# NousResearch/hermes-agent · Per-client and per-request model routing support

## Open now

- 0 issues, 1 PR open on GitHub.
- PR #10773: feat(api-server): honor request-level model field for per-request model selection

These issues and pull requests concern adding model routing capabilities to the API server, allowing model selection to be configured per-client via model_routes config and per-request via the request-level model field.

## TL;DR

- 4 member(s): 0 open issue(s), 1 closed issue(s), 1 open PR(s), 1 merged PR(s), 1 PR(s) closed without merging.
- A fix was merged to main in PR #57028. Verify it addresses your case.
- **Recommended:** read PR #57028 (and any competing PRs below) before filing. Confirm the merged change covers your symptom.

## Maintainer resolution verdict

- Internal-tracker verdict over this cluster's issue members: `unresolved`. This is distinct from the fix status above. Weigh it yourself.

## How this cluster was grouped

- Link evidence (`STRUCTURAL_SPANNING`): every member is connected by structural evidence: shared changed files, or a closing/superseded reference.
- Pairwise link density `0.50`: the share of member pairs that carry a link, counting a structural link in full and a relevance-only link at 0.36. `1.00` means every pair is linked.

## Closed / duplicate issues

- #3155: \[Feature\]: Support per-client model routing in the API server

## Prior & competing PRs

- **#57028: feat(api-server): per-client model routing via model_routes (#3176 salvage)**. merged to main · fix tier `verified`
  - maintainer direction: category `merged`
  - files: `cli-config.yaml.example`, `gateway/platforms/api_server.py`, `scripts/release.py`, `tests/gateway/test_api_server.py`
  - fix hunks:
  - `cli-config.yaml.example`

    ```diff
    @@ -571,6 +571,41 @@ max_concurrent_sessions: null
     # explicitly want one shared "room brain" per group/channel.
     group_sessions_per_user: true
     
    +# ─────────────────────────────────────────────────────────────────────────────
    +# API Server — per-client model routing
    +# ─────────────────────────────────────────────────────────────────────────────
    +# Route different API clients to different models/providers on a single
    +# Hermes deployment.  Clients choose a backend by sending a specific string
    +# as the OpenAI ``model`` field.  Unmapped model values fall back to the
    +# global model configured in the ``model:`` section above, and an explicit
    +# session /model override always wins over a route.
    +#
    +# Configure via the ``platforms.api_server.extra.model_routes`` gateway
    +# config block:
    +#
    +#   platforms:
    +#     api_server:
    +#       enabled: true
    +#       extra:
    +#         key: "your-api-server-secret"
    +#         model_routes:
    +#           # Xiaozhi clients send model="minimax-m2" → routed to MiniMax via OpenRouter
    +#           minimax-m2:
    +#             model: "minimax/minimax-m1"
    +#             provider: "openrouter"          # optional — overrides global provider
    +#             # api_key: "[redacted:secret-assignment]"             # optional — per-route UPSTREAM provider
    +#             #                                 key (NOT caller auth; never logged)
    +#             # base_url: "https://..."       # optional — per-route base URL
    +#           # GPT clients keep their own alias
    +#           gpt-5:
    +#             model: "openai/gpt-5"
    +#             provider: "openrouter"
    +#
    +# Configured aliases are automatically listed by GET /v1/models so clients
    +# can discover them without manual coordination.  Caller authentication is
    +# unchanged: every request still authenticates with the global API server
    +# key (``extra.key`` / API_SERVER_KEY).
    +
     # ─────────────────────────────────────────────────────────────────────────────
     # Gateway Streaming
     # ─────────────────────────────────────────────────────────────────────────────
    ```
  - `gateway/platforms/api_server.py`

    ```diff
    @@ -6,7 +6,7 @@
     - POST /v1/responses               — OpenAI Responses API format (stateful via previous_response_id; X-Hermes-Session-Key supported)
     - GET  /v1/responses/{response_id} — Retrieve a stored response
     - DELETE /v1/responses/{response_id} — Delete a stored response
    -- GET  /v1/models                  — lists hermes-agent as an available model
    +- GET  /v1/models                  — lists hermes-agent and any configured model_routes aliases
     - GET  /v1/capabilities            — machine-readable API capabilities for external UIs
     - GET  /api/sessions               — list client-visible Hermes sessions
     - POST /api/sessions               — create an empty Hermes session
    ```

