", project_id="proj-1", refresh_token="refresh-1"
+    )
+    payload = agy._build_generate_payload(
+        account,
+        "claude-opus-4-5-thinking",
+        [ChatMessage(role="user", content="سلام")],
+        system="",
+        temperature=0.4,
+        max_tokens=512,
+        reasoning_effort="turbo",
+    )
+    budget = payload["request"]["generationConfig"]["thinkingConfig"]["thinkingBudget"]
+    assert budget == 8192
+
+
 def test_response_parsing_skips_thoughts():
     model = AiModel(provider_id="p", internal_name="m", display_name_for_admin="m")
     response = agy._parse_generate_response(
@@ -713,3 +772,96 @@ async def test_available_models(google, decrypted):
     adapter = agy.AgyModelAdapter(loader=loader)
     models = await adapter.available_models()
     assert models == ["claude-sonnet-4-5", "gemini-3-flash"]
+
+
+# ---------------------------------------------------------------------------
+# Live per-account model catalog (the console's model dropdown source)
+# ---------------------------------------------------------------------------
+
+
+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")
+    )
+    assert models == ["claude-sonnet-4-5", "gemini-3-flash"]
+    # The list is fetched from Google every call — no cached catalog.
+    fetches = [c for c in google if c["url"].endswith("fetchAvailableModels")]
+    assert len(fetches) == 1
+
+
+async def test_available_models_for_account_without_material():
+    """A credential with no refresh token can never list models."""
+    adapter = agy.AgyModelAdapter()
+    credential = ModelCredential(
+        provider_id="p",
+        name="empty@gmail.com",
+        encrypted_api_key="PLAINTEXT::{}",
+        email="empty@gmail.com",
+    )
+    with pytest.raises(ProviderUnavailableError):
+        await adapter.available_models_for(credential)
+
+
+async def test_account_models_endpoint(client, google, platform_owner_token):
+    """GET /accounts/{id}/models returns that account's live catalog."""
+    started = await client.post(
+        "/api/admin/agy/accounts/login/start",
+        headers=auth_header(platform_owner_token),
+    )
+    state = parse_qs(urlparse(started.json()["auth_url"]).query)["state"][0]
+    account_id = (
+        await client.post(
+            "/api/admin/agy/accounts/login/complete",
+            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}"
+                ),
+            },
+        )
+    ).json()["id"]
+
+    response = await client.get(
+        f"/api/admin/agy/accounts/{account_id}/models",
+        headers=auth_header(platform_owner_token),
+    )
+    assert response.status_code == 200, response.text
+    assert response.json() == ["claude-sonnet-4-5", "gemini-3-flash"]
+
+    missing = await client.get(
+        "/api/admin/agy/accounts/does-not-exist/models",
+        headers=auth_header(platform_owner_token),
+    )
+    assert missing.status_code == 404
+
+
+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()
+        if url.endswith("/v1internal:generateContent"):
+            return _FakeResponse(json_data=GOOD_GENERATE)
+        return await default_handler(method, url, json_body, headers, data)
+
+    calls = patch_google(monkeypatch, handler)
+    model = AiModel(
+        provider_id="p",
+        internal_name="claude-sonnet-4-5-thinking",
+        display_name_for_admin="f",
+        reasoning_effort="low",
+    )
+    adapter = agy.AgyModelAdapter()
+    await adapter.complete(
+        model=model,
+        credentials=[_agy_credential(email="one@gmail.com")],
+        messages=_MESSAGES,
+    )
+    generate = [c for c in calls if c["url"].endswith("generateContent")]
+    assert generate[0]["json"]["request"]["generationConfig"]["thinkingConfig"] == {
+        "thinkingBudget": 4096,
+        "includeThoughts": True,
+    }

apps/api/tests/test_e2e_smoke.py

@@ -1,7 +1,9 @@
 """§18.3 — the minimal end-to-end smoke path.
 
 One test walks the whole product in order, through the real HTTP API and the
-real application services, on the mock adapters:
+real application services, against the agy-shaped catalogue (the seeded fixture
+installs the agy provider and its models) with the fake model provider from
+conftest standing in for the Google round-trip:
 
     signup -> organization -> project -> repo -> preview domain -> sandbox
            -> employee -> manager message -> task -> run -> usage -> audit

apps/web/src/app/(app)/admin/layout.tsx

@@ -17,7 +17,6 @@ import { useSession } from "@/lib/session";
 const SECTIONS = [
   { href: "/admin", label: "نمای کلی" },
   { href: "/admin/agy-accounts", label: "اکانت‌های agy" },
-  { href: "/admin/providers", label: "ارائه‌دهنده‌ها" },
   { href: "/admin/models", label: "مدل‌ها" },
   { href: "/admin/routing", label: "مسیریابی" },
   { href: "/admin/rate-cards", label: "تعرفه‌ها" },

apps/web/src/app/(app)/admin/models/page.tsx

@@ -1,7 +1,14 @@
 "use client";
 
 import * as React from "react";
-import type { AdminModel, AdminProvider, Capability } from "@agy/contracts";
+import {
+  effortsForModel,
+  effortLabel,
+  type AdminModel,
+  type AgyAccount,
+  type Capability,
+  type ReasoningEffort,
+} from "@agy/contracts";
 import {
   Badge,
   Button,
@@ -17,17 +24,34 @@ import {
 } from "@agy/ui";
 import { ApiError, api } from "@/lib/api";
 
