Skip to main content
Coding agents are log sources too. Most of them ship an OpenTelemetry exporter, so you point it at your intake URL from Settings > Log streams and Sazabi stores each session’s prompts, tool calls, model requests, token usage, and cost as searchable log records. Use this to see how your team uses its agents, what they cost, and where sessions fail. Each page covers the agent’s own telemetry settings and what it sends. The setup is manual: you configure the agent, and Sazabi does not connect to the vendor’s account. Two related setups are documented elsewhere:
  • To use Sazabi from your coding agent through the Sazabi MCP server, see Connect your coding agent.
  • To hand issue fixes to a cloud agent such as Codex, Cursor, or Devin, see Integrations.

​
In this section

Claude Code

Session telemetry from the CLI, Claude Code on the web, and GitHub Actions.

Codex

Session telemetry through Codex’s built-in OpenTelemetry exporter.

Cursor

Team agent telemetry through a Cursor OpenTelemetry export destination.

Devin CLI

Session telemetry through an otel block in the Devin CLI config.

Factory Droid

Usage metrics or GenAI spans from Factory’s org-managed settings.

Gemini CLI

Session telemetry through a telemetry block in Gemini CLI settings.

GitHub Copilot

Copilot CLI and VS Code agent traces and events.

oh-my-pi

Session spans with model, token usage, and cost, without prompt text.

OpenHands

Conversation runs, agent steps, tool calls, and LLM requests.

Hermes Agent

Sessions, model responses, and tool results via the traceverse OpenTelemetry plugin.

OpenClaw

Agent runs, model calls, tool executions, and Gateway logs via the diagnostics-otel plugin.

Claude Agent SDK

Telemetry from agents you build with the Claude Agent SDK.