feat(onboarding): verify Test connection without a model id (#792)

opensquilla-webui/src/composables/setup/useSetupCatalog.privacy.test.ts

@@ -5627,16 +5627,19 @@ describe('useSetupCatalog optional provider credentials', () => {
         source: 'not_required',
         probeReady: false,
       })
+      // The model id no longer gates the probe (#792): only the still-empty
+      // non-model required field (Base URL) blocks it.
       expect(credential?.probeDisabledReason).toBe(
-        'Complete required fields before verifying: Model, Base URL.',
+        'Complete required fields before verifying: Base URL.',
       )
 
       api.probeProviderConnection()
       expect(rpcCall.mock.calls.some(call => call[0] === 'onboarding.provider.probe')).toBe(false)
 
-      api.updateProviderField('model', 'test-model')
       api.updateProviderField('base_url', 'https://custom.example.test/v1')
       credential = api.providerPanel.value.credentialPanel
+      // An empty model is now allowed: reachability is verified via the
+      // model-list endpoint.
       expect(credential?.probeReady).toBe(true)
       expect(credential?.probeDisabledReason).toBe('')
 
@@ -5645,7 +5648,6 @@ describe('useSetupCatalog optional provider credentials', () => {
       expect(rpcCall).toHaveBeenCalledWith('onboarding.provider.probe', {
         providerId,
         baseUrl: 'https://custom.example.test/v1',
-        model: 'test-model',
       })
       app.unmount()
     },

opensquilla-webui/src/composables/setup/useSetupCatalog.ts

@@ -1521,19 +1521,20 @@ const providerProbeModel = computed(() => {
 
 const providerProbeMissingFields = computed(() => {
   if (!providerForm.selectedProvider.value) return []
+  // Stored/draft profiles probe through onboarding.llmProfile[.draft].probe,
+  // which still resolves a concrete deployment model, so keep requiring one.
   if (providerSelectionKind.value === 'profile') {
     return providerProbeModel.value ? [] : [t('setup.common.model')]
   }
+  // For a draft primary-provider config the model id no longer gates the
+  // probe: an empty model makes onboarding.provider.probe verify reachability
+  // via the model-list endpoint instead of a chat turn (#792).
   return providerFields.value
     .filter(field => field.required === true && !isProviderCredentialField(field))
-    .filter(field => {
-      const value = field.name === 'model'
-        && editingPrimaryProvider.value
-        && hasConfiguredPrimaryProvider.value
-        ? currentFormModelValue()
-        : providerForm.fieldValue(field, currentProviderConfig.value)
-      return !String(value ?? '').trim()
-    })
+    .filter(field => field.name !== 'model')
+    .filter(field => !String(
+      providerForm.fieldValue(field, currentProviderConfig.value) ?? '',
+    ).trim())
     .map(providerProbeFieldLabel)
 })
 

