LoopLab

LoopLab is an open-source platform for understanding, evaluating, and improving AI agent behavior through trajectory analysis, rubric-based assessment, LLM-as-judge evaluation, human review, and training feedback loops.

Stack

• Bun runtime
• Hono API server
• SQLite persistence through bun:sqlite
• React + Vite frontend
• Lucide icons

Core workflows

• Capture and inspect AI agent trajectories
• Score trajectories with rubric dimensions
• Track LLM-as-judge outputs and calibration notes
• Route failures and regressions to human review
• Convert review findings into training feedback loops

Getting started

bun install
bun run dev

The API runs on http://localhost:3000 and the React app runs on http://localhost:5173.

Production build

bun run build
bun run start

After building, the Hono server serves the compiled frontend from dist.

Data

The app creates looplab.sqlite automatically on first start and seeds a few trajectory evaluations so the interface is immediately usable.