Prepare
Resolve the repository, exact commit, entrypoint and dependencies.
How it works
Sulcus runs your agent, so it can watch and stop it.
Sulcus Cloud checks out your repository, runs your entrypoint in a resource-limited container and instruments your framework automatically. Because Sulcus launches the run, it can stream what happens and end it when you ask.
Execute and instrument
Start the container and instrument the framework your app already uses.
Stream and retain
Stream events to Sulcus Cloud live, and keep the run’s history.
Cloud Git runWhat runs where
01
You, in the browserConfigure the run · watch it live · stop it
start · stoplive events
02
Sulcus CloudSign-in · projects and runs · token limit · run history
launch · stopruntime events
03
Resource-limited Docker containerOperated by Sulcus · fixed CPU, memory and time limits
- Your repo at the resolved commit
- LangGraph · CrewAI · OpenAI Agents SDK
- Sulcus instrumentation
Who does what
Your code decides the work.
Your application still defines the prompts, agents, tools, graphs, crews and handoffs. Sulcus doesn’t set goals for your agent or assemble a workflow for you.
Sulcus operates the run.
Sulcus Cloud starts the container, activates the framework instrumentation, collects events, tracks status and token usage, and keeps the run record. It can stop the run because it started it.
Sulcus doesn’t attach to agents already running on your own infrastructure. Cloud runs are started by Sulcus.
One run, step by step
From configuration to history.
- 01 · Prepare
Resolve
Load the project from your workspace and copy its current configuration into the new run.
- 02 · Prepare
Validate
Check the HTTPS repository URL, Git reference, Python entrypoint, framework and dependency mode.
- 03 · Prepare
Check out
Clone the repository inside the container, resolve the reference to an exact commit and install dependencies.
- 04 · Execute
Instrument
Activate the LangGraph, CrewAI or OpenAI Agents SDK instrumentation for the framework you selected.
- 05 · Execute
Run
Start your normal Python entrypoint with runtime secrets, fixed CPU, memory and time limits, and outbound network access.
- 06 · Stream
Stream
Send ordered, redacted runtime events to Run Details as they happen, and enforce the token limit if you set one.
- 07 · Stream
Retain
Keep the configuration snapshot, resolved commit, timestamps, outcome, token totals and event history.
Workspace-scoped access
Projects and runs belong to a workspace. Every request is checked against your workspace membership on the server.
A fixed kind of workload
Sulcus accepts a validated repository, reference and entrypoint — not arbitrary Docker images, commands, host paths or flags.
Redacted events
Runtime events and failure diagnostics are redacted for the secrets you supply and for common credential patterns before they reach Cloud. Redaction is best effort.
Honest isolation limits
Containers run with resource and filesystem restrictions, but they aren’t a sandbox for hostile code. Run repositories you trust.