StatefoldCoding agents that keep working
while you’re away.
Run Claude Code, Codex, OpenCode, Pi and Gemini CLI in cloud workspaces. Let them brief each other, review each other’s work, and leave you a diff you can ship.
Explore the workspaceHow it works
- 01
Start a task.
From the browser, a GitHub repository, or
statefold task newwith your uncommitted work. - 02
Agents work together in the cloud.
Each task gets its own workspace. Agents follow your saved workflow, or start tasks on other harnesses and message them, like a Codex build asking Claude Code for a review.
- 03
Review and ship.
Read the conversation, inspect the diff and the running preview, then open a pull request.
- Claude Code
- Codex
- OpenCode
- Pi
Gemini CLI
A real workspace,
not a chat window.
Files, terminal, running preview and the full diff sit next to the conversation. When the agent finishes, the work stays saved and you can review it whenever you come back.
Agents that talk
across harnesses.
A Codex task can brief a Claude Code reviewer, wait for its findings, and reply with a fix. Each agent keeps its own workspace and native conversation. Messages are attributed to the task that sent them, and files move only when an agent retrieves them.
Review the auth changes in src/auth. Check token expiry.
Two findings: expired tokens aren’t rejected; missing test for refresh.
Fixed both. Please re-check.
For repeatable work, save a plan → build → review workflow that hands off through files.
- 01
Plan
Claude CodeDefine the approachwritesplan.md - 02
Build
CodexImplement the planreadsplan.mdwritesbuild.md - 03
Review
Gemini CLIApproval gate: ask youreadsbuild.md
An example. Each task has its own workspace and native conversation; files carry the work forward.
Files carry the handoff.
Plans, code, and review notes stay with the work. The next agent starts from those files.
You decide what runs next.
Continue automatically, pause for approval, or loop back to an earlier stage with a repeat limit.
Your project,
remembered.
Opt-in project memory and versioned skills give every task the same decisions and conventions, without re-explaining them.
Inspect, edit or forget saved notes. Running tasks keep their pinned skill versions.
Questions
Is Statefold a coding agent?
No. Statefold is the workspace and workflow layer around native coding agents. You choose the harness, model, connection, and instructions for each stage.
Can agents on different harnesses work together?
Yes. Tasks in the same project can message each other, wait for each other’s results and pull each other’s files, whichever harness each one runs. Each agent keeps its own native conversation; Statefold doesn’t merge them.
Can I use my own accounts?
Yes. Connect your own provider accounts and API keys, and see which connections have quota before a stage runs. Supported sign-in methods vary by harness.
What happens to my work when the agent stops?
The workspace is saved. Review the conversation, diff and files later, download the result, or open a pull request. You can also steer a running agent or queue the next instruction.
Do I have to push my local work first?
No. The CLI snapshots your current Git workspace, including uncommitted work, without changing your local branch or index. Review cloud work in the browser and download the result separately.