src/opensquilla/gateway/rpc_onboarding.py

@@ -1020,7 +1020,7 @@ async def _probe_primary_provider(
     config: Any,
     usage_event_sink: Any,
 ) -> dict[str, Any]:
-    """Live one-token probe of a candidate provider config (nothing is saved)."""
+    """Live probe of a candidate provider config without saving it."""
     provider_id = command.provider_id
     cfg = config
     api_key = str(command.api_key or "")
@@ -1061,19 +1061,42 @@ async def _probe_primary_provider(
         if not proxy:
             proxy = str(getattr(cfg.llm, "proxy", "") or "")
     model = str(command.model or "")
+    allow_default_api_key_env = not same_provider or reuse_stored_credentials
     with _validation_error("onboarding.provider.invalid"):
-        result = await _usage_accounted_provider_probe(
-            usage_event_sink,
-            provider_id=str(provider_id),
-            model=model,
-            api_key=api_key,
-            api_key_env=api_key_env,
-            base_url=base_url,
-            proxy=proxy,
-            allow_default_api_key_env=(
-                not same_provider or reuse_stored_credentials
-            ),
-        )
+        if model.strip():
+            result = await _usage_accounted_provider_probe(
+                usage_event_sink,
+                provider_id=str(provider_id),
+                model=model,
+                api_key=api_key,
+                api_key_env=api_key_env,
+                base_url=base_url,
+                proxy=proxy,
+                allow_default_api_key_env=allow_default_api_key_env,
+            )
+        else:
+            # A model is unnecessary for an endpoint/credential connectivity
+            # check; model discovery exercises that path without a chat turn.
+            from opensquilla.onboarding.probe import (
+                ProviderProbeResult,
+                discover_provider_models,
+            )
+
+            listing = await discover_provider_models(
+                provider_id=str(provider_id),
+                api_key=api_key,
+                api_key_env=api_key_env,
+                base_url=base_url,
+                proxy=proxy,
+                allow_default_api_key_env=allow_default_api_key_env,
+            )
+            result = ProviderProbeResult(
+                ok=listing.ok,
+                provider_id=str(provider_id),
+                model="",
+                failure_kind=listing.failure_kind,
+                message=listing.detail,
+            )
     saved_model = str(getattr(cfg.llm, "model", "") or "").strip()
     if (
         same_provider

tests/test_gateway/test_rpc_onboarding.py

@@ -345,9 +345,7 @@ def sync_primary(self, provider_config):
             sync_calls.append(provider_config)
 
     ctx = _admin_ctx()
-    ctx.config = GatewayConfig(
-        llm={"provider": "dashscope", "model": "qwen3.7-plus"}
-    )
+    ctx.config = GatewayConfig(llm={"provider": "dashscope", "model": "qwen3.7-plus"})
     ctx.config.config_path = str(tmp_path / "c.toml")
     ctx.provider_selector = FakeSelector()
 
@@ -375,10 +373,7 @@ def sync_primary(self, provider_config):
     assert res.error is None, res.error
     assert ctx.config.squilla_router.default_tier == "c0"
     assert ctx.config.squilla_router.tiers["c0"]["provider"] == "volcengine"
-    assert (
-        ctx.config.squilla_router.tiers["c0"]["model"]
-        == "doubao-seed-1-6-251015"
-    )
+    assert ctx.config.squilla_router.tiers["c0"]["model"] == "doubao-seed-1-6-251015"
     assert ctx.config.llm.provider == "dashscope"
     assert ctx.config.llm.model == "qwen3.7-plus"
     assert len(sync_calls) == 1
@@ -389,10 +384,7 @@ def sync_primary(self, provider_config):
     assert persisted["llm"]["provider"] == "dashscope"
     assert persisted["llm"]["model"] == "qwen3.7-plus"
     assert persisted["squilla_router"]["default_tier"] == "c0"
-    assert (
-        persisted["squilla_router"]["tiers"]["c0"]["model"]
-        == "doubao-seed-1-6-251015"
-    )
+    assert persisted["squilla_router"]["tiers"]["c0"]["model"] == "doubao-seed-1-6-251015"
 
 
 @pytest.mark.asyncio
@@ -501,9 +493,7 @@ async def test_router_catalog_rpc(tmp_path, monkeypatch):
 
 
 @pytest.mark.asyncio
-async def test_ensemble_configure_partial_payload_updates_and_persists(
-    tmp_path, monkeypatch
-):
+async def test_ensemble_configure_partial_payload_updates_and_persists(tmp_path, monkeypatch):
     monkeypatch.setenv("OPENSQUILLA_GATEWAY_CONFIG_PATH", str(tmp_path / "c.toml"))
     from opensquilla.gateway.config import GatewayConfig
 
@@ -540,9 +530,7 @@ async def test_ensemble_configure_partial_payload_updates_and_persists(
 