    ```diff
    @@ -832,6 +832,22 @@ def __init__(self, config: PlatformConfig):
             self._model_name: str = self._resolve_model_name(
                 extra.get("model_name", os.getenv("API_SERVER_MODEL_NAME", "")),
             )
    +        # model_routes: maps incoming ``model`` field values to specific
    +        # provider/model configs so one API server instance can serve
    +        # multiple clients on different backends.
    +        #
    +        # Config format (platforms.api_server.extra in the gateway config):
    +        #   model_routes:
    +        #     minimax-m2:          # alias the client sends as the "model" field
    +        #       model: "minimax/minimax-m1"
    +        #       provider: "openrouter"   # optional — resolved via the provider
    +        #                                # credential chain when set
    +        #       api_key: "sk-…"          # optional — per-route UPSTREAM provider
    +        #                                # key override (NOT caller auth; never logged)
    +        #       base_url: "https://…"    # optional — per-route base URL override
    +        self._model_routes: Dict[str, Dict[str, Any]] = self._parse_model_routes(
    +            extra.get("model_routes"),
    +        )
             self._app: Optional["web.Application"] = None
             self._runner: Optional["web.AppRunner"] = None
             self._site: Optional["web.TCPSite"] = None
    ```

    ```diff
    @@ -1118,6 +1134,78 @@ def _ensure_session_db(self):
         # Agent creation helper
         # ------------------------------------------------------------------
     
    +    @staticmethod
    +    def _parse_model_routes(raw: Any) -> Dict[str, Dict[str, Any]]:
    +        """Validate and normalize the ``model_routes`` config block.
    +
    +        Accepts a mapping of ``alias -> {model, provider?, api_key?, base_url?}``.
    +        Invalid shapes are dropped (never raised) so a config typo can't take
    +        the whole API server down.  Route values are coerced to strings.
    +
    +        Security: per-route ``api_key`` values are UPSTREAM provider
    +        credentials (used to call the routed model's backend), not caller
    +        authentication — callers still authenticate with the global
    +        API_SERVER_KEY bearer token via ``_check_auth``.  Route api_keys must
    +        never be logged; only alias names and non-secret fields may appear in
    +        logs.
    +        """
    +        if not isinstance(raw, dict):
    +            if raw:
    +                logger.warning(
    +                    "api_server model_routes ignored: expected a mapping, got %s",
    +                    type(raw).__name__,
    +                )
    +            return {}
    +
    +        allowed_keys = ("model", "provider", "api_key", "base_url")
    +        routes: Dict[str, Dict[str, Any]] = {}
    +        for alias, cfg in raw.items():
    +            alias_str = str(alias).strip()
    +            if not alias_str or not isinstance(cfg, dict):
    +                logger.warning(
    +                    "api_server model_routes: dropping invalid route entry %r", alias_str or alias
    +                )
    +                continue
    +            route = {
    +                key: str(cfg[key]).strip()
    +                for key in allowed_keys
    +                if cfg.get(key) is not None and str(cfg[key]).strip()
    +            }
    +            if not route.get("model"):
    +                logger.warning(
    +                    "api_server model_routes: route %r has no 'model'; dropping", alias_str
    +                )
    +                continue
    +            routes[alias_str] = route
    +        return routes
    +
    +    def _resolve_route(self, model_alias: Any) -> Optional[Dict[str, Any]]:
    +        """Return the model_routes entry for *model_alias*, or None."""
    +        if not self._model_routes or not isinstance(model_alias, str):
    +            return None
    +        return self._model_routes.get(model_alias)
    +
    +    def _session_model_override_for(self, session_key: Optional[str]) -> Optional[Dict[str, Any]]:
    +        """Return the gateway's session ``/model`` override for *session_key*, if any.
    +
    +        The gateway tracks per-session ``/model`` switches in
    +        ``GatewayRunner._session_model_overrides``.  API-server requests that
    +        share such a session key must keep honouring the explicit session
    +        override even when the request's ``model`` field matches a configured
    +        route — a user-issued ``/model`` always wins over static config.
    +        """
    +        if not session_key:
    +            return None
    +        try:
    +            from gateway.run import _gateway_runner_ref
    +            runner = _gateway_runner_ref()
    +            if runner is None:
    +                return None
    +            override = runner._session_model_overrides.get(session_key)
    +            return dict(override) if isinstance(override, dict) else None
    +        except Exception:
    +            return None
    +
         def _create_agent(
             self,
             ephemeral_system_prompt: Optional[str] = None,
    ```

