fix(cli): resolve embedding provider explicitly and surface lane status in sl search (#192)

* feat(cli): add tryUseManagedLocalEmbeddingsDaemon for read-only callers

* feat(cli): add resolveProjectEmbeddingProvider helper

* fix(cli): wire sl search through resolveProjectEmbeddingProvider so semantic lane works

* fix(cli): wire wiki/knowledge search through resolveProjectEmbeddingProvider

* feat(cli): surface embeddings-unavailable status when sl search returns empty

* refactor(cli): route admin reindex through resolveProjectEmbeddingProvider

* refactor: pass embeddingProvider into ingest/scan instead of resolving inside @ktx/context

* refactor(mcp): resolve embedding provider in CLI factory, pass into context ports

* refactor(context): delete MANAGED_SENTENCE_TRANSFORMERS_BASE_URL sentinel

* refactor(cli): delete sentinel-based managed-embeddings indirection

* chore: scrub stale managed-embeddings sentinel references from tests and smoke script

* chore: unexport unused EmbeddingResolutionMode alias

* fix(cli): force pathPrefix="" when targeting the managed embeddings daemon

The managed daemon serves /embeddings/compute directly. The default
pathPrefix in @ktx/llm is /api, so omitting sentenceTransformers from
ktx.yaml produced /api/embeddings/compute -> 404. The resolver now
sets pathPrefix='' explicitly when wiring the managed daemon URL,
matching what the daemon actually exposes.
This commit is contained in:
Andrey Avtomonov 2026-05-21 02:21:22 +02:00 committed by GitHub
parent 56a967278a
commit 9d92c79988
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
36 changed files with 750 additions and 442 deletions

View file

@ -8,7 +8,6 @@ import { noopLogger, SessionWorktreeService } from '../core/index.js';
import type { KtxSemanticLayerComputePort } from '../daemon/index.js';
import {
createRuntimeToolDescriptorFromAiTool,
createLocalKtxEmbeddingProviderFromConfig,
createLocalKtxLlmRuntimeFromConfig,
KtxIngestEmbeddingPortAdapter,
RuntimeAgentRunner,
@ -16,6 +15,7 @@ import {
type KtxLlmRuntimePort,
type KtxRuntimeToolSet,
} from '../llm/index.js';
import type { KtxEmbeddingProvider } from '@ktx/llm';
import type { KtxLocalProject } from '../project/index.js';
import { ktxLocalStateDbPath } from '../project/index.js';
import { PromptService } from '../prompts/index.js';
@ -114,6 +114,7 @@ export interface CreateLocalBundleIngestRuntimeOptions {
queryExecutor?: KtxSqlQueryExecutorPort;
jobIdFactory?: () => string;
logger?: KtxLogger;
embeddingProvider?: KtxEmbeddingProvider | null;
}
export interface LocalBundleIngestRuntime {
@ -669,7 +670,7 @@ export function createLocalBundleIngestRuntime(
mkdirSync(join(options.project.projectDir, '.ktx/cache/local-ingest'), { recursive: true });
const store = new SqliteBundleIngestStore({ dbPath });
const contextStore = new SqliteContextEvidenceStore({ dbPath });
const embeddingProvider = createLocalKtxEmbeddingProviderFromConfig(options.project.config.ingest.embeddings);
const embeddingProvider = options.embeddingProvider ?? null;
const embedding = embeddingProvider ? new KtxIngestEmbeddingPortAdapter(embeddingProvider) : new NoopEmbeddingPort();
const connections = new LocalConnectionCatalog(options.project, options.queryExecutor);
const rootFileStore = options.project.fileStore;

