Compare commits

...

2 Commits

Author SHA1 Message Date
Diego Rodrigues de Sa e Souza
6b83df1a75 Merge branch 'release/v3.8.50' into fix/11233-lmstudio-embedding-baseurl 2026-08-23 13:18:24 -03:00
Xiangzhe
592a7efc18 fix(embeddings): honor configured LM Studio connection URL via lm-studio alias (#11233)
The dashboard stores LM Studio connections under the hyphenated provider id
"lm-studio", but the embedding registry keys the provider as "lmstudio"
with no alias. As a result, "lm-studio/<model>" embedding requests failed
with a 400 unknown-provider error, and "lmstudio/<model>" requests always
hit the hardcoded http://localhost:1234/v1/embeddings endpoint, ignoring the
baseUrl of the configured connection.

Mirror the ollama-local pattern from #2824/#9225:

- embeddingRegistry: add "lm-studio" -> "lmstudio" to
  EMBEDDING_PROVIDER_ALIASES (registry key unchanged so existing
  "lmstudio/<model>" clients keep working).
- embeddings service: extend the optional keyless-connection hydration to
  lmstudio; getProviderCredentials("lmstudio") already resolves the
  "lm-studio" connection via the provider search pool/alias, and a
  selection/rate-limit failure still proceeds without credentials.
- embeddings handler: apply the same baseUrl override + normalization
  (strip trailing slashes and /v1, /v1/chat/completions, /v1/embeddings
  suffixes, then rebuild <host>/v1/embeddings) to lmstudio, keeping the
  static localhost fallback when no connection or empty baseUrl.

TDD: tests/unit/lmstudio-connection-baseurl-11233.test.ts failed on the
alias, override and service-hydration asserts before the fix and passes
after; ollama-local (#2824) and lmstudio registry (#7601) sibling tests
remain green.
2026-08-23 12:49:59 -03:00
4 changed files with 167 additions and 25 deletions

View File

@@ -413,6 +413,11 @@ export const EMBEDDING_PROVIDERS: Record<string, EmbeddingProvider> = {
const EMBEDDING_PROVIDER_ALIASES: Record<string, string> = {
jina: "jina-ai",
voyage: "voyage-ai",
// The dashboard stores LM Studio connections under the hyphenated provider
// id "lm-studio" while the embedding registry keys the provider "lmstudio"
// (#11233). Alias the dashboard id so "lm-studio/<model>" resolves instead
// of failing with an unknown-provider 400.
"lm-studio": "lmstudio",
};
/** Family name used by clients; Jina's public SKU is omni-small. */

View File

@@ -182,12 +182,8 @@ export async function handleEmbedding({
)
: [];
const nativeModalities = [
...(isJinaNativeEmbeddingInput(body.input)
? collectJinaNativeModalities(body.input)
: []),
...(isGeminiNativeEmbeddingInput(body.input)
? collectGeminiNativeModalities(body.input)
: []),
...(isJinaNativeEmbeddingInput(body.input) ? collectJinaNativeModalities(body.input) : []),
...(isGeminiNativeEmbeddingInput(body.input) ? collectGeminiNativeModalities(body.input) : []),
].filter((modality) => modality !== "text");
if (structuredItems.length > 0 || nativeModalities.length > 0) {
const supportedModalities = getEmbeddingModelModalities(providerConfig, model);
@@ -266,7 +262,10 @@ export async function handleEmbedding({
}
let upstreamUrl = providerConfig.baseUrl;
if (provider === "ollama-local") {
if (provider === "ollama-local" || provider === "lmstudio") {
// Keyless local servers (#2824 ollama-local, #11233 lmstudio): honor the
// configured connection's baseUrl when one was hydrated, and fall back to
// the static localhost registry default otherwise.
const configuredBaseUrl = credentials?.providerSpecificData?.baseUrl;
const rawBaseUrl =
typeof configuredBaseUrl === "string" && configuredBaseUrl.trim().length > 0
@@ -277,11 +276,11 @@ export async function handleEmbedding({
// (CodeQL js/polynomial-redos) since baseUrl is operator-configured
// per-connection data. See open-sse/utils/urlSanitize.ts.
const normalizedBaseUrl = stripTrailingSlashes(rawBaseUrl.trim());
const ollamaHost = normalizedBaseUrl
const localServerHost = normalizedBaseUrl
.replace(/\/v1\/(?:chat\/completions|embeddings)$/i, "")
.replace(/\/api\/chat$/i, "")
.replace(/\/v1$/i, "");
upstreamUrl = `${ollamaHost}/v1/embeddings`;
upstreamUrl = `${localServerHost}/v1/embeddings`;
}
let normalizeProviderResponse:
((data: Record<string, unknown>) => Record<string, unknown>) | null = null;
@@ -321,10 +320,7 @@ export async function handleEmbedding({
// become N embeddings. Native multimodal parts take the same path.
const useGeminiNativeTransport =
providerConfig.structuredInputProtocol === "gemini-embed-content" &&
(isGeminiEmbedding2Family(model) ||
canonicalStructured ||
geminiNative ||
jinaNative);
(isGeminiEmbedding2Family(model) || canonicalStructured || geminiNative || jinaNative);
if (providerConfig.structuredInputProtocol === "jina-v1" && jinaNative && canonicalStructured) {
try {
@@ -462,13 +458,7 @@ export async function handleEmbedding({
// best-effort.
if (connectionId) {
try {
await markAccountUnavailable(
connectionId,
response.status,
errorText,
provider,
model
);
await markAccountUnavailable(connectionId, response.status, errorText, provider, model);
} catch {
// swallow — the upstream error response takes priority
}

