fix(ollama): route models by advertised capability (#11087) — port of #11088 to the release line

changelog.d/fixes/11271-ollama-capability-routing.md

@@ -0,0 +1 @@
+- **fix(ollama):** Ollama Local models are no longer flattened to `chat` at sync time — the synced store persists every advertised capability and chat filtering moves to read time, so `/v1/embeddings` and `/v1/images/generations` stop rejecting models the daemon reports as capable ([#11271](https://github.com/diegosouzapw/OmniRoute/pull/11271)) — thanks @yourspraveen

open-sse/services/combo/autoStrategy.ts

@@ -28,6 +28,7 @@ import type {
   ResolvedComboTarget,
 } from "./types.ts";
 import { extractSessionAffinityKey } from "@/sse/services/auth";
+import { filterChatSelectableModels } from "../modelEndpointPolicy.ts";
 import { DEFAULT_INTENT_CONFIG, type IntentClassifierConfig } from "../intentClassifier.ts";
 import { getTaskFitness } from "../autoCombo/taskFitness.ts";
 import {
@@ -470,10 +471,13 @@ export async function expandAutoComboCandidatePool(
       // catalog only when the user has none. This keeps catalog-only models
       // (e.g. openrouter/auto) out of pure-auto pools when the operator only
       // synced a subset (e.g. OpenRouter with importFreeModelsOnly).
-      const [syncedModels, customModels] = await Promise.all([
+      // #11088 (option 1): the synced store now persists non-chat models too —
+      // chat combo pools must keep filtering them out at read time.
+      const [syncedModelsRaw, customModels] = await Promise.all([
         getSyncedAvailableModels(providerId),
         getCustomModels(providerId),
       ]);
+      const syncedModels = filterChatSelectableModels(providerId, syncedModelsRaw);
       const hiddenModels = hiddenModelsMap.get(providerId);
       const userVisibleIds = new Set<string>();
       for (const m of syncedModels) if (m.id && !hiddenModels?.has(m.id)) userVisibleIds.add(m.id);

src/app/api/providers/[id]/models/discovery/helpers.ts

@@ -1,5 +1,11 @@
 import { isSelfHostedChatProvider } from "@/shared/constants/providers";
 import { getStaticModelsForProvider, type LocalCatalogModel } from "@/lib/providers/staticModels";
+import { SAFE_OUTBOUND_FETCH_PRESETS, safeOutboundFetch } from "@/shared/network/safeOutboundFetch";
+import { getProviderValidationGuard } from "@/shared/network/outboundUrlGuardPolicy";
+import {
+  buildOllamaShowUrl,
+  enrichOllamaModelsWithCapabilities,
+} from "@/lib/providerModels/ollamaCapabilities";
 
 export type JsonRecord = Record<string, unknown>;
 
@@ -102,3 +108,35 @@ export function buildNamedOpenAiStyleHeaders(
 
   return headers;
 }
+
+// #11087 — Ollama's OpenAI-compatible /v1/models response carries no capability
+// data, so every local model looked like a chat model and image/embedding
+// requests were routed to text-only models. Probe /api/show per model (bounded
+// concurrency, failures degrade to the unenriched entry) to recover the
+// advertised capabilities. Lives here rather than inline in route.ts to keep the
+// route file under its frozen file-size cap.
+export async function enrichOllamaLocalModels(
+  models: unknown[],
+  baseUrl: string,
+  proxy: unknown,
+  token: string | null | undefined
+): Promise<JsonRecord[]> {
+  const showUrl = buildOllamaShowUrl(baseUrl);
+  return enrichOllamaModelsWithCapabilities(models, async (modelId) => {
+    try {
+      const showResponse = await safeOutboundFetch(showUrl, {
+        ...SAFE_OUTBOUND_FETCH_PRESETS.modelsProbe,
+        // Same guard tier as the discovery probe above: local-first, so LAN
+        // Ollama hosts are reachable while the outbound guard stays enforced.
+        guard: getProviderValidationGuard(),
+        proxyConfig: proxy,
+        method: "POST",
+        headers: buildOptionalBearerHeaders(token),
+        body: JSON.stringify({ model: modelId, verbose: false }),
+      });
+      return showResponse.ok ? await showResponse.json() : null;
+    } catch {
+      return null;
+    }
+  });
+}

