Run supported agent frameworks in Sulcus, connect Claude Code and Codex from your computer, or observe Google ADK and Microsoft Copilot Studio where they already run. See what your agents do, and step in where Sulcus has control.
Free during early access.
No agent yet? Run our demo: github.com/sulcus-dev/langgraph-loop-demo. It needs no API key. Watch it live, press Stop, or set a token limit and see it fail.
Illustration with sample data, not a live run: a hosted run with 38,420 of a 50,000 token limit used, recent model and tool activity, and a Stop run action.
The problem
It keeps calling the model or the same tool. Tokens climb; nothing moves forward.
A step hangs inside the workflow. From outside, it just looks busy.
A coding agent wants to run a command or change a file, and you’d rather decide than find out afterwards.
Observe and control
Tracing tools record what an agent did. Sulcus can also act, and how much depends on where the agent runs.
Sulcus runs your Python agent from a public Git repository.
Claude Code and Codex keep running where you started them.
The agent keeps running where it already runs. Sulcus works from its telemetry.
Integrations
Graph and node steps, model calls and tool activity where LangChain callbacks expose them, with reported token usage.
Crew, agent and task activity, plus model and tool events where CrewAI exposes them.
Agent lifecycle, handoffs, model and local tool calls and guardrail events for SDK 0.22.2 and later 0.22 releases.
Agents, sub-agents, tool calls and model calls from ADK’s own OpenTelemetry spans. Also works for an ADK service you run yourself.
Sessions on your computer appear as runs. Approve or deny supported permission requests from Sulcus.
Experimental. Sessions started through Sulcus appear as runs, with approvals for commands and file changes.
Sulcus reads your environment’s Application Insights telemetry and shows each conversation as a run, while the agent keeps running in Microsoft’s environment.
Custom agents · Preview
No framework, or one Sulcus doesn’t integrate? Your agent can report to Sulcus over a plain HTTP API. There’s no SDK: a few JSON requests, with curl, Python and JavaScript examples.
Sulcus sees only what your agent sends. An approval works only if your agent asks and acts only on approved, and a stop request only if your agent checks for it and stops itself. Sulcus can’t stop or block an agent that ignores the API.
Privacy
Anything a hosted app prints, including a model response, becomes part of the run history. Redaction is best effort, so avoid printing sensitive data.
FAQ
For hosted runs, no: Sulcus instruments LangGraph, CrewAI, the OpenAI Agents SDK or Google ADK when your entrypoint runs, so you don’t add Sulcus imports. Claude Code and Codex need the Sulcus command-line tool on your computer. A deployed Google ADK service needs three environment variables, and Microsoft Copilot Studio is connected in the browser with nothing added to the agent.
Verify your email, then connect Claude Code or Codex, a deployed Google ADK service or a Microsoft Copilot Studio environment. Anyone can sign up and run an agent from a public GitHub repo. Hosted runs support LangGraph, CrewAI, OpenAI Agents SDK and Google ADK.
It depends on where the agent runs. A run hosted by Sulcus, yes: Stop run ends the run and its container. An agent on your computer, no: you can approve or deny supported actions, but ending observation leaves Claude Code or Codex running. A custom agent connected over the API, only if it cooperates: Request stop is a request your agent has to check for and honor by stopping itself, and Sulcus can’t stop an agent that ignores it. A service deployed elsewhere is observed only.
The framework integrations don’t collect prompt text, model responses or tool inputs and outputs. Hosted runs do keep what your application prints to stdout and stderr, up to a limit, so anything your code prints, including a model response, appears in the run history. Avoid printing sensitive data.
No. It’s a guard based on the token counts your framework reports. Once reported usage reaches the limit, Sulcus blocks the next model call or ends the run, depending on the framework. A request already in flight can finish and exceed the limit, so it isn’t a ceiling on what your provider bills.
You do, through your own provider. Where a run needs a model, you supply your own provider API key, and that usage may be billed by your provider. Sulcus shows token counts, not costs.