View file

@ -34,6 +34,7 @@ export interface RunLocalIngestOptions {
semanticLayerCompute?: KtxSemanticLayerComputePort;
queryExecutor?: KtxSqlQueryExecutorPort;
logger?: KtxLogger;
embeddingProvider?: import('@ktx/llm').KtxEmbeddingProvider | null;
}
export interface LocalIngestMcpOptions
@ -172,6 +173,7 @@ async function runScheduledPullJob(options: {
semanticLayerCompute?: KtxSemanticLayerComputePort;
queryExecutor?: KtxSqlQueryExecutorPort;
logger?: KtxLogger;
embeddingProvider?: import('@ktx/llm').KtxEmbeddingProvider | null;
}): Promise<LocalIngestResult> {
const runtime = createLocalBundleIngestRuntime(options);
const jobId = options.jobId ?? runtime.nextJobId();
@ -225,6 +227,7 @@ export async function runLocalIngest(options: RunLocalIngestOptions): Promise<Lo
semanticLayerCompute: options.semanticLayerCompute,
queryExecutor: options.queryExecutor,
logger: options.logger,
embeddingProvider: options.embeddingProvider,
});
}
@ -403,6 +406,7 @@ export async function runLocalMetabaseIngest(
semanticLayerCompute: options.semanticLayerCompute,
queryExecutor: options.queryExecutor,
logger: options.logger,
embeddingProvider: options.embeddingProvider,
});
} catch (error) {
child = await recordLocalMetabaseChildFailure({

View file

@ -37,7 +37,6 @@ export {
summarizeKtxLlmDebugRequest,
} from './debug-request-recorder.js';
export {
MANAGED_SENTENCE_TRANSFORMERS_BASE_URL,
createLocalKtxEmbeddingProviderFromConfig,
createLocalKtxLlmProviderFromConfig,
createLocalKtxLlmRuntimeFromConfig,

View file

@ -5,7 +5,6 @@ import {
type KtxProjectLlmConfig,
} from '../project/config.js';
import {
MANAGED_SENTENCE_TRANSFORMERS_BASE_URL,
createLocalKtxEmbeddingProviderFromConfig,
createLocalKtxLlmProviderFromConfig,
resolveLocalKtxEmbeddingConfig,
@ -151,13 +150,13 @@ describe('local KTX embedding config', () => {
});
});
it('returns null when sentence-transformers base_url is still the unresolved managed sentinel', () => {
it('returns null when sentence-transformers has no base_url (managed daemon delegation)', () => {
const config: KtxProjectEmbeddingConfig = {
backend: 'sentence-transformers',
model: 'all-MiniLM-L6-v2',
dimensions: 384,
sentenceTransformers: {
base_url: MANAGED_SENTENCE_TRANSFORMERS_BASE_URL,
base_url: '',
pathPrefix: '',
},
};

View file

@ -22,8 +22,6 @@ interface LocalConfigDeps {
createAiSdkRuntime?: (deps: { llmProvider: KtxLlmProvider }) => KtxLlmRuntimePort;
}
export const MANAGED_SENTENCE_TRANSFORMERS_BASE_URL = 'managed:local-embeddings';
function resolveOptional(value: string | undefined, env: NodeJS.ProcessEnv): string | undefined {
return resolveKtxConfigReference(value, env) || undefined;
}
@ -149,7 +147,7 @@ export function resolveLocalKtxEmbeddingConfig(
}
if (config.backend === 'sentence-transformers') {
const baseURL = config.sentenceTransformers?.base_url;
if (!baseURL || baseURL === MANAGED_SENTENCE_TRANSFORMERS_BASE_URL) {
if (!baseURL) {
return null;
}
return {