+/**
+ * Model catalogue (platform owner only).
+ *
+ * agy is the only provider, and nothing about a model is typed by hand: the
+ * console asks the selected *account* for its live catalogue, the owner picks a
+ * model and a reasoning effort, and the rest is metadata. The model id itself
+ * never leaves the console — a customer only ever sees routing choices.
+ */
+
 const CAPABILITIES: Capability[] = ["text", "vision", "tools", "reasoning", "long_context", "code"];
 
+/** The single provider the platform talks to. */
+const PROVIDER_OPTIONS = [{ value: "agy", label: "agy (Google Antigravity)" }];
+
 export default function AdminModelsPage() {
   const [models, setModels] = React.useState<AdminModel[]>([]);
-  const [providers, setProviders] = React.useState<AdminProvider[]>([]);
+  const [accounts, setAccounts] = React.useState<AgyAccount[]>([]);
   const [loading, setLoading] = React.useState(true);
   const [error, setError] = React.useState<string | null>(null);
 
-  const [providerId, setProviderId] = React.useState("");
+  // The add-model cascade: provider -> account -> model -> effort.
-  const [internalName, setInternalName] = React.useState("");
+  const [providerId, setProviderId] = React.useState("agy");
-  const [displayName, setDisplayName] = React.useState("");
+  const [accountId, setAccountId] = React.useState("");
+  const [catalog, setCatalog] = React.useState<string[] | null>(null);
+  const [catalogLoading, setCatalogLoading] = React.useState(false);
+  const [catalogError, setCatalogError] = React.useState<string | null>(null);
+  const [modelId, setModelId] = React.useState("");
+  const [effort, setEffort] = React.useState("");
   const [caps, setCaps] = React.useState<Capability[]>(["text"]);
   const [inputPrice, setInputPrice] = React.useState("");
   const [outputPrice, setOutputPrice] = React.useState("");
@@ -38,13 +62,12 @@ export default function AdminModelsPage() {
     setLoading(true);
     setError(null);
     try {
-      const [nextModels, nextProviders] = await Promise.all([
+      const [nextModels, nextAccounts] = await Promise.all([
         api.adminModels(),
-        api.adminProviders(),
+        api.agyAccounts(),
       ]);
       setModels(nextModels);
-      setProviders(nextProviders);
+      setAccounts(nextAccounts);
-      setProviderId((current) => current || nextProviders[0]?.id || "");
     } catch (err) {
       setError(err instanceof ApiError ? err.message : "دریافت مدل‌ها ناموفق بود.");
     } finally {
@@ -56,6 +79,72 @@ export default function AdminModelsPage() {
     void load();
   }, [load]);
 