    ```diff
    @@ -1127,6 +1215,7 @@ def _create_agent(
             tool_start_callback=None,
             tool_complete_callback=None,
             gateway_session_key: Optional[str] = None,
    +        route: Optional[Dict[str, Any]] = None,
         ) -> Any:
             """
             Create an AIAgent instance using the gateway's runtime config.
    ```

    ```diff
    @@ -1142,6 +1231,11 @@ def _create_agent(
             key is meant to persist across transcripts so long-term memory
             providers (e.g. Honcho) can scope their per-chat state correctly
             — matching the semantics of the native gateway's ``session_key``.
    +
    +        ``route`` is an optional ``model_routes`` entry (per-client model
    +        routing).  When set — and no session ``/model`` override exists for
    +        this session — its model/provider/api_key/base_url override the
    +        global defaults for this agent instance only.
             """
             from run_agent import AIAgent
             from gateway.run import (
    ```

    ```diff
    @@ -1169,6 +1263,51 @@ def _create_agent(
             if runtime_model:
                 model = runtime_model
     
    +        # Per-client model routing (model_routes config).  The route was
    +        # resolved from the request's ``model`` field by the HTTP handler.
    +        # Precedence (highest first): session ``/model`` override → model_routes
    +        # route → global config — an explicit user-issued ``/model`` on the
    +        # session always beats static per-client route config.
    +        session_override = self._session_model_override_for(
    +            gateway_session_key or session_id
    +        )
    +        if route and not session_override:
    +            if route.get("provider"):
    +                # Resolve real credentials for the routed provider (mirrors
    +                # the channel_overrides path in gateway/run.py) so a route
    +                # without an explicit api_key/base_url still gets the right
    +                # provider auth instead of the default provider's key.
    +                try:
    +                    from gateway.run import _resolve_runtime_agent_kwargs_for_provider
    +                    provider_kwargs = _resolve_runtime_agent_kwargs_for_provider(
    +                        route["provider"]
    +                    )
    +                    provider_kwargs.pop("model", None)
    +                    runtime_kwargs.update(provider_kwargs)
    +                except Exception:
    +                    # Fall back to just switching the provider name; explicit
    +                    # per-route api_key/base_url below can still complete auth.
    +                    runtime_kwargs["provider"] = route["provider"]
    +            if route.get("model"):
    +                model = route["model"]
    +            # Per-route secrets are upstream provider credentials. Never log
    +            # them (compare _check_auth: caller auth stays the global bearer
    +            # key checked with hmac.compare_digest).
    +            if route.get("api_key"):
    +                runtime_kwargs["api_key"] = route["api_key"]
    +            if route.get("base_url"):
    +                runtime_kwargs["base_url"] = route["base_url"]
    +            logger.debug(
    +                "api_server model route applied: model=%s provider=%s",
    +                model,
    +                runtime_kwargs.get("provider"),
    +            )
    +        elif route and session_override:
    +            logger.debug(
    +                "api_server model route skipped: session /model override wins for %s",
    +                gateway_session_key or session_id,
    +            )
    +
             user_config = _load_gateway_config()
             enabled_toolsets = sorted(_get_platform_tools(user_config, "api_server"))
     
    ```

    ```diff
    @@ -1255,25 +1394,40 @@ async def _handle_health_detailed(self, request: "web.Request") -> "web.Response
             })
     
         async def _handle_models(self, request: "web.Request") -> "web.Response":
    -        """GET /v1/models — return hermes-agent as an available model."""
    +        """GET /v1/models — list hermes-agent and any configured model_routes aliases."""
             auth_err = self._check_auth(request)
             if auth_err:
                 return auth_err
     
