Summary

Gemini provider: Auto-detects AIza key prefix → routes to Google's OpenAI-compatible embedding endpoint (gemini-embedding-001, free tier)
LAT_LLM_ENDPOINT: Generic escape hatch for any OpenAI-compatible embedding server (Ollama, LM Studio, vLLM). Takes highest priority when set. Pair with LAT_LLM_MODEL to override model name.
dimensions in request body: Ensures providers that default to higher dimensions (Gemini → 3072) truncate to the expected 1536. Also correct for OpenAI.

Changes

File	Change
`src/search/provider.ts`	Add `gemini` provider, `customProvider()` function, `LAT_LLM_ENDPOINT`/`LAT_LLM_MODEL` env var support in `detectProvider()`
`src/search/embeddings.ts`	Pass `dimensions` in embedding request body
`tests/search.test.ts`	6 new tests: Gemini detection, endpoint override, trailing slash handling, custom provider

Complements #36 — this handles the "custom API endpoint" tier, #36 handles "bundled local inference"
Follows @jeffmm's proposed priority chain from #36 discussion
Gemini uses Google's OpenAI-compatible endpoint, so no format changes needed
91 additions, 1 deletion. All existing tests pass.

npm run typecheck passes
npx vitest run tests/search.test.ts — all 14 tests pass (8 existing + 6 new)
Full test suite: 158 passed, 2 pre-existing pnpm not found failures unchanged
Manually verified Gemini semantic search works with LAT_LLM_KEY=$GEMINI_API_KEY lat search "query"

a76ecc0Mar 25

vercel[bot]7mo ago

@kgray-wasteology is attempting to deploy a commit to the Yury Selivanov's projects Team on Vercel.

A member of the Team first needs to authorize it.

grumdreviewed6mo ago