Radar

Your AI agent's company clerk. Hands off the legwork so the thinking stays sharp.

Radar is a delegation worker for AI coding agents. It lets a high-reasoning host model hand off mechanical I/O work to a cheaper OpenAI-compatible worker model: bulk file reading, codebase search, long-output summarization, and boilerplate drafting.

Radar is built by ContextRail, the standards layer for AI-ready teams.

The rule is simple:

The host agent thinks. Radar handles the paperwork.

Quick Start

Requires Node.js 20 or newer.

pnpm install -g @contextrail/radar
export RADAR_API_KEY=sk-...
radar ask -p package.json README.md -q "What is this package for?"

Tools

| Tool | Use it when... | | ----------- | ----------------------------------------------------- | | ask | You know which files should be read | | search | You need to find relevant files before reading | | summarize | You have a long log, transcript, diff, or test output | | write | You need a draft that matches an existing pattern |

Radar runs as both:

• an MCP server with radar-mcp
• a CLI with radar

Documentation

The full docs are published at radar.contextrail.app. Source lives in docs/:

• Getting Started
• Delegation Model
• Configuration
• MCP Clients
• CLI Reference
• Documentation Updates Recipe

Agent Routing Rules

Installing Radar does not automatically modify a consuming project. Copy the routing guidance into the project where your AI agent works:

• AGENTS.md for general agent instructions.
• .cursor/rules/radar-delegation.mdc for Cursor.
• CLAUDE.md for Claude Code.

Use Radar for paperwork. Keep debugging, architecture, security judgment, safety-critical work, and exact edits with the host agent.

Development

git clone https://github.com/contextrail/radar
cd radar
pnpm install
pnpm run check
pnpm run build
pnpm run docs:dev

Release

Releases are automated with semantic-release. Use Conventional Commits, merge to main, and the release workflow publishes to npm and creates a GitHub Release.

License

MIT