+  /** Reload only the accounts — the provider step of the cascade. */
+  const loadAccounts = React.useCallback(async () => {
+    try {
+      setAccounts(await api.agyAccounts());
+    } catch (err) {
+      setError(err instanceof ApiError ? err.message : "دریافت اکانت‌ها ناموفق بود.");
+    }
+  }, []);
+
+  /**
+   * Live catalogue of the selected account — fetched from Google on every
+   * account change (and on explicit refresh), never from a cached list.
+   */
+  const loadCatalog = React.useCallback(async (id: string) => {
+    if (!id) {
+      setCatalog(null);
+      setCatalogError(null);
+      return;
+    }
+    setCatalogLoading(true);
+    setCatalogError(null);
+    try {
+      setCatalog(await api.agyAccountModels(id));
+    } catch (err) {
+      setCatalog(null);
+      setCatalogError(
+        err instanceof ApiError
+          ? err.message
+          : "دریافت کاتالوگ مدل‌ها از agy ناموفق بود.",
+      );
+    } finally {
+      setCatalogLoading(false);
+    }
+  }, []);
+
+  const onProviderChange = (value: string) => {
+    setProviderId(value);
+    // One provider exists, but the flow is wired for real: picking it reloads
+    // the account list, which is what the next step depends on.
+    setAccountId("");
+    setModelId("");
+    setEffort("");
+    setCatalog(null);
+    setCatalogError(null);
+    void loadAccounts();
+  };
+
+  const onAccountChange = (value: string) => {
+    setAccountId(value);
+    setModelId("");
+    setEffort("");
+    void loadCatalog(value);
+  };
+
+  const onModelChange = (value: string) => {
+    setModelId(value);
+    setEffort("");
+    // Sensible defaults the owner can still adjust: everything answers text,
+    // and a thinking-capable model is a reasoning model.
+    setCaps(
+      effortsForModel(value).length > 0
+        ? ["text", "reasoning"]
+        : ["text"],
+    );
+  };
+
   const toggleCap = (cap: Capability) => {
     setCaps((current) =>
       current.includes(cap) ? current.filter((c) => c !== cap) : [...current, cap],
@@ -65,8 +154,8 @@ export default function AdminModelsPage() {
   const onCreate = async (event: React.FormEvent) => {
     event.preventDefault();
     setFormError(null);
-    if (!providerId) {
+    if (!modelId) {
-      setFormError("ابتدا یک ارائه‌دهنده بسازید.");
+      setFormError("ابتدا یک مدل از کاتالوگ اکانت انتخاب کنید.");
       return;
     }
     if (caps.length === 0) {
@@ -76,16 +165,15 @@ export default function AdminModelsPage() {
     setCreating(true);
     try {
       const created = await api.adminCreateModel({
-        provider_id: providerId,
+        internal_name: modelId,
-        internal_name: internalName.trim(),
-        display_name_for_admin: displayName.trim() || internalName.trim(),
         capabilities: caps,
         input_price_per_1m: Number(inputPrice) || 0,
         output_price_per_1m: Number(outputPrice) || 0,
+        reasoning_effort: effort || undefined,
       });
       setModels((current) => [created, ...current]);
-      setInternalName("");
+      setModelId("");
-      setDisplayName("");
+      setEffort("");
       setInputPrice("");
       setOutputPrice("");
       setCaps(["text"]);
@@ -96,53 +184,93 @@ export default function AdminModelsPage() {
     }
   };
 
-  const providerName = (id: string) => providers.find((p) => p.id === id)?.name ?? id;
+  const efforts: ReasoningEffort[] = modelId ? effortsForModel(modelId) : [];
+  const accountOptions = accounts.map((a) => ({
+    value: a.id,
+    label: a.email || a.name || a.id,
+  }));
+  const modelOptions = (catalog ?? []).map((id) => ({ value: id, label: id }));
+  const effortOptions = efforts.map((e) => ({ value: e, label: effortLabel(e) }));
 
   return (
     <div className="anim-stagger flex flex-col gap-5">
       <header className="flex flex-col gap-1">
         <h1 className="text-lg font-medium text-slate-100">مدل‌ها</h1>
         <p className="text-sm text-slate-400">
-          نام داخلی مدل‌ها فقط در این کنسول دیده می‌شود. مشتری هرگز نام مدل یا ارائه‌دهنده را
+          agy تنها ارائه‌دهندهٔ سامانه است. برای افزودن مدل، اکانت را انتخاب کنید؛
-          نمی‌بیند.
+          کاتالوگ مدل‌ها لحظه‌ای و مستقیم از همان اکانت خوانده می‌شود. نام مدل فقط
+          در این کنسول دیده می‌شود و مشتری هرگز آن را نمی‌بیند.
         </p>
       </header>
 
       <Card>
         <CardTitle>افزودن مدل</CardTitle>
         <CardHint>
-          قیمت‌ها به «واحد پول به ازای هر یک‌میلیون توکن» وارد می‌شوند و مبنای محاسبهٔ هزینهٔ
+          قیمت‌ها به «واحد پول به ازای هر یک‌میلیون توکن» وارد می‌شوند و مبنای
-          داخلی هستند.
+          محاسبهٔ هزینهٔ داخلی هستند.
         </CardHint>
         <form onSubmit={onCreate} className="mt-4 flex flex-col gap-4">
-          <div className="grid gap-4 sm:grid-cols-3">
+          <div className="grid gap-4 sm:grid-cols-2">
             <Select
               label="ارائه‌دهنده"
               required
               value={providerId}
-              onChange={(e) => setProviderId(e.target.value)}
+              onChange={(e) => onProviderChange(e.target.value)}
-              options={
+              options={PROVIDER_OPTIONS}
-                providers.length === 0
-                  ? [{ value: "", label: "ارائه‌دهنده‌ای وجود ندارد" }]
-                  : providers.map((p) => ({ value: p.id, label: p.name }))
-              }
             />
-            <Field
+            <Select
-              label="نام داخلی"
+              label="اکانت agy"
               required
-              value={internalName}
+              value={accountId}
-              onChange={(e) => setInternalName(e.target.value)}
+              onChange={(e) => onAccountChange(e.target.value)}
-              placeholder="gpt-4o-mini"
+              options={
-              className="ltr-inline"
+                accountOptions.length === 0
-            />
+                  ? [{ value: "", label: "اکانتی ثبت نشده — ابتدا اکانت agy بسازید" }]
-            <Field
+                  : accountOptions
-              label="نام نمایشی برای مالک"
+              }
-              value={displayName}
+              hint="کاتالوگ مدل‌ها از همین اکانت خوانده می‌شود."
-              onChange={(e) => setDisplayName(e.target.value)}
-              placeholder="مدل سریع و ارزان"
             />
           </div>
 