 
 @pytest.mark.asyncio
-async def test_ensemble_configure_accepts_full_camel_case_payload(
-    tmp_path, monkeypatch
-):
+async def test_ensemble_configure_accepts_full_camel_case_payload(tmp_path, monkeypatch):
     monkeypatch.setenv("OPENSQUILLA_GATEWAY_CONFIG_PATH", str(tmp_path / "c.toml"))
     res = await get_dispatcher().dispatch(
         "r1",
@@ -575,9 +563,7 @@ async def test_ensemble_configure_accepts_full_camel_case_payload(
 
 
 @pytest.mark.asyncio
-async def test_ensemble_configure_rejects_out_of_range_proposer_retries(
-    tmp_path, monkeypatch
-):
+async def test_ensemble_configure_rejects_out_of_range_proposer_retries(tmp_path, monkeypatch):
     monkeypatch.setenv("OPENSQUILLA_GATEWAY_CONFIG_PATH", str(tmp_path / "c.toml"))
     res = await get_dispatcher().dispatch(
         "r1",
@@ -751,10 +737,7 @@ async def test_image_generation_configure_redacts_api_key(tmp_path, monkeypatch)
 
     data = tomllib.loads(target.read_text())
     assert data["image_generation"]["enabled"] is True
-    assert (
-        data["image_generation"]["primary"]
-        == "openrouter/google/gemini-3.1-flash-image-preview"
-    )
+    assert data["image_generation"]["primary"] == "openrouter/google/gemini-3.1-flash-image-preview"
     assert data["image_generation"]["providers"]["openrouter"]["api_key"] == "sk-or"
 
 
@@ -1132,10 +1115,7 @@ async def test_image_generation_configure_can_disable_legacy_invalid_config(
     assert data["image_generation"]["enabled"] is False
     assert data["image_generation"]["primary"] == "openrouter/google//image"
     assert data["image_generation"]["fallbacks"] == ["openai/"]
-    assert (
-        data["image_generation"]["providers"]["openrouter"]["base_url"]
-        == "not-a-url"
-    )
+    assert data["image_generation"]["providers"]["openrouter"]["base_url"] == "not-a-url"
 
 
 @pytest.mark.asyncio
@@ -1171,9 +1151,7 @@ async def test_onboarding_status_marks_legacy_image_endpoint_mismatch_degraded(
     ctx = _read_ctx()
     ctx.config = GatewayConfig()
     ctx.config.image_generation.enabled = True
-    ctx.config.image_generation.primary = (
-        "openrouter/google/gemini-3.1-flash-image-preview"
-    )
+    ctx.config.image_generation.primary = "openrouter/google/gemini-3.1-flash-image-preview"
     openrouter_provider = ctx.config.image_generation.providers.openrouter
     openrouter_provider.api_key = "sk-synthetic-image"
     openrouter_provider.base_url = "https://api.openai.com/v1"
@@ -1416,9 +1394,7 @@ async def test_memory_embedding_configure_updates_ctx_config(tmp_path, monkeypat
 
 
 @pytest.mark.asyncio
-async def test_memory_embedding_configure_auto_can_store_remote_fallback(
-    tmp_path, monkeypatch
-):
+async def test_memory_embedding_configure_auto_can_store_remote_fallback(tmp_path, monkeypatch):
     monkeypatch.setenv("OPENSQUILLA_GATEWAY_CONFIG_PATH", str(tmp_path / "c.toml"))
     from opensquilla.gateway.config import GatewayConfig
 
@@ -1534,9 +1510,7 @@ async def test_provider_configure_does_not_persist_runtime_api_key(tmp_path, mon
 
 
 @pytest.mark.asyncio
-async def test_provider_configure_persists_explicit_replacement_for_env_key(
-    tmp_path, monkeypatch
-):
+async def test_provider_configure_persists_explicit_replacement_for_env_key(tmp_path, monkeypatch):
     monkeypatch.setenv("OPENROUTER_API_KEY", "startup-key")
     from opensquilla.gateway.config import GatewayConfig
 
@@ -1921,9 +1895,7 @@ async def fake_discover(**kwargs):
 
