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.
A managed agent is an agent you write as files: an agent.md folder with its prompt, tools, skills and files. You check and run it on your machine, deploy it to your workspace, and run it on Helix Cloud by name.

The package

This is the invoice-pricing agent. It prices an invoice from a product catalog and a pricing policy. The Quickstart builds it file by file.
agent.md declares every other part. Only what it declares is packaged. See agent.md for the fields and Tools and skills for writing each part.

The essentials

Four commands take the folder from your machine to Helix Cloud. prod is an Environment in your workspace; create it once with mutagent env set prod INVOICE_CURRENCY=USD. You need the CLI, Bun 1.3.14 for the compiler, and, for deploy and the cloud run, a sign-in with a workspace selected. run also needs Helix and the model’s API key on your machine. The Quickstart lists each requirement with the command that checks it.
  1. check compiles the folder and validates it. It sends no request and runs no tool code.
  2. run runs the task with Helix on your machine. It creates nothing in the workspace.
  3. deploy uploads the package as revision v1 and makes it the active revision in prod.
  4. mutagent helix agent @invoice-pricing starts a sandbox on Helix Cloud, copies the active revision into it, and runs the task.

Manage agents with mutagent agent

check, run and pack work on the folder on your machine. deploy, list, inspect, activate and retire work on the agents in your workspace. Add --json to any of them for the full result: one object on stdout, with success true exactly when the exit code is 0. Exit codes: 0 success, 1 failure (usage errors included), 2 key expired or invalid, 3 not signed in or no workspace. See Errors and exit codes.

check

run

run takes a folder or its agent.md. To run a deployed agent, use mutagent helix agent @<slug>.

pack

pack writes the archive that deploy uploads, and uploads nothing. You do not need it to deploy.

deploy

Deploying changed content stores the next revision, v2, and makes it active. Deploying unchanged content reuses the existing revision.

list

slots shows each Environment with its active revision, or retired.

inspect

activate

activate makes an earlier or later revision active, which is how you roll back. It also enables a retired slot again.

retire

retire asks for --force (or -f) every time, --json included, so a slot is never retired by accident.
A new run in a retired slot is refused with MANAGED_AGENT_SLOT_RETIRED. Every flag, argument and refusal is in the mutagent agent reference.

What the platform keeps

An Environment names a slot and loads its variables and secrets into the sandbox, an LLM provider supplies the key for the model that agent.md names, and a sandbox provider decides where each run happens.

Where next

Quickstart

Build the invoice-pricing agent from an empty folder, deploy it, change it and roll it back.

agent.md

Every field of agent.md and the package it builds.

Tools and skills

Write a tool module, add built-in tools, write a skill, and package files and secrets.

Deployment

Revisions, slots and Environments: what deploy, activate, run, retire and archive check.

mutagent agent

Every mutagent agent command and its flags.

TypeScript SDK

Deploy and run managed agents from code.