Skip to main content
Early access. Cloud sessions and managed agent runs are not open to every account yet: we are letting accounts in gradually while we test. They run in a cloud sandbox operated by Mutagent, so there is nothing to host. A sandbox with nothing to do for 15 minutes stops; send the session a message and it wakes up, delivers your message and carries on in the same conversation.
Helix Cloud runs Helix in a cloud sandbox instead of on your machine. A run continues if you disconnect, and a script can start one with a single command. A cloud run uses the four things below. You set up the first three once per workspace. The fourth is the command you run. Start by signing in:

​
Environments

An Environment holds the variables and secrets your cloud sessions need, such as a GitHub token or a database URL. Your local shell variables are not sent to the sandbox, so anything a tool reads from its environment goes here.
Load it into a run with --env ci. Environments belong to the workspace. Secrets are stored encrypted and are never shown again. Model keys do not go in an Environment. They come from your LLM providers. More on Environments

​
Sandbox providers

A sandbox provider is where your cloud sessions run. Mutagent runs them in its own sandboxes, so there is nothing to set up. More on sandbox providers

​
LLM providers

Your sessions call your models with your own API keys, so the workspace needs your LLM providers. If Helix already works on your machine, copy its setup:
providers mirror shows what it will copy and asks first. It copies your API keys and model lists into the workspace and, if the workspace has no default model yet, sets one. helix models lists the models your sessions can use and marks the default with ★. Without a local Helix setup, add an LLM provider directly:
A run with no model available is refused before any sandbox starts. More on LLM providers and models

​
Run Helix in the cloud

You can run Helix in two ways.

​
Run a task

Helix completes the task, making as many model and tool calls as it needs, prints the final answer, and exits. The sandbox stops 15 minutes after the run’s last activity. If you need its files afterwards, mutagent helix session restore <reference> restores it, using the hs1_ reference the run printed.
  • mutagent helix runs the Helix orchestrator.
  • mutagent helix agent "<definition>" runs your own agent instead.
  • Add --prime to either one to use the Prime agent.
  • Use --mode json instead of -p to get events, one JSON object per line.
More on running a task

​
Run a session

For your own agent, use mutagent helix agent "<definition>" --rpc. The session keeps running and reads commands from stdin. The CLI prints the session’s reference, which starts with hs1_. Use it from any terminal:
After 15 minutes with no activity, the sandbox stops; an interactive session is checkpointed first. A turn the agent is still working on counts as activity, for up to 4 hours. To continue, just send it a message: the session wakes on a new sandbox from its last checkpoint and gets your message, under the same reference:
There is no terminal chat client for a cloud session. Use -p for a task, or --mode rpc and session send for a session. More on sessions

​
Run a managed agent

The orchestrator, Prime and your own agent definition are three things you can run. The fourth is a managed agent: an agent you write as an agent.md folder with its own tools, skills and files, deploy to the workspace, and run by its address, as a task with -p or as a session with --rpc.
More on managed agents

​
Options for tasks and sessions

Your repository and local files are not uploaded. The agent works inside the sandbox.

LLM providers and models

Mirror, add an LLM provider, and choose the default model.

Sessions

Send, watch, stop, checkpoint, and restore a session.

Command reference

Every Helix command and its flags.

Troubleshooting

Errors you may see and what to do about them.