CoderLM addresses the challenge of LLM agents efficiently exploring large codebases. It provides a Rust-based indexing server that leverages tree-sitter to expose a precise JSON API, enabling AI agents to query for specific code elements like symbols, implementations, and callers on demand. This avoids loading entire projects into context, facilitating deeper code understanding and navigation for AI tools.

How It Works

The project implements the Recursive Language Model (RLM) pattern, treating codebases as external data. A Rust server indexes projects using tree-sitter for supported languages, building a detailed symbol table with cross-references. Agents interact via a JSON API, requesting exact code snippets, function implementations, or symbol relationships. This approach replaces inefficient glob/grep/read cycles with targeted, index-backed lookups, offering significant performance and accuracy gains for AI code analysis.

Quick Start & Requirements

• Prerequisites: Rust toolchain (rustup recommended), Python 3 (stdlib only).
• Build & Run Server: Clone the repo, navigate to server/, run cargo build --release, then cargo run --release -- serve or use ./coderlm-daemon.sh.
• AI Integration:
  • Claude Code: Add marketplace source (claude /plugin marketplace add JaredStewart/coderlm) then install (claude plugin install coderlm).
  • Other Platforms: Use uv tool install coderlm --from git+https://github.com/JaredStewart/coderlm.git followed by coderlm --platform <platform>, or run directly via uvx.
• CLI: Python scripts are available for direct interaction (plugin/skills/coderlm/scripts/coderlm_cli.py).
• Docs: API reference available at server/REPL_to_API.md.

Highlighted Details

• Multi-Language Indexing: Supports Rust, Python, TypeScript, JavaScript, Go, Java, Scala via tree-sitter for full symbol tables. SQL uses regex fallbacks.
• Precise Context Retrieval: API allows querying for symbols, implementations, callers, tests, and grep patterns, returning exact source code.
• AI Agent Workflow: Integrated Claude Code plugin provides a structured /coderlm skill and hooks for automatic session initialization and prompt guidance.
• Platform Agnostic Generation: Tools exist to generate configurations for various AI platforms like Cursor, Windsurf, Copilot, Gemini, and Codex.

Maintenance & Community

No specific details regarding maintainers, community channels (e.g., Discord, Slack), or project roadmap were found in the provided README content.

Licensing & Compatibility

• License: MIT.
• Compatibility: The permissive MIT license allows for broad adoption, including use within commercial and closed-source applications.

Limitations & Caveats

Symbol resolution for SQL relies on regex, which may be less comprehensive than tree-sitter's capabilities for other supported languages.