Stateful sandbox
Default choiceAn isolated cloud workspace that pauses and resumes while keeping files and session state.
Pick this if the work does not need your machine.
Manyfold documentation
Run, connect, and orchestrate AI agents. Start with a hosted agent in a sandbox, or register your own computer and let an agent work in your repository.
First decision: where does your agent run?
Almost every setup question downstream depends on this one, so it goes first.
Self-owned computer
Your machineRuns on a Mac, Linux, or Windows machine you control, using its repository, tools, and network.
Pick this for a local repository, CLI, GPU, or VPN.
Cloud computer
Always onA long-running Manyfold cloud computer for services, connectors, and scheduled workflows.
Pick this when it has to stay online.
Then, in order
The first setup, in the order the getting started guide and the CLI install page describe it.
- Sign in
Sign in to Manyfold.
- Add a model provider
Add a model provider key in Settings -> Model providers, unless your workspace already has managed model access.
- Create your first agent
Create the agent from New agent, then choose its framework and runtime mode.
- Send the first task
Open the chat workspace and send the first task.
- Install the CLI
For terminal access, install the CLI after signing in, then run mf login and mf whoami.
Browse the docs
Where each section starts. The full tree is in the sidebar, open at every level.
Start
Understand the product, install mf, and create your first agent.
Build
Give an agent a workspace, a model and somewhere to run, then drive it from the terminal.
- Agents
- Runtimes
- Deployment
- CLI
Connect
Wire an agent to the tools your team already uses, and to your own applications.
- Channels
- Automations
- A2A
- API
Reference
Look up the exact shape of the API and the CLI, plus the machine-readable versions of these docs.
- CLI commands