    -        return web.json_response({
    -            "object": "list",
    -            "data": [
    -                {
    -                    "id": self._model_name,
    -                    "object": "model",
    -                    "created": int(time.time()),
    -                    "owned_by": "hermes",
    -                    "permission": [],
    -                    "root": self._model_name,
    -                    "parent": None,
    -                }
    -            ],
    -        })
    +        now = int(time.time())
    +        models = [
    +            {
    +                "id": self._model_name,
    +                "object": "model",
    +                "created": now,
    +                "owned_by": "hermes",
    +                "permission": [],
    +                "root": self._model_name,
    +                "parent": None,
    +            }
    +        ]
    +        # Expose configured model route aliases so clients can discover them.
    +        # Only the alias and resolved model name are exposed — never provider
    +        # credentials.
    +        for alias, route_cfg in self._model_routes.items():
    +            if alias == self._model_name:
    +                continue  # already listed above
    +            models.append({
    +                "id": alias,
    +                "object": "model",
    +                "created": now,
    +                "owned_by": "hermes",
    +                "permission": [],
    +                "root": route_cfg.get("model", alias),
    +                "parent": self._model_name,
    +            })
    +
    +        return web.json_response({"object": "list", "data": models})
     
         async def _handle_capabilities(self, request: "web.Request") -> "web.Response":
             """GET /v1/capabilities — advertise the stable API surface.
    ```

    ```diff
    @@ -2012,6 +2166,11 @@ async def _handle_chat_completions(self, request: "web.Request") -> "web.Respons
             model_name = body.get("model", self._model_name)
             created = int(time.time())
     
    +        # Per-client model routing: if the requested model matches a
    +        # configured model_routes alias, this request's agent is created
    +        # with that route's model/provider instead of the global default.
    +        route = self._resolve_route(model_name)
    +
             if stream:
                 import queue as _q
                 _stream_q: _q.Queue = _q.Queue()
    ```

    ```diff
    @@ -2094,6 +2253,7 @@ def _on_tool_complete(tool_call_id, function_name, function_args, function_resul
                     tool_complete_callback=_on_tool_complete,
                     agent_ref=agent_ref,
                     gateway_session_key=gateway_session_key,
    +                route=route,
                 ))
                 # Ensure SSE drain loops can terminate without relying on polling
                 # agent_task.done(), which can race with queue timeout checks.
    ```

    ```diff
    @@ -2113,6 +2273,7 @@ async def _compute_completion():
                     ephemeral_system_prompt=system_prompt,
                     session_id=session_id,
                     gateway_session_key=gateway_session_key,
    +                route=route,
                 )
     
             idempotency_key = request.headers.get("Idempotency-Key")
    ```

    ```diff
    @@ -3123,6 +3284,9 @@ async def _handle_responses(self, request: "web.Request") -> "web.Response":
             # groups the entire conversation under one session entry.
             session_id = stored_session_id or str(uuid.uuid4())
     
    +        # Per-client model routing for /v1/responses (see model_routes).
    +        route = self._resolve_route(body.get("model"))
    +
             stream = _coerce_request_bool(body.get("stream"), default=False)
             if stream:
                 # Streaming branch — emit OpenAI Responses SSE events as the
    ```

    ```diff
    @@ -3176,6 +3340,7 @@ def _on_tool_complete(tool_call_id, function_name, function_args, function_resul
                     tool_complete_callback=_on_tool_complete,
                     agent_ref=agent_ref,
                     gateway_session_key=gateway_session_key,
    +                route=route,
                 ))
                 # Ensure SSE drain loops can terminate without relying on polling
                 # agent_task.done(), which can race with queue timeout checks.
    ```

    ```diff
    @@ -3209,6 +3374,7 @@ async def _compute_response():
                     ephemeral_system_prompt=instructions,
                     session_id=session_id,
                     gateway_session_key=gateway_session_key,
    +                route=route,
                 )
     
             idempotency_key = request.headers.get("Idempotency-Key")
    ```

    ```diff
    @@ -3839,6 +4005,7 @@ async def _run_agent(
             tool_complete_callback=None,
             agent_ref: Optional[list] = None,
             gateway_session_key: Optional[str] = None,
    +        route: Optional[Dict[str, Any]] = None,
         ) -> tuple:
             """
             Create an agent and run a conversation in a thread executor.
    ```

    ```diff
    @@ -3846,6 +4013,10 @@ async def _run_agent(
             Returns ``(result_dict, usage_dict)`` where *usage_dict* contains
             ``input_tokens``, ``output_tokens`` and ``total_tokens``.
     