+          <div className="grid gap-4 sm:grid-cols-2">
+            <Select
+              label="مدل"
+              required
+              value={modelId}
+              onChange={(e) => onModelChange(e.target.value)}
+              options={
+                !accountId
+                  ? [{ value: "", label: "ابتدا یک اکانت انتخاب کنید" }]
+                  : modelOptions.length === 0
+                    ? [{ value: "", label: "کاتالوگی برای این اکانت نیست" }]
+                    : modelOptions
+              }
+              disabled={!accountId || catalogLoading || modelOptions.length === 0}
+            />
+            <Select
+              label="افورت استدلال"
+              value={effort}
+              onChange={(e) => setEffort(e.target.value)}
+              options={effortOptions}
+              disabled={efforts.length === 0}
+              hint={
+                efforts.length === 0
+                  ? "این مدل افورت استدلال ندارد."
+                  : "بودجهٔ تفکر مدل؛ روی مدل‌های غیراستدلالی نادیده گرفته می‌شود."
+              }
+            />
+          </div>
+
+          {(accountId && catalogLoading) || (accountId && catalog === null && !catalogError) ? (
+            <LoadingState label="در حال دریافت کاتالوگ مدل‌ها از اکانت…" />
+          ) : catalogError ? (
+            <ErrorState
+              title="دریافت کاتالوگ ناموفق"
+              message={catalogError}
+              onRetry={() => void loadCatalog(accountId)}
+            />
+          ) : null}
+
           <div className="grid gap-4 sm:grid-cols-2">
             <Field
               label="قیمت ورودی (به ازای ۱M توکن)"
@@ -178,7 +306,7 @@ export default function AdminModelsPage() {
           </div>
 
           <div className="flex justify-end">
-            <Button type="submit" loading={creating}>
+            <Button type="submit" loading={creating} disabled={!modelId}>
               افزودن مدل
             </Button>
           </div>
@@ -207,7 +335,7 @@ export default function AdminModelsPage() {
               <thead>
                 <tr className="border-b border-surface-border text-xs text-slate-400">
                   <th className="px-2 py-2 text-right font-medium">نام داخلی</th>
-                  <th className="px-2 py-2 text-right font-medium">ارائه‌دهنده</th>
+                  <th className="px-2 py-2 text-right font-medium">افورت</th>
                   <th className="px-2 py-2 text-right font-medium">توانایی‌ها</th>
                   <th className="px-2 py-2 text-right font-medium">ورودی</th>
                   <th className="px-2 py-2 text-right font-medium">خروجی</th>
@@ -219,10 +347,16 @@ export default function AdminModelsPage() {
                   <tr key={model.id} className="border-b border-surface-border/60">
                     <td className="px-2 py-2">
                       <span className="ltr-inline text-slate-200">{model.internal_name}</span>
-                      <p className="text-xs text-slate-500">{model.display_name_for_admin}</p>
+                      <p className="text-xs text-slate-500">
+                        {model.display_name_for_admin}
+                      </p>
                     </td>
-                    <td className="ltr-inline px-2 py-2 text-xs text-slate-400">
+                    <td className="px-2 py-2">
-                      {providerName(model.provider_id)}
+                      {model.reasoning_effort ? (
+                        <Badge tone="info">{effortLabel(model.reasoning_effort)}</Badge>
+                      ) : (
+                        <span className="text-xs text-slate-500">پیش‌فرض</span>
+                      )}
                     </td>
                     <td className="px-2 py-2">
                       <div className="flex flex-wrap gap-1">

apps/web/src/app/(app)/admin/page.tsx

@@ -52,7 +52,7 @@ export default function AdminOverviewPage() {
       <header className="flex flex-col gap-1">
         <h1 className="text-lg font-medium text-slate-100">کنسول مالک سامانه</h1>
         <p className="text-sm text-slate-400">
-          مدیریت ارائه‌دهنده‌ها، مدل‌ها، سیاست مسیریابی و تعرفه‌ها.
+          مدیریت اکانت‌های agy، مدل‌ها، سیاست مسیریابی و تعرفه‌ها.
         </p>
       </header>
 
