agenttrail

Overview of agenttrail

What it is

Agenttrail is local observability for AI coding agents. It watches coding agents' available activity and turns it into a live view of their work. It does not run the agents, assign their work, or decide that a task is finished. The project provides two views: Agenttrail Map and Agenttrail Kitchen. Map shows project structure, progress, and which components are changing. Kitchen visualizes tasks, role contributions, and completed work as a shared cooking scene.

Key features

• Two independent views: Map (component map, file activity, session trails) and Kitchen (chefs, order tickets, cooking, deliveries)
• Kitchen reads available local logs from Codex CLI, Claude Code, and Cursor agents automatically - Map combines declared intent in PLAN.md with observed file changes - No API key, copied files, or new agent session required for basic observation - Live file tree, dependency links, and session trails in Map - Kitchen can run with --example for a scripted demo without a real agent - Experimental preview; npm package includes graphics and fonts, no build step needed

Use cases

• See which part of a project is changing and where an agent needs attention - Follow current tasks, role contributions, and completed work in a visual kitchen scene - Use with coding agents such as Codex, Claude Code, and Cursor - Choose Map for a project overview and Kitchen to follow work as a shared cooking scene

Getting started

You need Node.js 20+, a browser with WebGL, and a local project folder. Run the following commands in the repo you want to watch: bash npx agenttrail-kitchen . The browser opens at localhost:4780 (or next free port). For the project map: bash npx agenttrail For full component map setup (writes files): bash npx agenttrail init

License and pricing

The license is MIT. The project is open source.

Based on the project's public README and repository metadata.