    +        *route* is an optional ``model_routes`` entry (resolved from the
    +        request's ``model`` field) that overrides the global model/provider
    +        for this specific request.
    +
             If *agent_ref* is a one-element list, the AIAgent instance is stored
             at ``agent_ref[0]`` before ``run_conversation`` begins.  This allows
             callers (e.g. the SSE writer) to call ``agent.interrupt()`` from
    ```

    ```diff
    @@ -3870,6 +4041,7 @@ def _run():
                         tool_start_callback=tool_start_callback,
                         tool_complete_callback=tool_complete_callback,
                         gateway_session_key=gateway_session_key,
    +                    route=route,
                     )
                     if agent_ref is not None:
                         agent_ref[0] = agent
    ```

    ```diff
    @@ -4085,6 +4257,9 @@ def _text_cb(delta: Optional[str]) -> None:
                 model=body.get("model", self._model_name),
             )
     
    +        # Per-client model routing for /v1/runs (see model_routes).
    +        route = self._resolve_route(body.get("model"))
    +
             async def _run_and_close():
                 try:
                     self._set_run_status(run_id, "running")
    ```

    ```diff
    @@ -4094,6 +4269,7 @@ async def _run_and_close():
                         stream_delta_callback=_text_cb,
                         tool_progress_callback=event_cb,
                         gateway_session_key=gateway_session_key,
    +                    route=route,
                     )
                     self._active_run_agents[run_id] = agent
     
    ```
  - `scripts/release.py`

    ```diff
    @@ -46,6 +46,7 @@
     # Auto-extracted from noreply emails + manual overrides
     AUTHOR_MAP = {
         "[redacted:email]": "Mibayy",  # PR #3243 salvage (/compact alias + preview/aggressive flags for /compress)
    +    "[redacted:email]": "Mibayy",  # PR #3176 salvage (api-server: per-client model routing via model_routes)
         "[redacted:email]": "CrazyBoyM",  # PR #55828 salvage (image_gen openai-codex: wire image-to-image / reference-image editing via Codex Responses input_image parts; magic-byte + read-guard + 25MB-cap + clamp-to-16 hardening)
         "r0gersm1th@users.noreply.github.com": "r0gersm1th",  # PR #3219 salvage (whatsapp bridge: resolve LID sender IDs to phone numbers in the message payload so phone-based allowlists match; commit authored by collaborator r0gersm1th, PR by @ajmeese7)
         "[redacted:email]": "Mibayy",  # PR #3296 salvage (status: provider label honors config.yaml model.base_url, not just OPENAI_BASE_URL env)
    ```
  - `tests/gateway/test_api_server.py`