 
 @pytest.mark.asyncio
-async def test_models_discover_unverified_provider_stays_empty_without_build(
-    tmp_path, monkeypatch
-):
+async def test_models_discover_unverified_provider_stays_empty_without_build(tmp_path, monkeypatch):
     monkeypatch.setenv("OPENSQUILLA_GATEWAY_CONFIG_PATH", str(tmp_path / "c.toml"))
 
     def _unexpected_build(*_args, **_kwargs):
@@ -1949,6 +1921,58 @@ def _unexpected_build(*_args, **_kwargs):
     }
 
 
+@pytest.mark.asyncio
+async def test_provider_probe_without_model_verifies_via_model_list(tmp_path, monkeypatch):
+    """An empty model probes reachability through the model-list endpoint
+    instead of raising ``Model is required`` (#792)."""
+    monkeypatch.setenv("OPENSQUILLA_GATEWAY_CONFIG_PATH", str(tmp_path / "c.toml"))
+    _stub_openai_transport(
+        monkeypatch,
+        httpx.Response(
+            200,
+            headers={"content-type": "application/json"},
+            content=b'{"data": [{"id": "gpt-x", "context_length": 32000}]}',
+        ),
+    )
+    res = await get_dispatcher().dispatch(
+        "r1",
+        "onboarding.provider.probe",
+        {"providerId": "openrouter", "apiKey": "sk-test"},
+        _admin_ctx(),
+    )
+    assert res.error is None, res.error
+    assert res.payload["ok"] is True
+    assert res.payload["model"] == ""
+    assert res.payload["failureKind"] == ""
+    # No chat round-trip happened; the chat-only timings stay at their
+    # never-reached-the-network sentinels.
+    assert res.payload["latencyMs"] == 0
+    assert res.payload["firstResponseMs"] is None
+
+
+@pytest.mark.asyncio
+async def test_provider_probe_without_model_reports_auth_failure(tmp_path, monkeypatch):
+    """A model-less probe surfaces a bad key through the same envelope."""
+    monkeypatch.setenv("OPENSQUILLA_GATEWAY_CONFIG_PATH", str(tmp_path / "c.toml"))
+    _stub_openai_transport(
+        monkeypatch,
+        httpx.Response(
+            401,
+            headers={"content-type": "application/json"},
+            content=b'{"error": {"message": "Incorrect API key provided"}}',
+        ),
+    )
+    res = await get_dispatcher().dispatch(
+        "r1",
+        "onboarding.provider.probe",
+        {"providerId": "openrouter", "apiKey": "sk-bad"},
+        _admin_ctx(),
+    )
+    assert res.error is None, res.error
+    assert res.payload["ok"] is False
+    assert res.payload["failureKind"] == "auth_invalid"
+
+
 @pytest.mark.asyncio
 async def test_image_models_discover_requires_admin_scope(tmp_path, monkeypatch):
     monkeypatch.setenv("OPENSQUILLA_GATEWAY_CONFIG_PATH", str(tmp_path / "c.toml"))
@@ -1965,9 +1989,7 @@ async def test_image_models_discover_requires_admin_scope(tmp_path, monkeypatch)
 
 
 @pytest.mark.asyncio
-async def test_image_models_discover_returns_image_specific_catalog(
-    tmp_path, monkeypatch
-):
+async def test_image_models_discover_returns_image_specific_catalog(tmp_path, monkeypatch):
     monkeypatch.setenv("OPENSQUILLA_GATEWAY_CONFIG_PATH", str(tmp_path / "c.toml"))
 
     async def _discover(provider_id: str):
@@ -1980,8 +2002,7 @@ async def _discover(provider_id: str):
         }
 
     monkeypatch.setattr(
-        "opensquilla.onboarding.image_generation_model_discovery."
-        "discover_image_generation_models",
+        "opensquilla.onboarding.image_generation_model_discovery.discover_image_generation_models",
         _discover,
     )