@@ -69,8 +69,8 @@ export default function AdminOverviewPage() {
       <Card>
         <CardTitle>آداپتورهای فعال</CardTitle>
         <CardHint>
-          این مقادیر از محیط خوانده می‌شوند. هر مقدار «mock» یعنی آن سرویس بیرونی هنوز
+          وضعیت سرویس‌های بیرونی. هر مقدار «mock» یعنی آن سرویس هنوز متصل نشده
-          متصل نشده است.
+          است؛ مدل‌ها فقط از agy (Google Antigravity) سرویس می‌شوند.
         </CardHint>
         <div className="mt-3 flex flex-wrap gap-2">
           {Object.entries(readiness?.checks.adapters ?? {}).map(([key, value]) => (
@@ -84,13 +84,13 @@ export default function AdminOverviewPage() {
       <div className="grid gap-4 lg:grid-cols-2">
         {[
           {
-            title: "ارائه‌دهنده‌ها و کلیدها",
+            title: "اکانت‌های agy",
-            hint: "ساخت ارائه‌دهنده و ذخیرهٔ کلید رمزنگاری‌شده. کلید خام هرگز نمایش داده نمی‌شود.",
+            hint: "ورود با Google و پول چرخشی اکانت‌ها. agy تنها ارائه‌دهندهٔ سامانه است.",
-            path: "/admin/providers",
+            path: "/admin/agy-accounts",
           },
           {
             title: "مدل‌ها",
-            hint: "نام داخلی، توانایی‌ها و قیمت ورودی/خروجی — فقط برای مالک سامانه.",
+            hint: "انتخاب مدل از کاتالوگ زندهٔ اکانت، افورت استدلال و قیمت — فقط برای مالک سامانه.",
             path: "/admin/models",
           },
           {