src/app/api/providers/[id]/models/route.ts

@@ -108,6 +108,7 @@ import {
   mergeSpecialtyCatalogIntoLiveModels,
   buildOptionalBearerHeaders,
   buildNamedOpenAiStyleHeaders,
+  enrichOllamaLocalModels,
 } from "./discovery/helpers";
 import {
   fetchAntigravityDiscoveryModelsCached,
@@ -794,6 +795,8 @@ export async function GET(
             models = isNamedOpenAIStyleProvider(provider)
               ? normalizeOpenAiLikeModelsResponse(data, provider)
               : data.data || data.models || [];
+            if (provider === "ollama-local")
+              models = await enrichOllamaLocalModels(models, baseUrl, proxy, token);
             break; // Success!
           }
 

src/app/api/v1/images/generations/route.ts

@@ -23,6 +23,10 @@ import { getComboByName } from "@/lib/db/combos";
 import { getAllCustomModels } from "@/lib/db/models";
 import { resolveProxyForConnection } from "@/lib/db/settings";
 import { resolveImageRouteModel } from "@/lib/images/imageRouteModel";
+import {
+  resolveLocalSyncedEndpointRoute,
+  type LocalSyncedEndpointRoute,
+} from "@/lib/providerModels/syncedEndpointRouting";
 import { runWithProxyContext } from "@omniroute/open-sse/utils/proxyFetch.ts";
 import { attachOmniRouteMetaHeaders } from "@/domain/omnirouteResponseMeta";
 import { calculateModalCost } from "@/lib/usage/costCalculator";
@@ -145,6 +149,16 @@ async function postHandler(request, context) {
   // Parse model to get provider
   let { provider, model: requestedModel } = parseImageModel(body.model);
   let isCustomModel = false;
+  let syncedEndpointRoute: LocalSyncedEndpointRoute | null = null;
+
+  if (!provider) {
+    syncedEndpointRoute = await resolveLocalSyncedEndpointRoute(body.model, "images");
+    if (syncedEndpointRoute) {
+      provider = syncedEndpointRoute.provider;
+      body.model = `${syncedEndpointRoute.provider}/${syncedEndpointRoute.model}`;
+      isCustomModel = true;
+    }
+  }
 
   // If not in built-in registry, check custom models tagged for images
   if (!provider) {
@@ -231,9 +245,8 @@ async function postHandler(request, context) {
     credentials = await getProviderCredentialsWithQuotaPreflight(
       provider,
       null,
-      null,
-      requestedModel
-    );
+      syncedEndpointRoute?.connectionIds ?? null,
+      requestedModel    );
     if (!credentials) {
       return errorResponse(
         HTTP_STATUS.BAD_REQUEST,

src/lib/embeddings/service.ts

@@ -31,6 +31,7 @@ import { isPrivateHost, isCloudMetadataHost } from "@/shared/network/outboundUrl
 import { calculateCost } from "@/lib/usage/costCalculator";
 import { attachOmniRouteMetaHeaders } from "@/domain/omnirouteResponseMeta";
 import { generateRequestId } from "@/shared/utils/requestId";
+import { resolveLocalSyncedEndpointRoute } from "@/lib/providerModels/syncedEndpointRouting";
 
 type ValidatedEmbeddingBody = Record<string, unknown> & { model: string };
 type ProviderCredentialsResult = Awaited<ReturnType<typeof getProviderCredentials>>;
@@ -164,21 +165,74 @@ export async function createEmbeddingResponse(
         model: options.resolvedModel ?? body.model,
       }
     : parseEmbeddingModel(body.model, dynamicProviders);
-  const { provider, model: resolvedModel } = parsedModel;
+  let { provider, model: resolvedModel } = parsedModel;
+  // #11088: a bare local-model request routes through the connection that
+  // advertises the requested endpoint — only when no explicit resolvedProvider
+  // already won above (explicit resolution takes precedence).
+  const syncedEndpointRoute = options.resolvedProvider
+    ? null
+    : await resolveLocalSyncedEndpointRoute(body.model, "embeddings");
+  if (syncedEndpointRoute) {
+    provider = syncedEndpointRoute.provider;
+    resolvedModel = syncedEndpointRoute.model;
+  }
   if (!provider) {
     return errorResponse(
       HTTP_STATUS.BAD_REQUEST,
       `Invalid embedding model: ${body.model}. Use format: provider/model`
     );
   }
 
+  let credentials: ProviderCredentialsResult | null = null;
   let providerConfig: EmbeddingProvider | null =
     options.resolvedProvider ||
     dynamicProviders.find((dp) => dp.id === provider) ||
     getEmbeddingProvider(provider) ||
     null;
   let credentialsProviderId = provider;
 