    ```diff
    @@ -3798,3 +3798,278 @@ async def test_capabilities_advertises_session_key_header(self, adapter):
                 assert resp.status == 200
                 data = await resp.json()
                 assert data["features"]["session_key_header"] == "X-Hermes-Session-Key"
    +
    +
    +# ---------------------------------------------------------------------------
    +# Per-client model routing (model_routes)
    +# ---------------------------------------------------------------------------
    +
    +
    +def _make_routing_adapter(routes) -> APIServerAdapter:
    +    """Create an adapter with model_routes configured."""
    +    config = PlatformConfig(enabled=True, extra={"model_routes": routes})
    +    return APIServerAdapter(config)
    +
    +
    +def _patch_create_agent_runtime(monkeypatch, captured: dict, fake_agent_cls):
    +    """Stub out every external dependency of _create_agent."""
    +    monkeypatch.setattr("run_agent.AIAgent", fake_agent_cls)
    +    monkeypatch.setattr(
    +        "gateway.run._resolve_runtime_agent_kwargs",
    +        lambda: {
    +            "provider": "openrouter",
    +            "api_key": "[redacted:secret-assignment]",
    +            "base_url": "https://openrouter.ai/api/v1",
    +            "api_mode": "chat_completions",
    +        },
    +    )
    +    monkeypatch.setattr("gateway.run._resolve_gateway_model", lambda: "global/model")
    +    monkeypatch.setattr("gateway.run._load_gateway_config", lambda: {})
    +    monkeypatch.setattr(
    +        "gateway.run.GatewayRunner._load_reasoning_config", staticmethod(lambda: {})
    +    )
    +    monkeypatch.setattr(
    +        "gateway.run.GatewayRunner._load_fallback_model", staticmethod(lambda: None)
    +    )
    +    monkeypatch.setattr("gateway.run._current_max_iterations", lambda: 90)
    +    monkeypatch.setattr("hermes_cli.tools_config._get_platform_tools", lambda *_: set())
    +
    +
    +class TestModelRoutesParsing:
    +    def test_valid_routes_are_parsed(self):
    +        routes = {"minimax-m2": {"model": "minimax/minimax-m1", "provider": "openrouter"}}
    +        adapter = _make_routing_adapter(routes)
    +        assert adapter._model_routes == routes
    +
    +    def test_non_dict_routes_config_is_ignored(self):
    +        adapter = _make_routing_adapter("not-a-dict")
    +        assert adapter._model_routes == {}
    +
    +    def test_route_without_model_is_dropped(self):
    +        adapter = _make_routing_adapter({"bad": {"provider": "openrouter"}})
    +        assert adapter._model_routes == {}
    +
    +    def test_route_with_non_dict_value_is_dropped(self):
    +        adapter = _make_routing_adapter({"bad": "gpt-5", "good": {"model": "openai/gpt-5"}})
    +        assert set(adapter._model_routes) == {"good"}
    +
    +    def test_unknown_route_keys_are_stripped(self):
    +        adapter = _make_routing_adapter(
    +            {"a": {"model": "m", "provider": "p", "evil_extra": "x"}}
    +        )
    +        assert adapter._model_routes["a"] == {"model": "m", "provider": "p"}
    +
    +    def test_resolve_route_lookup(self):
    +        adapter = _make_routing_adapter({"minimax-m2": {"model": "minimax/minimax-m1"}})
    +        assert adapter._resolve_route("minimax-m2") == {"model": "minimax/minimax-m1"}
    +        assert adapter._resolve_route("unknown-model") is None
    +        assert adapter._resolve_route(None) is None
    +        assert adapter._resolve_route(123) is None
    +
    +    def test_no_routes_configured(self):
    +        adapter = _make_routing_adapter({})
    +        assert adapter._resolve_route("hermes-agent") is None
    +
    +
    +class TestModelRoutesModelsEndpoint:
    +    @pytest.mark.asyncio
    +    async def test_models_endpoint_lists_route_aliases(self):
    +        routes = {
    +            "minimax-m2": {"model": "minimax/minimax-m1", "provider": "openrouter"},
    +            "gpt-5": {"model": "openai/gpt-5"},
    +        }
    +        adapter = _make_routing_adapter(routes)
    +        app = _create_app(adapter)
    +        async with TestClient(TestServer(app)) as cli:
    +            resp = await cli.get("/v1/models")
    +            assert resp.status == 200
    +            data = await resp.json()
    +            ids = {m["id"] for m in data["data"]}
    +            assert adapter._model_name in ids
    +            assert "minimax-m2" in ids
    +            assert "gpt-5" in ids
    +
    +    @pytest.mark.asyncio
    +    async def test_models_endpoint_route_alias_fields_and_no_secrets(self):
    +        routes = {"my-alias": {"model": "openai/gpt-5", "api_key": "[redacted:secret-assignment]"}}
    +        adapter = _make_routing_adapter(routes)
