FAQ

Is Manyfold just a chat wrapper?

No. Manyfold hosts real agent runtimes with workspace files, terminal access, resumable sessions, provider settings, skills, and optional channel connections.

Which agent should I choose first?

For repository work, start with Claude Code or Codex. Choose Gemini CLI if your workflow already depends on Gemini. Use Hermes Agent or OpenClaw when you need a framework-style agent for connectors, services, scheduled jobs, or product workflows.

Can I attach files when I chat with an agent?

Yes. Use the attachment button in the chat composer to add images or documents to a message, and the agent receives them along with your text. Most agents support attachments, including Claude Code, Codex, Gemini CLI, OpenClaw, Hermes, and Dify. For a Dify agent, make sure file upload is enabled for the connected Dify app and that the app allows the file types you send.

Do I need my own model keys?

You can bring your own provider keys. Some workspaces may also have managed model access. The creation flow shows the options available to your account.

What is the difference between a stateful sandbox and a cloud computer?

A stateful sandbox is best for interactive coding and task work. It can pause and resume while keeping the workspace state.

A cloud computer is best for always-on work such as services, connectors, or scheduled workflows. Rent it first from Settings -> Plan & billing -> Buy container, then attach agents to it.

Can I run agents on my own machine?

Yes. Use Settings -> Self-owned computers to issue a token, then register the machine with the mf CLI. The registered machine appears as a self-owned computer and is useful when the agent needs access to local files, hardware, or a private network that is not available from the cloud workspace.

Can I connect one agent to multiple channels?

Yes. Create one channel per external bot or app. Keep labels clear so you know which team, room, or workflow each channel serves.

Where should I put secrets?

Put model keys in Settings -> Model providers and channel tokens in Settings -> Channels. Do not paste secrets into chat prompts, files, issue descriptions, or public logs.

Can agents make mistakes?

Yes. AI-generated work can be wrong or incomplete. Review important outputs, inspect code changes, and run tests before relying on an agent’s work.

How do I reduce unexpected costs?

Use clear prompts, stop runs that are going in the wrong direction, review Settings -> Usage, and confirm the agent is using the intended provider and model.

How do I get support?

Contact support from your Manyfold workspace or email customer-support@netmind.ai. Include the agent name, approximate time, and a short description of what happened. Do not include secrets.