Open source · MIT

See what your AI agents
actually do.

Every tool call, token, and dollar your AI coding agents spend — on one dashboard. Then set guardrails that stop the risky ones before they run.

npm install -g @klars/agentobs

Requires Node 22.5+. Runs entirely on your machine — no account, no signup.

The AgentObs dashboard: spend for the week, tool call and error-rate tiles, an activity chart, and tables of tools and sessions. — The local dashboard — cost, tool calls, errors and blocked calls at a glance.

Local-firstNothing leaves your machine

Zero configOne command to start

MIT licensedRead every line

Three commands

1
Install and initialise
```
npm install -g @klars/agentobs
agentobs init
```
Prints a hook config block. Paste it into ~/.claude/settings.json.
2
Open the dashboard
```
agentobs dashboard
```
Runs at 127.0.0.1:4300. Tool calls appear within seconds.
3
Add guardrails
```
agentobs policy init
```
Blocks rm -rf and curl | sh out of the box.

Stop it before it runs

Guardrails are evaluated inside the same hook that does the logging, so a blocked call never executes. Check exactly what a rule will do before it fires mid-task:

Terminal demo: agentobs policy test blocks an rm -rf command, then agentobs stats shows cost, calls, errors and blocked totals.

Honest cost

A model missing from your pricing table shows —, never $0.00. One wrong number and you stop trusting all of them.

Honest fidelity

Process-wrapped sessions are labelled coarse. The UI never implies detail it doesn't have.

Fails open

A malformed policy degrades to allow-everything and says so. A guardrail that wedges your agent is worse than none.

Privacy is the whole point

AgentObs sits in the middle of everything your agent does. That only works if it is trustworthy by construction, not by promise.

No network calls. No telemetry, no analytics, no account.
Secrets are redacted before anything touches disk. AWS, Anthropic, OpenAI, GitHub, GitLab, Slack, Stripe and Google keys, JWTs, PEM blocks, KEY=value pairs, Authorization: headers and URL credentials.
The redaction rules are unit-tested — the tests are the guarantee, and they caught real leaks during development.
Uninstalling is rm -rf ~/.agentobs. That is the entire footprint.

Read the source Documentation

See what your AI agents actually do.