View File

@@ -249,11 +249,14 @@ export async function createEmbeddingResponse(
`[${provider}] All ${credentials.expiredCount || 1} connection(s) authentication expired — please reconnect in the dashboard`
);
}
} else if (provider === "ollama-local") {
// Ollama is keyless, but a configured connection can still provide a
// custom local host. Hydrate that optional connection without imposing an
// authentication requirement, then keep the static localhost default when
// no connection exists.
} else if (provider === "ollama-local" || provider === "lmstudio") {
// Ollama and LM Studio are keyless, but a configured connection can still
// provide a custom local host. Hydrate that optional connection without
// imposing an authentication requirement, then keep the static localhost
// default when no connection exists. getProviderCredentials("lmstudio")
// resolves the dashboard's hyphenated "lm-studio" connection via the
// provider search pool/alias (#11233); a selection or rate-limit failure
// must not break the flow — proceed without credentials.
const localCredentials = await getProviderCredentials(credentialsProviderId);
if (
localCredentials &&

View File

@@ -0,0 +1,144 @@
import test from "node:test";
import assert from "node:assert/strict";
import { mkdtempSync, rmSync } from "node:fs";
import { tmpdir } from "node:os";
import { join } from "node:path";
const TEST_DATA_DIR = mkdtempSync(join(tmpdir(), "omniroute-lmstudio-embedding-11233-"));
process.env.DATA_DIR = TEST_DATA_DIR;
const { parseEmbeddingModel } = await import("../../open-sse/config/embeddingRegistry.ts");
const { handleEmbedding } = await import("../../open-sse/handlers/embeddings.ts");
const core = await import("../../src/lib/db/core.ts");
const { createProviderConnection } = await import("../../src/lib/db/providers.ts");
const { createEmbeddingResponse } = await import("../../src/lib/embeddings/service.ts");
test.after(() => {
core.resetDbInstance();
rmSync(TEST_DATA_DIR, { recursive: true, force: true });
});
// Issue #11233: the dashboard stores LM Studio connections under the provider
// id "lm-studio" (hyphenated), but the embedding registry keys the provider as
// "lmstudio" with no alias. Two symptoms resulted:
// 1. "lm-studio/<model>" embedding requests failed with 400 unknown provider.
// 2. "lmstudio/<model>" requests always hit the hardcoded localhost:1234
// endpoint, ignoring the baseUrl of the configured connection.
// The fix mirrors the ollama-local pattern from #2824/#9225: an embedding
// provider alias plus optional (non-auth) connection hydration and the same
// baseUrl normalization in the handler.
test("lm-studio model strings resolve to the lmstudio embedding provider", () => {
assert.deepEqual(parseEmbeddingModel("lm-studio/nomic-embed-text"), {
provider: "lmstudio",
model: "nomic-embed-text",
});
});
test("lmstudio routes to the configured connection baseUrl", async () => {
const originalFetch = globalThis.fetch;
let capturedUrl: string | null = null;
globalThis.fetch = async (url) => {
capturedUrl = String(url);
return new Response(
JSON.stringify({
data: [{ object: "embedding", embedding: [0.1, 0.2], index: 0 }],
usage: { prompt_tokens: 2, total_tokens: 2 },
}),
{ status: 200, headers: { "content-type": "application/json" } }
);
};
try {
const result = await handleEmbedding({
body: { model: "lmstudio/nomic-embed-text", input: "hello" },
resolvedProvider: {
id: "lmstudio",
baseUrl: "http://localhost:1234/v1/embeddings",
authType: "none",
authHeader: "none",
models: [],
},
resolvedModel: "nomic-embed-text",
credentials: {
providerSpecificData: { baseUrl: "http://192.168.1.50:1234/v1" },
},
log: null,
});
assert.equal(result.success, true);
} finally {
globalThis.fetch = originalFetch;
}
assert.equal(capturedUrl, "http://192.168.1.50:1234/v1/embeddings");
});
test("lmstudio keeps the static localhost default without credentials", async () => {
const originalFetch = globalThis.fetch;
let capturedUrl: string | null = null;
globalThis.fetch = async (url) => {
capturedUrl = String(url);
return new Response(
JSON.stringify({
data: [{ object: "embedding", embedding: [0.3, 0.4], index: 0 }],
usage: { prompt_tokens: 2, total_tokens: 2 },
}),
{ status: 200, headers: { "content-type": "application/json" } }
);
};
try {
const result = await handleEmbedding({
body: { model: "lmstudio/nomic-embed-text", input: "hello" },
credentials: null,
log: null,
});
assert.equal(result.success, true);
} finally {
globalThis.fetch = originalFetch;
}
assert.equal(capturedUrl, "http://localhost:1234/v1/embeddings");
});
test("lmstudio service hydrates the lm-studio connection host without requiring a key", async () => {
await createProviderConnection({
provider: "lm-studio",
authType: "none",
name: "LAN LM Studio",
isActive: true,
providerSpecificData: { baseUrl: "http://10.20.0.60:1234/v1/" },
});
const originalFetch = globalThis.fetch;
let captured: { url: string; headers: Record<string, string> } | null = null;
globalThis.fetch = async (url, options = {}) => {
captured = {
url: String(url),
headers: (options.headers as Record<string, string>) || {},
};
return new Response(
JSON.stringify({
data: [{ object: "embedding", embedding: [0.5, 0.6], index: 0 }],
usage: { prompt_tokens: 2, total_tokens: 2 },
}),
{ status: 200, headers: { "content-type": "application/json" } }
);
};
try {
const response = await createEmbeddingResponse({
model: "lm-studio/nomic-embed-text",
input: "hello",
});
assert.equal(response.status, 200);
} finally {
globalThis.fetch = originalFetch;
}
assert.ok(captured);
assert.equal(captured.url, "http://10.20.0.60:1234/v1/embeddings");
assert.equal(captured.headers.Authorization, undefined);
});