    +        app = _create_app(adapter)
    +        async with TestClient(TestServer(app)) as cli:
    +            resp = await cli.get("/v1/models")
    +            data = await resp.json()
    +            alias_entry = next(m for m in data["data"] if m["id"] == "my-alias")
    +            assert alias_entry["root"] == "openai/gpt-5"
    +            assert alias_entry["parent"] == adapter._model_name
    +            # per-route api_key must never leak through the discovery endpoint
    +            assert "sk-route-secret" not in json.dumps(data)
    +
    +
    +class TestModelRoutesHandlers:
    +    @pytest.mark.asyncio
    +    async def test_chat_completions_passes_route_to_run_agent(self):
    +        routes = {"minimax-m2": {"model": "minimax/minimax-m1", "provider": "openrouter"}}
    +        adapter = _make_routing_adapter(routes)
    +        app = _create_app(adapter)
    +        async with TestClient(TestServer(app)) as cli:
    +            with patch.object(adapter, "_run_agent", new_callable=AsyncMock) as mock_run:
    +                mock_run.return_value = (
    +                    {"final_response": "hi", "messages": [], "api_calls": 1},
    +                    {"input_tokens": 5, "output_tokens": 5, "total_tokens": 10},
    +                )
    +                resp = await cli.post("/v1/chat/completions", json={
    +                    "model": "minimax-m2",
    +                    "messages": [{"role": "user", "content": "hello"}],
    +                })
    +                assert resp.status == 200
    +                kwargs = mock_run.call_args.kwargs
    +                assert kwargs.get("route") == {
    +                    "model": "minimax/minimax-m1", "provider": "openrouter",
    +                }
    +
    +    @pytest.mark.asyncio
    +    async def test_chat_completions_no_route_for_unknown_model(self):
    +        adapter = _make_routing_adapter({"minimax-m2": {"model": "minimax/minimax-m1"}})
    +        app = _create_app(adapter)
    +        async with TestClient(TestServer(app)) as cli:
    +            with patch.object(adapter, "_run_agent", new_callable=AsyncMock) as mock_run:
    +                mock_run.return_value = (
    +                    {"final_response": "hi", "messages": [], "api_calls": 1},
    +                    {"input_tokens": 5, "output_tokens": 5, "total_tokens": 10},
    +                )
    +                resp = await cli.post("/v1/chat/completions", json={
    +                    "model": "unknown-model",
    +                    "messages": [{"role": "user", "content": "hello"}],
    +                })
    +                assert resp.status == 200
    +                assert mock_run.call_args.kwargs.get("route") is None
    +
    +    @pytest.mark.asyncio
    +    async def test_responses_api_passes_route_to_run_agent(self):
    +        routes = {"xiaozhi": {"model": "minimax/minimax-m1", "provider": "openrouter"}}
    +        adapter = _make_routing_adapter(routes)
    +        app = _create_app(adapter)
    +        async with TestClient(TestServer(app)) as cli:
    +            with patch.object(adapter, "_run_agent", new_callable=AsyncMock) as mock_run:
    +                mock_run.return_value = (
    +                    {"final_response": "hi", "messages": [], "api_calls": 1},
    +                    {"input_tokens": 5, "output_tokens": 5, "total_tokens": 10},
    +                )
    +                resp = await cli.post("/v1/responses", json={
    +                    "model": "xiaozhi",
    +                    "input": "hello",
    +                })
    +                assert resp.status == 200
    +                assert mock_run.call_args.kwargs.get("route") == {
    +                    "model": "minimax/minimax-m1", "provider": "openrouter",
    +                }
    +
    +
    +class TestModelRoutesAgentCreation:
    +    def test_route_overrides_model_and_credentials(self, monkeypatch):
    +        captured = {}
    +
    +        class FakeAgent:
    +            def __init__(self, **kwargs):
    +                captured.update(kwargs)
    +
    +        _patch_create_agent_runtime(monkeypatch, captured, FakeAgent)
    +        adapter = _make_routing_adapter(
    +            {"alias": {
    +                "model": "minimax/minimax-m1",
    +                "api_key": "[redacted:secret-assignment]",
    +                "base_url": "https://route.example/v1",
    +            }}
    +        )
    +        monkeypatch.setattr(adapter, "_ensure_session_db", lambda: None)
    +        monkeypatch.setattr(adapter, "_session_model_override_for", lambda *_: None)
    +
    +        agent = adapter._create_agent(
    +            session_id="s1", route=adapter._resolve_route("alias")
