Tuneloop Documentation

Tuneloop is an analytics platform for AI coding agents. It captures session transcripts from the harnesses your team uses, links every session to the outcomes it produced — merged PRs, resolved tickets, shipped features — and creates private benchmarks from your own repos so every model and harness gets evaluated on your code, not a public leaderboard.

Session transcripts never leave your infrastructure.

Architecture

The system has three server-side components running on your infrastructure, plus lightweight hooks on machines where your agents run — developers' machines or servers where background agents run.

tuneloop-ingest

Session lifecycle hooks installed wherever your agents run — developer machines or background agent servers. When a session ends, the hook uploads the transcript to your Tuneloop server automatically.

Web

API server that receives transcripts, serves the dashboard, and handles GitHub and Jira webhooks.

Workers

Background processors that parse transcripts through the adapter/processor pipeline: normalize sessions, extract artifacts (commits, PRs, tickets), run LLM enrichment, and execute benchmark runs.

Postgres

Single data store for sessions, artifacts, and metrics.

Explore the docs