Create a workspace, run an agent, and inspect the files it saves. Use Macrofold Cloud, a self-hosted deployment, or the free local simulator. The steps are the same.
To integrate your application instead, use the API quickstart or coding-agent setup prompt.
1. Sign in
Open the dashboard and create an account. Verify your email before using protected features. For local development, use the demo account from the local setup guide.
Choose the organization you want to work in. Workspaces, API keys, and usage belong to that organization.
2. Create a workspace
Open Workspaces, choose New workspace, and give it a name. A persistent workspace keeps its files across runs. Start with an empty workspace; you can connect GitHub or upload files afterward.
A worktree is the workspace's independent working folder and branch. Two agents can take turns over the same files; use separate worktrees when they need to write at the same time. See sharing a worktree.
3. Start a run
Create a run from the dashboard. Select your workspace, harness, and an available model. For an empty workspace, try “Create hello.txt containing Hello world.” Set a runtime and spending limit, then submit. No connected apps, saved agent preset, or schedule is needed for this first task.
On a real deployment, managed execution uses prepaid credits. To use your own model account, add a provider connection and select BYOK. Compute and authorized tools can still use platform credits. The local simulator uses synthetic credits and does not call a model.
4. Follow the work
Open the run to follow status, output, tool activity, and any provider-exposed reasoning summaries. A run may wait for capacity, an earlier worktree task, or input from you. The run view explains its current state.
You can leave the page and return later. Closing the browser does not cancel the run. Use Cancel when you want to stop it.
5. Review and continue
When persistence finishes, open the worktree files and its checkpoint history. Execution, persistence, and Git synchronization have separate outcomes; check all three before relying on a remote Git update.
For a real-model run, open hello.txt to confirm the requested file was saved. Ask a follow-up to read it and write a second file. The simulator demonstrates this lifecycle with scripted files and output instead of interpreting that prompt.
Continue the same session to preserve a compatible native conversation, or start a new session over the saved files. Your files do not depend on keeping the earlier execution environment running. After this first result, reuse an agent or explore ongoing workflows.
Next steps
- Core concepts: learn what persists and what a run represents.
- API quickstart: submit a task from your application.
- CLI guide: link a directory and stream remote work.
- Connections: add provider keys, MCP servers, and apps.