View file

@ -174,7 +174,7 @@ describe('createLocalProjectMcpContextPorts', () => {
driver: 'postgres',
url: 'env:DATABASE_URL',
};
const ports = createLocalProjectMcpContextPorts(project);
const ports = createLocalProjectMcpContextPorts(project, { embeddingService: null });
expect(Object.keys(ports).sort()).toEqual([
'connections',
@ -216,6 +216,7 @@ describe('createLocalProjectMcpContextPorts', () => {
localScan: {
createConnector,
},
embeddingService: null,
});
expect(Object.keys(ports).sort()).toContain('sqlExecution');
@ -269,6 +270,7 @@ describe('createLocalProjectMcpContextPorts', () => {
localScan: {
createConnector,
},
embeddingService: null,
});
const result = await ports.sqlExecution?.execute(
@ -313,6 +315,7 @@ describe('createLocalProjectMcpContextPorts', () => {
localScan: {
createConnector: vi.fn(async () => connector),
},
embeddingService: null,
});
await expect(
@ -332,7 +335,7 @@ describe('createLocalProjectMcpContextPorts', () => {
url: 'env:DATABASE_URL',
};
await seedScanReport(project.projectDir);
const ports = createLocalProjectMcpContextPorts(project);
const ports = createLocalProjectMcpContextPorts(project, { embeddingService: null });
await expect(
ports.entityDetails?.read({
@ -358,7 +361,7 @@ describe('createLocalProjectMcpContextPorts', () => {
driver: 'postgres',
url: 'env:DATABASE_URL',
};
const ports = createLocalProjectMcpContextPorts(project);
const ports = createLocalProjectMcpContextPorts(project, { embeddingService: null });
await expect(
ports.entityDetails?.read({
@ -411,7 +414,7 @@ describe('createLocalProjectMcpContextPorts', () => {
'Seed dictionary profile',
);
const ports = createLocalProjectMcpContextPorts(project);
const ports = createLocalProjectMcpContextPorts(project, { embeddingService: null });
await expect(ports.dictionarySearch?.search({ values: ['paid'] })).resolves.toMatchObject({
searched: [{ connectionId: 'warehouse', status: 'ready' }],
@ -432,7 +435,7 @@ describe('createLocalProjectMcpContextPorts', () => {
url: 'env:DATABASE_URL',
};
const ports = createLocalProjectMcpContextPorts(project);
const ports = createLocalProjectMcpContextPorts(project, { embeddingService: null });
await expect(ports.dictionarySearch?.search({ values: ['paid'] })).resolves.toEqual({
searched: [
@ -601,7 +604,7 @@ describe('createLocalProjectMcpContextPorts', () => {
'seed scan report',
);
const ports = createLocalProjectMcpContextPorts(project);
const ports = createLocalProjectMcpContextPorts(project, { embeddingService: null });
const results = await ports.discover?.search({ query: 'paid orders', connectionId: 'warehouse', limit: 10 });
expect(results).toEqual(
@ -635,7 +638,7 @@ describe('createLocalProjectMcpContextPorts', () => {
'ktx@example.com',
'Seed wiki',
);
const ports = createLocalProjectMcpContextPorts(project);
const ports = createLocalProjectMcpContextPorts(project, { embeddingService: null });
await expect(ports.knowledge?.read({ userId: 'local-user', key: 'revenue' })).resolves.toMatchObject({
key: 'revenue',
@ -680,7 +683,7 @@ describe('createLocalProjectMcpContextPorts', () => {
'',
].join('\n'),
});
const ports = createLocalProjectMcpContextPorts(project);
const ports = createLocalProjectMcpContextPorts(project, { embeddingService: null });
await expect(
ports.semanticLayer?.readSource({ connectionId: 'warehouse', sourceName: 'orders' }),
@ -692,7 +695,7 @@ describe('createLocalProjectMcpContextPorts', () => {
it('rejects path traversal keys before touching the project directory', async () => {
const project = await initKtxProject({ projectDir: tempDir });
const ports = createLocalProjectMcpContextPorts(project);
const ports = createLocalProjectMcpContextPorts(project, { embeddingService: null });
await expect(
ports.knowledge?.read({
@ -746,7 +749,7 @@ describe('createLocalProjectMcpContextPorts', () => {
})),
generateSources: vi.fn(),
};
const ports = createLocalProjectMcpContextPorts(project, { semanticLayerCompute });
const ports = createLocalProjectMcpContextPorts(project, { semanticLayerCompute, embeddingService: null });
await expect(
ports.semanticLayer?.query({
@ -817,6 +820,7 @@ describe('createLocalProjectMcpContextPorts', () => {
const ports = createLocalProjectMcpContextPorts(project, {
semanticLayerCompute: compute,
queryExecutor,
embeddingService: null,
});
const result = await ports.semanticLayer?.query({

