I built codix because AI coding agents waste a lot of tokens grepping and guessing their way through codebases. They search for symbols by name, get imprecise results back, and sometimes just hallucinate definitions that don't exist.
codix indexes your code into a local SQLite database using tree-sitter, then provides precise symbol lookups, reference finding, and relationship queries (callers, callees, implementations, supertypes).
It also supports refactoring: `codix rename` finds every occurrence of a symbol across the codebase using AST analysis (not text search), shows you a dry-run diff, and applies it with --apply.
cd your-project
codix init
codix find UserService
codix callers save
codix impls Repository
codix rename save persist # dry-run
codix rename save persist --apply # apply
It supports Go, Java, JavaScript/TypeScript, Python, and Rust. The index auto-refreshes on every query, so it's always current.
The output is designed to be token-efficient for agents, but it works fine for humans too. Adding it to your agent is one line in CLAUDE.md or AGENTS.md.
MIT licensed, written in Rust, ~5k lines. Contributions welcome — especially for new language plugins.
codix indexes your code into a local SQLite database using tree-sitter, then provides precise symbol lookups, reference finding, and relationship queries (callers, callees, implementations, supertypes).
It also supports refactoring: `codix rename` finds every occurrence of a symbol across the codebase using AST analysis (not text search), shows you a dry-run diff, and applies it with --apply.
It supports Go, Java, JavaScript/TypeScript, Python, and Rust. The index auto-refreshes on every query, so it's always current.
The output is designed to be token-efficient for agents, but it works fine for humans too. Adding it to your agent is one line in CLAUDE.md or AGENTS.md.
MIT licensed, written in Rust, ~5k lines. Contributions welcome — especially for new language plugins.