Give an agent a task, follow its progress, and keep the work it produces. Macrofold runs Codex, Claude Code, OpenCode, Hermes, DeepSeek Harness, and Pi through the same API, terminal, and dashboard—without a separate integration for each harness.
Workspace files outlive individual runs. Start with one useful task, then reuse its files, agent setup, and connected tools for the next. From one task to ongoing work explains how the pieces fit together.
Start building
- Build with AI: copy a setup prompt into your coding agent and describe your feature.
- API quickstart: create a workspace, submit a task, and print its result.
- Dashboard quickstart: run and review a task without writing code.
Choose where Macrofold runs
| Option | You manage | Start here |
|---|---|---|
| Macrofold Cloud | Your application, workspaces, permissions, and usage | Connect to Cloud |
| Self-hosted | Your Macrofold deployment and its infrastructure | Self-hosting guide |
The same API, SDK, CLI, and feature guides apply to both. Use the origin and credentials for your deployment; enabled integrations, model availability, and limits depend on its configuration. For a free first experiment, use local simulation.
Learn the essentials
Start with a workspace for files, an agent for the job, and a run to do one task. You can use the workspace's main worktree before learning about parallel branches or saved conversations. Core concepts explains the resources and what persists.
- Explicit-context decisions: submit evidence for typed decisions, bounded investigations and reviewable task proposals.
- Runs and agents: select a harness, stream output, continue, and cancel.
- Worktrees and Git: edit files, restore checkpoints, and synchronize a repository.
- Share files between agents: take turns in one worktree or use separate branches in parallel.
- Connections: model keys, named app accounts, and MCP tools.
- Triggers and schedules: start work from Slack, webhooks, or recurring prompts.
- Dashboard and billing: review work, manage access, and control usage.
Build a customer-facing agent
Read the customer identity guide, then run the personal-agent application. It keeps customer ownership, named agents, files, conversations and schedules connected. Add the optional file memory starter or use customer-data recipes for an existing database or MCP service. Each guide includes setup, examples, recovery, validation and an explicit AI reading map.
Go deeper
API overview · SDKs in five languages · CLI · Streaming · Errors and retries · OpenAPI · Troubleshooting
Develop and operate
Local simulation, Docker, and cloud staging are development modes. Self-hosting covers operating a deployment. Architecture, the codebase map, and contributing explain how to change Macrofold itself.
Anyone can contribute a harness through the Unified Harness Interface, reusing the existing execution, streaming, and persistence system.