Skip to main content
Terma gives engineering teams one place to see what their AI coding agents are actually doing. It captures sessions from agents like Claude Code and Codex, prices every model call, tells subscription-covered spend from metered spend, and stamps each commit with the sessions that produced it — so you can trace agent spend all the way to shipped code.

Quickstart

Create a team, install the CLI, and see your first session.

How Terma works

Teams, sessions, principals, attribution, and cost views.

Claude Code

What Terma captures from Claude Code and how.

CLI reference

Four commands: setup, doctor, update, and teardown.

​
How it works

1

Set up your machine once

Run terma setup. It signs you in, chooses your team, points your coding agents at a relay on your machine, and writes their machine-wide hooks. Nothing is written into your repositories: your team’s collection policy decides which repositories are collected.
2

Work as usual

Your agents export their usage to Terma as they run. Terma’s hooks record which files each session edits and, when your team’s policy turns on commit stamping, stamp your commits with Agent-Session-Id and Agent-Tool trailers. Nothing waits on the network.
3

See where the effort goes

The dashboard shows spend and a live burn rate, every session with a full replay, usage by person and model, the pull requests agent sessions contributed to, and a burn map of agent work that has — and has not — reached Git.

​
Supported tools

Cursor, OpenCode, and Antigravity are marked Coming Soon in terma setup; each integration page describes what Terma will capture.
Terma is in private beta. If sign-up asks for an invite code, enter the one from your invite. People invited by an organization don’t need a code.