View file

@ -1,7 +1,6 @@
import { type KtxSqlQueryExecutorPort, localConnectionInfoFromConfig } from '../connections/index.js';
import type { KtxEmbeddingPort } from '../core/index.js';
import type { KtxSemanticLayerComputePort } from '../daemon/index.js';
import { createLocalKtxEmbeddingProviderFromConfig, KtxIngestEmbeddingPortAdapter } from '../llm/index.js';
import type { KtxLocalProject } from '../project/index.js';
import { createKtxEntityDetailsService, type KtxScanConnector, type LocalScanMcpOptions } from '../scan/index.js';
import { createKtxDiscoverDataService } from '../search/index.js';
@ -15,7 +14,7 @@ interface CreateLocalProjectMcpContextPortsOptions {
queryExecutor?: KtxSqlQueryExecutorPort;
sqlAnalysis?: SqlAnalysisPort;
localScan?: LocalScanMcpOptions;
embeddingService?: KtxEmbeddingPort | null;
embeddingService: KtxEmbeddingPort | null;
}
function dialectForDriver(driver: string | undefined): string {
@ -133,12 +132,9 @@ async function executeValidatedReadOnlySql(
export function createLocalProjectMcpContextPorts(
project: KtxLocalProject,
options: CreateLocalProjectMcpContextPortsOptions = {},
options: CreateLocalProjectMcpContextPortsOptions,
): KtxMcpContextPorts {
const configuredEmbeddingProvider = createLocalKtxEmbeddingProviderFromConfig(project.config.ingest.embeddings);
const embeddingService =
options.embeddingService ??
(configuredEmbeddingProvider ? new KtxIngestEmbeddingPortAdapter(configuredEmbeddingProvider) : null);
const embeddingService = options.embeddingService;
const ports: KtxMcpContextPorts = {
connections: {
async list() {

View file

@ -143,7 +143,6 @@ describe('@ktx/context package exports', () => {
expect(root.assertSearchBackendConformanceCase).toBeTypeOf('function');
expect(root.assertSearchBackendCapabilities).toBeTypeOf('function');
expect(root.createLocalKtxEmbeddingProviderFromConfig).toBeTypeOf('function');
expect(root.MANAGED_SENTENCE_TRANSFORMERS_BASE_URL).toBe('managed:local-embeddings');
expect(agent).toBeDefined();
expect(agent.AgentRunnerService).toBeTypeOf('function');
expect(root.AgentRunnerService).toBeTypeOf('function');

View file

@ -36,7 +36,7 @@ const vertexProviderSchema = z
const sentenceTransformersSchema = z
.strictObject({
base_url: z.string().default('').describe('Base URL of the sentence-transformers HTTP server. Empty string uses the managed local runtime.'),
base_url: z.string().default('').describe('Base URL of the sentence-transformers HTTP server. Leave empty (or omit) to use the project-managed local daemon.'),
pathPrefix: z.string().optional().describe('Optional URL path prefix prepended to embedding requests.'),
})
.describe('Sentence-transformers embedding server configuration.');