apps/web/src/app/(app)/admin/providers/page.tsx

@@ -1,269 +0,0 @@
-"use client";
-
-import * as React from "react";
-import type { AdminCredential, AdminProvider, ProviderType } from "@agy/contracts";
-import {
-  Badge,
-  Button,
-  Card,
-  CardHint,
-  CardTitle,
-  EmptyState,
-  ErrorState,
-  Field,
-  LoadingState,
-  Select,
-} from "@agy/ui";
-import { ApiError, api } from "@/lib/api";
-
-const PROVIDER_TYPES: { value: ProviderType; label: string }[] = [
-  { value: "mock", label: "آزمایشی (mock)" },
-  { value: "openai_compatible", label: "سازگار با OpenAI" },
-  { value: "anthropic", label: "Anthropic" },
-  { value: "gemini", label: "Gemini" },
-  { value: "openrouter", label: "OpenRouter" },
-  { value: "agy", label: "agy (Google Antigravity)" },
-];
-
-export default function AdminProvidersPage() {
-  const [providers, setProviders] = React.useState<AdminProvider[]>([]);
-  const [loading, setLoading] = React.useState(true);
-  const [error, setError] = React.useState<string | null>(null);
-
-  const [name, setName] = React.useState("");
-  const [type, setType] = React.useState<ProviderType>("openai_compatible");
-  const [baseUrl, setBaseUrl] = React.useState("");
-  const [creating, setCreating] = React.useState(false);
-  const [formError, setFormError] = React.useState<string | null>(null);
-
-  const load = React.useCallback(async () => {
-    setLoading(true);
-    setError(null);
-    try {
-      setProviders(await api.adminProviders());
-    } catch (err) {
-      setError(err instanceof ApiError ? err.message : "دریافت ارائه‌دهنده‌ها ناموفق بود.");
-    } finally {
-      setLoading(false);
-    }
-  }, []);
-
-  React.useEffect(() => {
-    void load();
-  }, [load]);
-
-  const onCreate = async (event: React.FormEvent) => {
-    event.preventDefault();
-    setFormError(null);
-    setCreating(true);
-    try {
-      const created = await api.adminCreateProvider({
-        name: name.trim(),
-        type,
-        base_url: baseUrl.trim(),
-      });
-      setProviders((current) => [created, ...current]);
-      setName("");
-      setBaseUrl("");
-    } catch (err) {
-      setFormError(err instanceof ApiError ? err.message : "ساخت ارائه‌دهنده ناموفق بود.");
-    } finally {
-      setCreating(false);
-    }
-  };
-
-  return (
-    <div className="anim-stagger flex flex-col gap-5">
-      <header className="flex flex-col gap-1">
-        <h1 className="text-lg font-medium text-slate-100">ارائه‌دهنده‌ها</h1>
-        <p className="text-sm text-slate-400">
-          هر ارائه‌دهنده می‌تواند چند کلید داشته باشد. کلیدها رمزنگاری‌شده ذخیره می‌شوند و فقط
-          به‌صورت ماسک‌شده برگردانده می‌شوند.
-        </p>
-      </header>
-
-      <Card>
-        <CardTitle>افزودن ارائه‌دهنده</CardTitle>
-        <form onSubmit={onCreate} className="mt-4 flex flex-col gap-4">
-          <div className="grid gap-4 sm:grid-cols-3">
-            <Field
-              label="نام"
-              required
-              minLength={2}
-              value={name}
-              onChange={(e) => setName(e.target.value)}
-              placeholder="مثلاً openai-main"
-              className="ltr-inline"
-            />
-            <Select
-              label="نوع"
-              value={type}
-              onChange={(e) => setType(e.target.value as ProviderType)}
-              options={PROVIDER_TYPES}
-            />
-            <Field
-              label="آدرس پایه"
-              value={baseUrl}
-              onChange={(e) => setBaseUrl(e.target.value)}
-              placeholder="https://api.example.com/v1"
-              className="ltr-inline"
-            />
-          </div>
-          <div className="flex justify-end">
-            <Button type="submit" loading={creating}>
-              افزودن
-            </Button>
-          </div>
-        </form>
-        {formError && (
-          <div className="mt-3">
-            <ErrorState title="ساخت ارائه‌دهنده ناموفق" message={formError} />
-          </div>
-        )}
-      </Card>
-
-      {loading ? (
-        <LoadingState label="در حال دریافت ارائه‌دهنده‌ها…" />
-      ) : error ? (
-        <ErrorState message={error} onRetry={load} />
-      ) : providers.length === 0 ? (
-        <EmptyState
-          title="ارائه‌دهنده‌ای ثبت نشده است"
-          description="با فرم بالا اولین ارائه‌دهنده را اضافه کنید."
-        />
-      ) : (
-        <div className="flex flex-col gap-3">
-          {providers.map((provider) => (
-            <ProviderCard key={provider.id} provider={provider} />
-          ))}
-        </div>
-      )}
-    </div>
-  );
-}
-
-function ProviderCard({ provider }: { provider: AdminProvider }) {
-  const [open, setOpen] = React.useState(false);
-  const [credentials, setCredentials] = React.useState<AdminCredential[]>([]);
-  const [loading, setLoading] = React.useState(false);
-  const [error, setError] = React.useState<string | null>(null);
-
-  const [credName, setCredName] = React.useState("");
-  const [apiKey, setApiKey] = React.useState("");
-  const [saving, setSaving] = React.useState(false);
-  const [formError, setFormError] = React.useState<string | null>(null);
-
-  const loadCredentials = React.useCallback(async () => {
-    setLoading(true);
-    setError(null);
-    try {
-      setCredentials(await api.adminCredentials(provider.id));
-    } catch (err) {
-      setError(err instanceof ApiError ? err.message : "دریافت کلیدها ناموفق بود.");
-    } finally {
-      setLoading(false);
-    }
-  }, [provider.id]);
-
-  React.useEffect(() => {
-    if (open) void loadCredentials();
-  }, [open, loadCredentials]);
-
-  const onAddCredential = async (event: React.FormEvent) => {
-    event.preventDefault();
-    setFormError(null);
-    setSaving(true);
-    try {
-      const created = await api.adminCreateCredential(provider.id, {
-        name: credName.trim(),
-        api_key: apiKey.trim(),
-      });
-      // The response carries the masked form only; the plaintext is gone.
-      setCredentials((current) => [...current, created]);
-      setCredName("");
-      setApiKey("");
-    } catch (err) {
-      setFormError(err instanceof ApiError ? err.message : "ذخیرهٔ کلید ناموفق بود.");
-    } finally {
-      setSaving(false);
-    }
-  };
-
-  return (
-    <Card>
-      <div className="flex flex-wrap items-center justify-between gap-3">
-        <div className="flex flex-col gap-1">
-          <div className="flex items-center gap-2">
-            <p className="ltr-inline text-sm font-medium text-slate-100">{provider.name}</p>
-            <Badge tone={provider.is_active ? "success" : "neutral"}>
-              {provider.is_active ? "فعال" : "غیرفعال"}
-            </Badge>
-          </div>
-          <p className="ltr-inline text-xs text-slate-500">
-            {provider.type}
-            {provider.base_url ? ` · ${provider.base_url}` : ""}
-          </p>
-        </div>
-        <Button variant="secondary" size="sm" onClick={() => setOpen((v) => !v)}>
-          {open ? "بستن کلیدها" : "مدیریت کلیدها"}
-        </Button>
-      </div>
-
-      {open && (
-        <div className="mt-4 flex flex-col gap-4 border-t border-surface-border pt-4">
-          {loading ? (
-            <LoadingState label="در حال دریافت کلیدها…" />
-          ) : error ? (
-            <ErrorState message={error} onRetry={loadCredentials} />
-          ) : credentials.length === 0 ? (
-            <p className="text-xs text-slate-500">کلیدی ثبت نشده است.</p>
-          ) : (
-            <ul className="flex flex-col gap-2">
-              {credentials.map((credential) => (
-                <li
-                  key={credential.id}
-                  className="flex items-center justify-between gap-3 rounded-lg border border-surface-border bg-surface-overlay px-3 py-2"
-                >
-                  <span className="text-xs text-slate-300">{credential.name}</span>
-                  <span className="ltr-inline font-mono text-xs text-slate-500">
-                    {credential.masked_key}
-                  </span>
-                </li>
-              ))}
-            </ul>
-          )}
-
-          <form onSubmit={onAddCredential} className="flex flex-col gap-3 sm:flex-row sm:items-end">
-            <div className="flex-1">
-              <Field
-                label="نام کلید"
-                required
-                value={credName}
-                onChange={(e) => setCredName(e.target.value)}
-                placeholder="production"
-                className="ltr-inline"
-              />
-            </div>
-            <div className="flex-1">
-              <Field
-                label="کلید API"
-                required
-                type="password"
-                minLength={4}
-                autoComplete="off"
-                value={apiKey}
-                onChange={(e) => setApiKey(e.target.value)}
-                hint="پس از ذخیره دیگر قابل بازیابی نیست."
-                className="ltr-inline"
-              />
-            </div>
-            <Button type="submit" loading={saving}>
-              ذخیره
-            </Button>
-          </form>
-          {formError && <ErrorState title="ذخیرهٔ کلید ناموفق" message={formError} />}
-        </div>
-      )}
-    </Card>
-  );
-}

