Understand any codebase. Ask CodeLensa.
Trace behavior, find symbols, and investigate architecture with cited answers that jump directly to the code.
Ask CodeLensa
How does hybrid search rank code?
The retrieval pipeline combines lexical and vector candidates with reciprocal-rank fusion, then expands high-confidence symbols through the dependency graph.
hybrid.ts:84–112
Results are deduplicated by chunk identity before the final relevance sort.
For investigative queries, neighboring callers and callees are added within a bounded graph depth.
Repository-first intelligence
From Git URL to explainable answers.
Connect
Add a public or private GitHub repository.
Index
AST-aware chunks, symbols, embeddings, and relationships.
Understand
Ask questions with answers grounded in exact source lines.
Designed for trust
Every answer has a path back to source.
CodeLensa keeps retrieval visible: strategy, files, symbols, timing, and exact line citations are first-class product data.
Hybrid retrieval
Vector, lexical, graph, and reranking strategies.
Code graph
Explore callers, callees, imports, and symbol relationships.
Commit-aware
Every index is tied to branch and commit metadata.
Evaluation built in
Measure hit rate, citation quality, latency, and confidence.