Introduction
Trenchcoat monitors, meters, and reports on your team's AI agent usage.
Trenchcoat is a SaaS platform that gives engineering teams visibility into how they use AI agents. It ingests telemetry from Claude Code via a lightweight plugin, stores it in a managed database, and surfaces analytics through a dashboard — sessions, tool usage, token attribution, and cost.
What Trenchcoat tracks
- Sessions — every time a developer opens Claude Code, a session is created
- Tools — every tool call (Read, Edit, Bash, etc.) is captured with name and duration
- Agents — subagent invocations are tracked separately so you can see which agents your team runs most
- Tokens & cost — input/output token counts per session, mapped to current model pricing
Who is this for?
Installing the plugin?
Developers — install the Claude Code plugin, connect your API key, and start sending telemetry in minutes.
Using the dashboard?
Engineering managers — create an account, explore the analytics dashboard, and invite your team.
How it works
Claude Code → Plugin hooks → Local JSONL → Batch push → Trenchcoat API → DashboardThe plugin runs Python hooks on each Claude Code lifecycle event. Events are written locally first, under ~/.claude/trenchcoat/, and pushed to the Trenchcoat API when a session ends — not on a timer. Local recording works with no account at all.
By default only metadata is captured: prompt length rather than prompt text, result size rather than tool output, and no git branch. Content capture is opt-in twice over — you enable it locally and the key's scopes must permit it, or the server strips it on arrival.