Skip to main content
Bento watches your AI agents in production, learns what “broken” looks like for your app, and tracks the problems it finds so you can fix them.

​
Set up with an AI coding tool

The fastest way in. Paste this into Claude Code, Cursor, Codex, OpenCode, or Windsurf. It installs Bento’s Agent Skills and hands the agent the full integration playbook:
For per-tool MCP setup, see the quickstart.
Bento traces view

​
What you get

Traces

Every span from every framework, captured over OpenTelemetry. Open the span tree, or follow a signal straight to the call that broke.

Signals

Describe a failure mode in plain English. Bento trains a detector on your own traces, fires it in real time, and backfills your history.

Issues

Bento groups recurring failures into tracked issues, the front door for triage, with status, severity, and the runs behind each one.

Clusters

See what your agent is doing at a glance. Similar findings group into clusters, so patterns surface before you go looking.

​
Start in three minutes

Quickstart

Install the SDK, send your first trace.

Integrations

One line to trace Google ADK.

​
Pick your SDK

Python

pip install bentolabs-sdk. Google ADK integration, plus manual tracking for everything else.

TypeScript

Python is the supported SDK. Track the TypeScript SDK here.

​
From the terminal

CLI

uv tool install bentolabs-cli. Triage issues, inspect signals, drive analytics. Every dashboard surface, from your shell or CI.