Guides

Connect a model to a Mandala computer, move code from another provider, and fix what goes wrong.

• Claude computer useDrive a Mandala desktop or browser with Claude's toolsets, in Python and TypeScript.
• OpenAI computer useDrive a Mandala computer with the computer tool in OpenAI's Responses API.
• Gemini computer useDrive a Mandala computer with Gemini's computer use tool in its desktop environment.
• LangChain agentsGive a LangChain agent a Mandala computer as a set of tools.
• Hermes AgentRun NousResearch's Hermes Agent on a Mandala computer from the hermes template, with your model key bound as a secret, and drive it by API.
• OpenClawRun OpenClaw on a Mandala computer from the openclaw template, set it up with your model key bound as a secret, and drive it by API.
• Claude Managed Agents workerRun a self-hosted sandbox for Anthropic's Claude Managed Agents on a Mandala computer, from the managed-agents template, with the environment key bound as a secret.
• Run your agent next to a computerHost the agent process on Fly.io, DigitalOcean or Hetzner Cloud, pass it your Mandala API key as a platform secret, and clean up its computers.
• Moving from Orgo or E2BMap Orgo and E2B Desktop code to the Mandala SDK, and what behaves differently.
• TroubleshootingCommon failures when creating and driving Mandala computers, what causes them, and the fix.

A skill for agents

The Mandala skill tells an agent when a cloud computer is the right tool, how to get access, the working loop, and which failures are worth a retry. An agent that supports skills can load it from https://app.mandala.computer/skill.md, and llms.txt links to it. Save it as SKILL.md in a skills directory, for example ~/.claude/skills/mandala-computer/SKILL.md for Claude Code.