apps/web/src/lib/api.ts

@@ -418,26 +418,6 @@ export const api = {
     ),
 
   // --- platform owner -------------------------------------------------------
-  adminProviders: () =>
-    apiRequest<import("@agy/contracts").AdminProvider[]>("/api/admin/providers"),
-  adminCreateProvider: (body: {
-    name: string;
-    type: import("@agy/contracts").ProviderType;
-    base_url?: string;
-  }) =>
-    apiRequest<import("@agy/contracts").AdminProvider>("/api/admin/providers", {
-      method: "POST",
-      body,
-    }),
-  adminCredentials: (providerId: string) =>
-    apiRequest<import("@agy/contracts").AdminCredential[]>(
-      `/api/admin/providers/${providerId}/credentials`,
-    ),
-  adminCreateCredential: (providerId: string, body: { name: string; api_key: string }) =>
-    apiRequest<import("@agy/contracts").AdminCredential>(
-      `/api/admin/providers/${providerId}/credentials`,
-      { method: "POST", body },
-    ),
   // agy accounts: OAuth login instead of API keys.
   agyAccounts: () =>
     apiRequest<import("@agy/contracts").AgyAccount[]>("/api/admin/agy/accounts"),
@@ -466,16 +446,18 @@ export const api = {
       input_tokens: number;
       output_tokens: number;
     }>(`/api/admin/agy/accounts/${accountId}/test`, { method: "POST" }),
-  agyModels: () =>
+  // Live model catalog of one account — fetched from Google on every call, so
-    apiRequest<string[]>("/api/admin/agy/models"),
+  // the model dropdown always matches that subscription's entitlement.
+  agyAccountModels: (accountId: string) =>
+    apiRequest<string[]>(`/api/admin/agy/accounts/${accountId}/models`),
   adminModels: () => apiRequest<import("@agy/contracts").AdminModel[]>("/api/admin/models"),
   adminCreateModel: (body: {
-    provider_id: string;
     internal_name: string;
-    display_name_for_admin: string;
+    display_name_for_admin?: string;
     capabilities: import("@agy/contracts").Capability[];
     input_price_per_1m?: number;
     output_price_per_1m?: number;
+    reasoning_effort?: string;
   }) =>
     apiRequest<import("@agy/contracts").AdminModel>("/api/admin/models", {
       method: "POST",

apps/web/src/lib/nav.ts

@@ -27,10 +27,12 @@ export const NAV_ITEMS: NavItem[] = [
 /**
  * Platform-owner-only destinations. `admin/layout.tsx` mirrors this list and
  * renders it as a tab strip; the API is the real security boundary either way.
+ * agy is the only model provider, so there is no providers section anymore —
+ * accounts live at `/admin/agy-accounts`.
  */
 export const ADMIN_ITEMS: NavItem[] = [
   { href: "/admin", label: "نمای کلی کنسول", hint: "وضعیت سامانه" },
-  { href: "/admin/providers", label: "ارائه‌دهنده‌ها", hint: "مدل‌ها و کلیدها" },
+  { href: "/admin/agy-accounts", label: "اکانت‌های agy", hint: "ورود و پول چرخشی" },
   { href: "/admin/models", label: "مدل‌ها", hint: "قابلیت و قیمت" },
   { href: "/admin/routing", label: "مسیریابی", hint: "fast، cheap، reasoning" },
   { href: "/admin/rate-cards", label: "تعرفه‌ها", hint: "قیمت واحدها" },