View file

@ -813,16 +813,16 @@ describe('local scan enrichment', () => {
}
});
it('resolves gateway LLM providers and OpenAI embeddings from local scan config', () => {
it('resolves gateway LLM providers and passes injected embedding provider through to scan enrichment', () => {
const createKtxLlmProvider = vi.fn(() => ({
getModel: vi.fn().mockReturnValue({ modelId: 'provider/language-model', provider: 'gateway' }),
}));
const createKtxEmbeddingProvider = vi.fn(() => ({
const embeddingProvider = {
dimensions: 1536,
maxBatchSize: 8,
embed: vi.fn(),
[['embed', 'Many'].join('')]: vi.fn(),
}));
};
const providers = createLocalScanEnrichmentProvidersFromConfig(
{
@ -844,8 +844,8 @@ describe('local scan enrichment', () => {
},
{
createKtxLlmProvider: createKtxLlmProvider as any,
createKtxEmbeddingProvider: createKtxEmbeddingProvider as any,
env: { OPENAI_API_KEY: 'openai-key' }, // pragma: allowlist secret
embeddingProvider: embeddingProvider as any,
},
);
@ -854,8 +854,5 @@ describe('local scan enrichment', () => {
expect(createKtxLlmProvider).toHaveBeenCalledWith(
expect.objectContaining({ backend: 'gateway', modelSlots: { default: 'provider/language-model' } }),
);
expect(createKtxEmbeddingProvider).toHaveBeenCalledWith(
expect.objectContaining({ backend: 'openai', model: 'provider/embedding-model' }),
);
});
});

View file

@ -1,4 +1,4 @@
import type { createKtxEmbeddingProvider, createKtxLlmProvider } from '@ktx/llm';
import type { createKtxEmbeddingProvider, createKtxLlmProvider, KtxEmbeddingProvider } from '@ktx/llm';
import {
createDefaultLocalIngestAdapters,
getLocalStageOnlyIngestStatus,
@ -6,11 +6,7 @@ import {
runLocalStageOnlyIngest,
type SourceAdapter,
} from '../ingest/index.js';
import {
createLocalKtxEmbeddingProviderFromConfig,
createLocalKtxLlmRuntimeFromConfig,
KtxScanEmbeddingPortAdapter,
} from '../llm/index.js';
import { createLocalKtxLlmRuntimeFromConfig, KtxScanEmbeddingPortAdapter } from '../llm/index.js';
import type { KtxProjectLlmConfig, KtxScanEnrichmentConfig, KtxScanRelationshipConfig } from '../project/config.js';
import type { KtxLocalProject } from '../project/index.js';
import { ktxLocalStateDbPath } from '../project/local-state-db.js';
@ -55,6 +51,7 @@ export interface RunLocalScanOptions {
enrichmentProviders?: KtxLocalScanEnrichmentProviders | null;
enrichmentStateStore?: SqliteLocalScanEnrichmentStateStore | null;
progress?: KtxProgressPort;
embeddingProvider?: KtxEmbeddingProvider | null;
}
export interface LocalScanRunResult {
@ -152,6 +149,7 @@ interface LocalScanEnrichmentProviderDeps {
createKtxEmbeddingProvider?: typeof createKtxEmbeddingProvider;
env?: NodeJS.ProcessEnv;
projectDir?: string;
embeddingProvider?: KtxEmbeddingProvider | null;
}
export function createLocalScanEnrichmentProvidersFromConfig(
@ -171,7 +169,7 @@ export function createLocalScanEnrichmentProvidersFromConfig(
...deps,
projectDir: deps.projectDir,
});
const embeddingProvider = createLocalKtxEmbeddingProviderFromConfig(config.embeddings, deps);
const embeddingProvider = deps.embeddingProvider ?? null;
if (!llmRuntime || !embeddingProvider) {
return null;
}
@ -371,6 +369,7 @@ export async function runLocalScan(options: RunLocalScanOptions): Promise<LocalS
? options.enrichmentProviders
: createLocalScanEnrichmentProvidersFromConfig(options.project.config.scan.enrichment, options.project.config.llm, {
projectDir: options.project.projectDir,
embeddingProvider: options.embeddingProvider ?? null,
})
: null;