Skip to main content
The openlit CLI is a single Go binary that connects your local tooling to an OpenLIT instance. Today it ships first-class observability for AI coding agents - Claude Code, Cursor, and Codex - by wiring into each vendor’s hook system, normalizing events onto OpenTelemetry’s gen_ai.* and OpenLIT’s coding_agent.* semantic conventions, and exporting them via OTLP to your collector. It talks only to your OpenLIT endpoint. There is no managed cloud - your telemetry stays on your infrastructure.
The CLI is the future control plane for OpenLIT. It’s built so new command groups can plug in over time - prompts, traces, eval, and migrate are already reserved in the codebase and will light up in subsequent releases without breaking the coding-agent integration. Today, the CLI is focused on coding-agent observability.

What you can do today

Observe coding agents

Wire Claude Code, Cursor, or Codex into OpenLIT with one idempotent command per vendor - no SDK, no code changes in your repos.

Configure once

Set your OTLP endpoint, API key, environment, and content-capture mode in a single config file or via environment variables.

Launch agents instrumented

openlit coding launch bootstraps the plugin and execs your agent in one step - telemetry on from the first turn.

Diagnose your setup

openlit doctor prints the resolved config, dials your OTLP endpoint, and lists installed plugins so you can confirm everything is wired up.

Command groups

How it works

The hook is the CLI’s hot path - invoked once per agent event. It is crash-isolated: a 5-second hard timeout with ~3 seconds reserved for OTLP flush, and it always exits 0 even when telemetry fails, so a broken pipeline never blocks your coding session.

Get started

Install the CLI

macOS, Linux, Windows, Docker, or build from source

Configuration

Endpoint, API key, content capture, and precedence rules

Command reference

Every command, subcommand, and flag

Coding agents setup

The end-to-end guide to wiring each vendor