GitHub radar
Give your AI coding agent a map of your code and team memory
Trail-code is a layer on top of your AI coding agent (Claude Code, Cursor, Codex) that builds a map of your code and remembers your team's decisions, so the agent guesses less and finds the right code faster.
It's a tool that scans your code locally to map functions and who calls them, and saves what you and your agent work out together as reusable notes.
When it helps
Useful if you work on the same project with an AI agent every day, especially in a team. Skip it for one-off scripts or a project you touch once a month.
Pros
- Runs offline, no AI models or internet needed to build the map
- Agent solved 33 of 50 tasks instead of 27 — with the map
- 46% fewer tool calls and 60% less time per run
- Team decisions are saved as files and committed to the repo
Cons
- Deep code summaries still need your own AI model and API key
- Adds another layer on top of an already-set-up agent
How to set it up — step by step
- 1Check you have Node.js 20 or newer installed — no GPU needed.
- 2Open a terminal in your project folder.
- 3Send the agent the text from the block below.
- 4Check the result: what code map and integrations the agent set up.
- 5Commit the setup: git add .claude .mcp.json .trail .gitignore .ignore && git commit -m "set up Trail-code".
- 6Work as usual in Claude Code — the agent keeps the map fresh and reuses past decisions.
Text for your agent
Copy this and send it to your agent — Claude Code, Codex, any of them:
Install Trail-code from https://github.com/trailhq/Trail-Code in my current repo: run `npm install -g @trailhq/trail-code && trail init`, wire it into Claude Code, and show me what code map and integrations it set up.
Runs on a regular laptop, no GPU needed — just Node.js 20 or newer.
Open on GitHub▌ More finds