+  if (syncedEndpointRoute) {
+    credentials = await getProviderCredentials(
+      provider,
+      null,
+      syncedEndpointRoute.connectionIds,
+      syncedEndpointRoute.model
+    );
+    if (!credentials) {
+      return errorResponse(
+        HTTP_STATUS.BAD_REQUEST,
+        `No credentials for embedding provider: ${provider}`
+      );
+    }
+    if ("allRateLimited" in credentials && credentials.allRateLimited) {
+      return unavailableResponse(
+        HTTP_STATUS.RATE_LIMITED,
+        `[${provider}] All accounts rate limited`,
+        credentials.retryAfter,
+        credentials.retryAfterHuman
+      );
+    }
+
+    const providerSpecificData = (credentials as { providerSpecificData?: Record<string, unknown> })
+      .providerSpecificData;
+    const configuredBaseUrl = providerSpecificData?.baseUrl;
+    if (typeof configuredBaseUrl !== "string" || configuredBaseUrl.trim().length === 0) {
+      return errorResponse(
+        HTTP_STATUS.BAD_REQUEST,
+        `No base URL configured for embedding provider: ${provider}`
+      );
+    }
+    let baseUrl = configuredBaseUrl.trim();
+    while (baseUrl.endsWith("/")) baseUrl = baseUrl.slice(0, -1);
+    providerConfig = {
+      id: provider,
+      baseUrl: baseUrl.endsWith("/embeddings") ? baseUrl : `${baseUrl}/embeddings`,
+      authType: "apikey",
+      authHeader: "bearer",
+      models: [],
+    };
+  }
+
   if (!providerConfig) {
     try {
       const allNodes = (await getCachedProviderNodes()) as unknown as EmbeddingProviderNodeRow[];
@@ -226,8 +280,7 @@ export async function createEmbeddingResponse(
     );
   }
 
-  let credentials: ProviderCredentialsResult | null = null;
-  if (providerConfig.authType !== "none") {
+  if (!credentials && providerConfig.authType !== "none") {
     credentials = await getProviderCredentials(credentialsProviderId);
     if (!credentials) {
       return errorResponse(

src/lib/providerModels/managedModelImport.ts

@@ -23,6 +23,7 @@ import {
 } from "@omniroute/open-sse/config/antigravityModelAliases.ts";
 import { filterChatSelectableModels } from "@omniroute/open-sse/services/modelEndpointPolicy.ts";
 import { filterSelectableModels } from "@omniroute/open-sse/services/modelLifecycle.ts";
+import { isSelfHostedChatProvider } from "@/shared/constants/providers";
 
 type JsonRecord = Record<string, unknown>;
 
@@ -253,10 +254,18 @@ export async function importManagedModels({
   const previousSyncedAvailableModels =
     previousSyncedAvailableModelsInput ??
     (await getSyncedAvailableModelsForConnection(providerId, connectionId));
-  const discoveredModels = filterChatSelectableModels(
+  // #11088 (option 1): self-hosted providers keep their non-chat models — chat
+  // filtering happens at read time (resolveLocalSyncedEndpointRoute). Every other
+  // provider keeps the import-time chat filter: the read-time path is gated on
+  // isSelfHostedChatProvider, so dropping it globally leaked image/video models
+  // into OpenAI chat selections (#11271).
+  const selectableModels = filterSelectableModels(
     providerId,
-    filterSelectableModels(providerId, normalizeDiscoveredModels(fetchedModels, providerId))
+    normalizeDiscoveredModels(fetchedModels, providerId)
   );
+  const discoveredModels = isSelfHostedChatProvider(providerId)
+    ? selectableModels
+    : filterChatSelectableModels(providerId, selectableModels);
   const candidateImportedModels = normalizeImportedModels(discoveredModels);
   const importedIds = new Set(candidateImportedModels.map((model) => model.id));
 

src/lib/providerModels/modelDiscovery.ts

@@ -6,7 +6,6 @@ import {
 } from "@/lib/db/models";
 import { CANONICAL_EFFORT_VALUES } from "@/shared/reasoning/effortStandardization";
 import { isObsoleteKiroModelAlias } from "@omniroute/open-sse/services/kiroModels.ts";
-import { filterChatSelectableModels } from "@omniroute/open-sse/services/modelEndpointPolicy.ts";
 import { filterSelectableModels } from "@omniroute/open-sse/services/modelLifecycle.ts";
 
