feat(docs): added vector search

[waleedlatif1]

Summary

• added vector search to docs

Type of Change

• New feature

Testing

Tested manually

Checklist

• Code follows project style guidelines
• Self-reviewed my changes
• Tests added/updated and passing
• No new warnings introduced
• I confirm that I have read and agree to the terms outlined in the Contributor License Agreement (CLA)

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Review	Updated (UTC)
docs	Ready	Preview, Comment	Dec 25, 2025 6:34pm

[greptile-apps]

Greptile Summary

Replaced fumadocs client-side search with a hybrid vector + keyword search system using PostgreSQL and OpenAI embeddings. Vector search (cosine similarity with pgvector) is enabled for English queries, while all languages get keyword search using PostgreSQL full-text search with locale-specific text search configurations. Results are interleaved to combine semantic and lexical matching.

Key Changes:

• Implemented hybrid search combining vector embeddings (cosine similarity) and PostgreSQL full-text search
• Added OpenAI text-embedding-3-small integration for query embeddings with proper error handling
• Search queries against docs_embeddings table with both vector and tsvector columns
• Results are deduplicated and interleaved using alternating selection from both result sets

Notes:

• The chunkTextTsv column is pre-generated with English tokenization (schema.ts:1306), so keyword search for non-English locales may have reduced effectiveness despite using locale-specific tsconfig in queries
• Vector search is English-only by design; other languages fall back to keyword-only search

Confidence Score: 4/5

• Safe to merge with proper testing of search functionality across different locales
• Implementation is solid with proper error handling and validation. Score reflects the locale/tsvector language configuration mismatch noted in previous review that limits non-English search effectiveness, though this appears to be a known design trade-off rather than a critical bug
• No files require special attention - validation issues from previous review have been addressed

Important Files Changed

Filename	Overview
apps/docs/lib/embeddings.ts	Added OpenAI embedding generation with proper validation for empty arrays
apps/docs/lib/db.ts	Simple re-export of database connection and schema from shared package
apps/docs/package.json	Added required dependencies for database operations and vector search
apps/docs/app/api/search/route.ts	Replaced fumadocs search with hybrid vector+keyword search; locale parameter retrieved but not used in search queries

Sequence Diagram

sequenceDiagram
    participant Client
    participant SearchAPI as GET /api/search
    participant OpenAI as OpenAI API
    participant DB as PostgreSQL
    
    Client->>SearchAPI: "GET ?query=...&locale=...&limit=10"
    
    alt Empty query
        SearchAPI->>Client: "Empty array"
    end
    
    SearchAPI->>SearchAPI: "Parse params"
    SearchAPI->>SearchAPI: "Compute tsConfig from locale"
    
    alt locale is en
        SearchAPI->>OpenAI: "POST /v1/embeddings"
        OpenAI->>SearchAPI: "embedding vector (1536 dims)"
        SearchAPI->>DB: "Vector search (cosine similarity)"
        DB->>SearchAPI: "vectorResults"
    end
    
    SearchAPI->>DB: "Keyword search (ts_rank)"
    DB->>SearchAPI: "keywordResults"
    
    SearchAPI->>SearchAPI: "Interleave & deduplicate"
    SearchAPI->>SearchAPI: "Map to format"
    SearchAPI->>Client: "JSON search results"

Loading

[greptile-apps]

_{4 files reviewed, 2 comments}

_{Edit Code Review Agent Settings | Greptile}

[waleedlatif1]

@greptile