Search by keywords, concepts or meaning. Find what you meant, not just what you typed. Retrieve it in Compact Form for resource-minded LLM injection or in User-Friendly mode for detailed explanations.
You don't always remember the exact words you used. You remember the idea — the problem you were solving, the concept you explored, the solution that clicked. ContextBridge's semantic search understands the meaning behind your query, not just the literal words.
Search for "authentication bug in my API" and find conversations that talked about JWT errors, token refresh failures, or OAuth issues — even if those exact words never appeared in your query.
Results span both your saved AI conversations, your synced codebase and files uploaded to or created with your AIs. Find the conversation where you discussed a contract, a disclosure, a research piece, a code function — and the file or function itself — in a single search.
Search happens in seconds across your entire knowledge base. No waiting, no indexing delays — results appear as you type.
Access your entire knowledge base without leaving your current AI conversation. The search modal opens directly in the page — find what you need, copy it, and continue without switching tabs.
ContextBridge does not rely on a single retrieval method. It combines proprietary semantic vector search, BM25 keyword matching, and entity detection, fusing results using Reciprocal Rank Fusion (RRF) for more accurate and robust retrieval than any single method alone.
Query text is embedded using a 1536-dimension model. Cosine similarity search runs against pre-computed chunk embeddings stored in Supabase. This captures semantic proximity — meaning over exact wording.
PostgreSQL full-text search with GIN indexes provides BM25-style keyword matching. This excels on precise terms — function names, error codes, identifiers — where semantic embeddings can be too diffuse. A proprietary normalization step prepares scores for fusion.
Vector and keyword result sets are merged using RRF, a rank-based fusion approach robust to score distribution differences between retrieval methods. This consistently outperforms score-weighted approaches in retrieval benchmarks.
ContextBridge applies proprietary post-processing to eliminate common retrieval artifacts — including score inflation from indirect associations — ensuring results reflect genuine relevance rather than coincidental co-occurrence. Contact [email protected] for details.
The current hybrid approach is already a significant step beyond simple keyword search. The next frontier is search that learns from your usage patterns — understanding which results you actually use, which queries repeatedly fail to find what you meant, and adapting retrieval weights accordingly. Combined with entity-aware intent detection and expanded source coverage, the goal is a search layer precise enough that finding something becomes as natural as remembering it.
Install ContextBridge and search your AI conversations by meaning.