Convergent

Convergent shows you every run of every AI agent you operate.

Convergent Control is the workspace for operating AI agents in production. It collects the traces and sessions your agents produce, turns them into reviewed insights, and shows what each agent did and why. Graders — platform detectors and ones your team authors — score production behavior continuously, so a regression surfaces as an insight instead of a support ticket.

From this workspace a team watches live production traffic, manages environments and ingestion keys, reviews provider-reported model usage and cost, and administers organization members. Instrumenting an agent takes a few lines of the Convergent Python SDK; the documentation covers setup end to end.

Signing in happens in a browser through WorkOS. A coding agent or another headless client should not sign in here: read /auth.md on this host to register for credentials that work at the API and the MCP server, /llms.txt for the page index, and /openapi.json for the API routes published on this origin.

Where to go

Documentation — Python SDK documentation, from install to production.

Agent index — Every public and workspace page, with instructions for coding agents.

Agent registration — How an agent gets credentials for the API and MCP server without a browser.

API reference — OpenAPI document for the control-plane routes published on this origin.

Service status — Whether the hosted control plane is reachable. No session required.