    +        )
    +
    +        assert isinstance(agent, FakeAgent)
    +        assert captured["model"] == "minimax/minimax-m1"
    +        assert captured["api_key"] == "sk-route"
    +        assert captured["base_url"] == "https://route.example/v1"
    +
    +    def test_route_provider_resolves_provider_credentials(self, monkeypatch):
    +        captured = {}
    +
    +        class FakeAgent:
    +            def __init__(self, **kwargs):
    +                captured.update(kwargs)
    +
    +        _patch_create_agent_runtime(monkeypatch, captured, FakeAgent)
    +        monkeypatch.setattr(
    +            "gateway.run._resolve_runtime_agent_kwargs_for_provider",
    +            lambda provider: {
    +                "provider": provider,
    +                "api_key": f"sk-{provider}",
    +                "base_url": f"https://{provider}.example/v1",
    +                "api_mode": "chat_completions",
    +            },
    +        )
    +        adapter = _make_routing_adapter(
    +            {"alias": {"model": "other/model", "provider": "otherprov"}}
    +        )
    +        monkeypatch.setattr(adapter, "_ensure_session_db", lambda: None)
    +        monkeypatch.setattr(adapter, "_session_model_override_for", lambda *_: None)
    +
    +        adapter._create_agent(session_id="s1", route=adapter._resolve_route("alias"))
    +
    +        assert captured["model"] == "other/model"
    +        assert captured["provider"] == "otherprov"
    +        assert captured["api_key"] == "sk-otherprov"
    +
    +    def test_no_route_keeps_global_model(self, monkeypatch):
    +        captured = {}
    +
    +        class FakeAgent:
    +            def __init__(self, **kwargs):
    +                captured.update(kwargs)
    +
    +        _patch_create_agent_runtime(monkeypatch, captured, FakeAgent)
    +        adapter = _make_routing_adapter({"alias": {"model": "other/model"}})
    +        monkeypatch.setattr(adapter, "_ensure_session_db", lambda: None)
    +        monkeypatch.setattr(adapter, "_session_model_override_for", lambda *_: None)
    +
    +        adapter._create_agent(session_id="s1", route=None)
    +
    +        assert captured["model"] == "global/model"
    +        assert captured["api_key"] == "sk-global"
    +
    +    def test_session_model_override_beats_route(self, monkeypatch):
    +        """A user-issued /model on the session must win over static route config."""
    +        captured = {}
    +
    +        class FakeAgent:
    +            def __init__(self, **kwargs):
    +                captured.update(kwargs)
    +
    +        _patch_create_agent_runtime(monkeypatch, captured, FakeAgent)
    +        adapter = _make_routing_adapter({"alias": {"model": "route/model", "api_key": "[redacted:secret-assignment]"}})
    +        monkeypatch.setattr(adapter, "_ensure_session_db", lambda: None)
    +        monkeypatch.setattr(
    +            adapter,
    +            "_session_model_override_for",
    +            lambda key: {"model": "session/override-model"},
    +        )
    +
    +        adapter._create_agent(session_id="s1", route=adapter._resolve_route("alias"))
    +
    +        # The route must NOT be applied — the session override path (global
    +        # runtime here, since the gateway applies /model separately) wins.
    +        assert captured["model"] == "global/model"
    +        assert captured["api_key"] == "sk-global"
    +
    +    def test_session_override_lookup_reads_gateway_runner(self, monkeypatch):
    +        """_session_model_override_for consults GatewayRunner._session_model_overrides."""
    +        adapter = _make_routing_adapter({})
    +
    +        class FakeRunner:
    +            _session_model_overrides = {"chan-1": {"model": "user/model"}}
    +
    +        monkeypatch.setattr("gateway.run._gateway_runner_ref", lambda: FakeRunner())
    +        assert adapter._session_model_override_for("chan-1") == {"model": "user/model"}
    +        assert adapter._session_model_override_for("chan-2") is None
    +        assert adapter._session_model_override_for(None) is None
    ```
- **#3176: feat(api-server): per-client model routing via model_routes config**. closed without merge
  - maintainer direction: category `merged_via`; implication `superseded`; refs `#57028`
- **#10773: feat(api-server): honor request-level model field for per-request model selection**. open
  - maintainer direction: category `duplicate_of`; implication `duplicate`; refs `#31139`, `#27678`, `#10773`
  - files: `gateway/platforms/api_server.py`

## Fix status

- Merged to main in PR #57028 (cited above). No release data is tracked. Verify downstream.
- Fix last verified (merged) `2026-07-02`.
- Subsystem: `gateway`