Observability for AI agents

See what your agents are actually doing.

Cost, performance, and quality for every Claude Code session, in one view. Built into the workflow you already use.

No credit card required

app.trenchcoat.io
Trenchcoat Overview dashboard showing session, event, and tool-use metrics with a daily activity chart and top tools breakdown

Under the coat

Everything a session did, on the record.

sessions

Session timeline

Replay any session turn by turn — every tool call, error, and stop reason, with time spent per phase.

cost

Cache-aware cost

Real cost per session and per agent, priced from live model rates with cache reads and writes counted.

agents

Agent tracing

Trace the full spawn tree a session launches — model, tool count, cost, and latency for every subagent.

evals

Evals

Tag sessions and record eval scores through the API to track agent quality over time.

skills

Skill analytics

See which skills your agents reach for, and how often, across every session.

teams

Team access

Share dashboards, manage API keys, and control who sees what.

For the engineer building agents

Find the tool call that went sideways.

  • Replay any session turn by turn, with every tool call, error, and stop reason.
  • Trace the subagents a session spawns, with model, cost, and latency for each.
  • See which skills your agents reach for, and how often.

For the leader paying for them

Know what every session cost.

  • Real cost per session and per agent, priced from live model rates.
  • Eval scores recorded through the API and tracked over time.
  • Shared dashboards and API keys, with control over who sees what.

Wire up your first session

About five minutes, start to finish.

  1. 01
    Create an API keySign up, then open Settings → API Keys. Keys start with ct_live_ and are shown once.
  2. 02
    Install the pluginTwo commands inside Claude Code add the marketplace and the plugin.
  3. 03
    Add your keyPut it in the env block of ~/.claude/settings.json, then start a new session.
  4. 04
    ConnectRun /trenchcoat:connect, start one more session, and work as usual. Sessions land in your dashboard.
> /plugin marketplace add pando-codes/pando-plugins
> /plugin install trenchcoat@pando-plugins
  # add your key to ~/.claude/settings.json
> /trenchcoat:connect
The commands that install and connect the Trenchcoat plugin in Claude Code.

The agents don’t know they’re being watched. We think they’d be proud.

Coming soon

More ways to plug in

Trenchcoat is expanding beyond Claude Code.

In development

OpenAI Agents SDK

Instrument agents built on the OpenAI Agents SDK and stream their telemetry into Trenchcoat.

In development

GitHub Copilot

A Copilot Extension that reports agent activity from your GitHub workflow into Trenchcoat.

Pull back the coat.

Connect the Trenchcoat plugin and events start flowing from your next session.