Works with the agentsyou already run.

Where your agent runs decides what Sulcus can do with it. Find yours, then follow its setup.

Hosted by Sulcus

Sulcus runs your Python agent from a public Git repository. Watch it live, stop it and set a token limit.

  • Graph and node steps, model and tool calls, tokens.

  • Crews, agents and tasks, model and tool calls, tokens.

  • Agents, handoffs, tools and guardrails, for SDK 0.22.2 and later 0.22 releases.

  • Agents, sub-agents, tools and model calls. Also works deployed.

On your computer

Early access

Claude Code, Codex and Cursor keep running where you started them. The Sulcus command-line tool is provided on request, and controls work while the computer is online.

  • Approve or deny supported permission requests, and stop it at its next action.

  • Codex

    Experimental

    In sessions started through Sulcus, approve or deny commands and file changes, and interrupt a turn.

  • Cursor

    Preview

    Each conversation appears as a run. Observation only.

Deployed elsewhere

The agent keeps running where it is. Sulcus reads its telemetry, so runs appear after the fact and can’t be stopped from Sulcus.

Your own agent

Preview

Any agent can report its runs over a plain HTTP API, ask a person to approve an action and check whether someone asked it to stop. It has to call the API itself.

Custom agents docs

See your first run in Sulcus.