 type JsonRecord = Record<string, unknown>;
@@ -379,9 +378,13 @@ export async function persistDiscoveredModels(
   connectionId: string,
   models: unknown
 ): Promise<SyncedAvailableModel[]> {
-  const normalized = filterChatSelectableModels(
+  // #11088 (option 1): the synced store is endpoint-agnostic — images/embeddings
+  // models must persist so per-connection endpoint routing (#11088) and the
+  // /v1/models catalog can see them. Chat selectability is applied at read time
+  // (auto-pool expansion, chat projections), not at write time.
+  const normalized = filterSelectableModels(
     providerId,
-    filterSelectableModels(providerId, normalizeDiscoveredModels(models, providerId))
+    normalizeDiscoveredModels(models, providerId)
   );
   await replaceSyncedAvailableModelsForConnection(providerId, connectionId, normalized);
   return normalized;

src/lib/providerModels/ollamaCapabilities.ts

@@ -0,0 +1,98 @@
+import { z } from "zod";
+
+type JsonRecord = Record<string, unknown>;
+
+const ollamaShowResponseSchema = z
+  .object({
+    capabilities: z.array(z.string().max(64)).max(32).optional(),
+  })
+  .passthrough();
+
+const OLLAMA_CAPABILITY_TO_ENDPOINT: Readonly<Record<string, string>> = {
+  completion: "chat",
+  embedding: "embeddings",
+  image: "images",
+};
+
+const MAX_CONCURRENT_SHOW_REQUESTS = 4;
+
+function asRecord(value: unknown): JsonRecord {
+  return value && typeof value === "object" && !Array.isArray(value) ? (value as JsonRecord) : {};
+}
+
+export function buildOllamaShowUrl(openAiBaseUrl: string): string {
+  let base = openAiBaseUrl.trim();
+  while (base.endsWith("/")) base = base.slice(0, -1);
+  base = base.replace(/\/(?:chat\/completions|completions|embeddings|images\/generations)$/i, "");
+  if (base.endsWith("/v1")) base = base.slice(0, -3);
+  return `${base}/api/show`;
+}
+
+export function applyOllamaShowCapabilities(model: unknown, showResponse: unknown): JsonRecord {
+  const record = asRecord(model);
+  const parsed = ollamaShowResponseSchema.safeParse(showResponse);
+  if (!parsed.success || !parsed.data.capabilities) return record;
+
+  const capabilities = Array.from(
+    new Set(parsed.data.capabilities.map((value) => value.trim().toLowerCase()).filter(Boolean))
+  );
+  const supportedEndpoints = Array.from(
+    new Set(
+      capabilities
+        .map((capability) => OLLAMA_CAPABILITY_TO_ENDPOINT[capability])
+        .filter((endpoint): endpoint is string => Boolean(endpoint))
+    )
+  );
+  if (supportedEndpoints.length === 0) return record;
+
+  const apiFormat = supportedEndpoints.includes("chat")
+    ? "chat-completions"
+    : supportedEndpoints.includes("embeddings")
+      ? "embeddings"
+      : "images-generations";
+
+  return {
+    ...record,
+    apiFormat,
+    supportedEndpoints,
+    ...(capabilities.includes("vision") ? { supportsVision: true } : {}),
+    ...(capabilities.includes("tools") ? { supportsTools: true } : {}),
+    ...(capabilities.includes("thinking") ? { supportsThinking: true } : {}),
+  };
+}
+
+export async function enrichOllamaModelsWithCapabilities(
+  models: unknown[],
+  fetchShow: (modelId: string) => Promise<unknown | null>
+): Promise<JsonRecord[]> {
+  const output: JsonRecord[] = new Array(models.length);
+  let nextIndex = 0;
+
+  const worker = async () => {
+    while (nextIndex < models.length) {
+      const index = nextIndex++;
+      const model = asRecord(models[index]);
+      const modelId =
+        typeof model.id === "string"
+          ? model.id
+          : typeof model.name === "string"
+            ? model.name
+            : typeof model.model === "string"
+              ? model.model
+              : null;
+      if (!modelId) {
+        output[index] = model;
+        continue;
+      }
+      try {
+        output[index] = applyOllamaShowCapabilities(model, await fetchShow(modelId));
+      } catch {
+        output[index] = model;
+      }
+    }
+  };
+
+  const workerCount = Math.min(MAX_CONCURRENT_SHOW_REQUESTS, Math.max(1, models.length));
+  await Promise.all(Array.from({ length: workerCount }, () => worker()));
+  return output;
+}