AI-ready codebase indexer. Scans your project and generates a minimal, structured context file (AGENTS.md / CLAUDE.md) so any AI agent — Claude Code, GPT-4, Gemini, or your own — instantly understands your codebase.
codexa init # scan project → generate AGENTS.md
codexa update # re-scan and refresh context
codexa stats # show project summary
codexa configure # interactive setup wizard
codexa index # build vector semantic index
codexa search "query" # semantic search over codebaseAI agents work better when they know:
- What language / stack the project uses
- Where entry points and key files are
- What the coding conventions are
- What they should and shouldn't touch
Writing this by hand is tedious. codexa does it in one command.
For large projects, codexa can index your code into a vector database so agents can find relevant context by semantic similarity — without re-scanning the whole project every time.
Unlike simple encoders, codexa uses a language-aware chunker (chunker.rs) that respects code boundaries (functions, classes, blocks) to keep semantic units together.
Running codexa configure creates codexa.toml in your project. You can enable SQLite or PostgreSQL backends for vector storage.
src/
main.rs # CLI entry point and commands
config.rs # codexa.toml config + wizard
indexer.rs # file scanner, language detection
generator.rs # markdown + json output
vector.rs # vector store + embeddings integration
chunker.rs # semantic code splitting logic (Smart Chunker)
PRs are welcome. The codebase is modular and designed to be easily extensible.
To add a new language — edit the detect_language() match in indexer.rs.
To improve chunking — look into chunker.rs.