apps/web/src/test/api.test.ts

@@ -167,4 +167,24 @@ describe("agy account client", () => {
     expect(calls[0].method).toBe("DELETE");
     expect(calls[0].url).toContain("/api/admin/agy/accounts/a1");
   });
+
+  it("fetches the live model catalog for one account", async () => {
+    const calls = stubJson(["gemini-3-flash", "gemini-3-pro-high"]);
+    const models = await api.agyAccountModels("a1");
+    expect(calls[0].method).toBe("GET");
+    expect(calls[0].url).toContain("/api/admin/agy/accounts/a1/models");
+    expect(models).toContain("gemini-3-pro-high");
+  });
+
+  it("creates a model from the cascade — no manual names", async () => {
+    const calls = stubJson({ id: "m1", internal_name: "gemini-3-pro-high" });
+    await api.adminCreateModel({
+      internal_name: "gemini-3-pro-high",
+      capabilities: ["text", "reasoning"],
+      reasoning_effort: "high",
+    });
+    expect(calls[0].url).toContain("/api/admin/models");
+    expect(calls[0].body).toContain("gemini-3-pro-high");
+    expect(calls[0].body).toContain("reasoning_effort");
+  });
 });

packages/contracts/src/index.ts

@@ -10,14 +10,52 @@ export type Role = "owner" | "admin" | "manager" | "member" | "viewer";
 
 export type MembershipStatus = "active" | "invited" | "suspended";
 
-/** Provider kinds the AI Control Plane can talk to. */
+/**
-export type ProviderType =
+ * Provider kinds the AI Control Plane can talk to. agy (Google Antigravity)
-  | "mock"
+ * is the only one — the platform no longer has a provider picker.
-  | "openai_compatible"
+ */
-  | "anthropic"
+export type ProviderType = "agy";
-  | "gemini"
+
-  | "openrouter"
+/** agy reasoning efforts, cheapest first. Mirrors the backend's list. */
-  | "agy";
+export type ReasoningEffort = "minimal" | "low" | "medium" | "high" | "xhigh";
+
+export const REASONING_EFFORTS: ReasoningEffort[] = [
+  "minimal",
+  "low",
+  "medium",
+  "high",
+  "xhigh",
+];
+
+const EFFORT_LABELS: Record<ReasoningEffort, string> = {
+  minimal: "حداقل",
+  low: "کم",
+  medium: "متوسط",
+  high: "زیاد",
+  xhigh: "بسیار زیاد",
+};
+
+/**
+ * Reasoning efforts a model may be configured with, or an empty list when the
+ * model cannot think — the backend rejects a thinking config everywhere else.
+ * Same marker rule as the agy adapter: thinking-capable ids contain one of
+ * `thinking` / `gemini-3` / `gemini-2.5-pro` / `gemini-2.5-flash`, image
+ * models are always excluded.
+ */
+export function effortsForModel(modelId: string): ReasoningEffort[] {
+  const lower = modelId.toLowerCase();
+  if (lower.includes("image")) return [];
+  const canThink =
+    lower.includes("thinking") ||
+    lower.includes("gemini-3") ||
+    lower.includes("gemini-2.5-pro") ||
+    lower.includes("gemini-2.5-flash");
+  return canThink ? [...REASONING_EFFORTS] : [];
+}
+
+export function effortLabel(effort: string): string {
+  return EFFORT_LABELS[effort as ReasoningEffort] ?? effort;
+}
 
 /** Rotation-pool lifecycle of one agy account. */
 export type AgyAccountStatus = "active" | "cooldown" | "error";
@@ -182,15 +220,6 @@ export interface Repository {
   private: boolean;
 }
 
-export interface AdminProvider {
-  id: string;
-  name: string;
-  type: string;
-  base_url: string;
-  is_active: boolean;
-  created_at: string;
-}
-
 export interface AdminModel {
   id: string;
   provider_id: string;
@@ -199,6 +228,8 @@ export interface AdminModel {
   capabilities: string[];
   input_price_per_1m: number;
   output_price_per_1m: number;
+  /** agy reasoning effort ("" = provider default). Owner-only metadata. */
+  reasoning_effort: string;
   is_active: boolean;
 }
 
@@ -228,21 +259,6 @@ export interface AdminOrganization {
   created_at: string;
 }
 
-/**
- * A stored provider credential.
- *
- * There is no field here that can carry a plaintext key — the API only ever
- * returns a masked form, and on list it returns a fixed mask because the
- * plaintext is unrecoverable by design.
- */
-export interface AdminCredential {
-  id: string;
-  provider_id: string;
-  name: string;
-  masked_key: string;
-  is_active: boolean;
-}
-
 /** Result of a routing dry-run: which model a mode would pick, and why. */
 export interface ModelChoice {
   mode: string;