JaredStewart/coderlm

Tree-sitter-powered code indexing server that gives LLM agents precise, on-demand access to symbols, implementations, callers, tests, and grep across multi-language projects - so they explore codebases through targeted queries instead of loading everything into context.

What it solves

CodeRLM addresses the problem of LLM agents struggling to navigate large codebases that exceed their context window. Instead of loading entire files or relying on basic keyword searches (glob/grep), it allows agents to recursively explore a codebase as external data, retrieving only the specific symbols, callers, and source code they need.

How it works

The project implements the Recursive Language Model (RLM) pattern using a two-part architecture:

  1. Indexing Server: A Rust-based server uses tree-sitter to parse supported languages and build a symbol table with cross-references for a project's files.
  2. Query Interface: LLM agents (such as Claude Code) use a JSON API to query this index. They can search for symbols by name, list functions within a file, find callers of a specific function, or retrieve exact source code ranges.

Who it’s for

Developers using AI coding assistants (like Claude Code, Cursor, Windsurf, or Copilot) who want their agents to have a more precise and efficient way to explore and understand unfamiliar or large-scale codebases.

Highlights

  • Tree-sitter Indexing: Provides deep structural understanding of code (functions, classes, methods) rather than simple text search.
  • Recursive Exploration: Implements the RLM pattern to decompose complex codebase queries into targeted lookups.
  • Multi-platform Support: Includes a dedicated Claude Code plugin and generators for other AI platforms like Cursor and Windsurf.
  • Broad Language Support: Supports a wide range of languages including Rust, Python, TypeScript, Go, Java, and more via tree-sitter.

Related

  • Project
    thewaltero/mythos-router

    A local CLI tool for verifiable AI-assisted coding that uses Strict Write Discipline to ensure AI file operations match the actual filesystem state, preventing hallucinations and broken writes.

  • Project
    nurettincoban/ai-prd-workflow

    A structured, document-driven workflow for AI coding agents that enforces RFC-based development to ensure project consistency, traceability, and long-term planning.

  • Project
    Meet-Miyani/compose-skill

    Compose Skill (Compose Kit) is a set of seven AI‑agent skills that guide large‑language‑model coding assistants to produce consistent, well‑architected Jetpack Compose and Compose Multiplatform code. It works by first loading a tiny decision tree that routes a prompt to the relevant parts of a project, then loading only the needed reference files and enforcing a house style (MVI, Koin DI, Navigation 3, etc.). The kit can be installed via `npx skills`, GitHub CLI, or agent‑specific plugins, and includes guard scripts for CI. Evaluations on multiple models (Claude, Gemini, DeepSeek, etc.) show substantial rubric score lifts (up to +30.8 points) and no regressions, proving its practical benefit for teams using AI agents for Android/KMP development.

  • Project
    opalinehq/cli

    A CLI tool that syncs local Claude Code and Codex sessions to a shared workspace for team analytics and session visibility.