TwinsGitHub
A GitHub twin for testing AI agents.
Test agents that open issues, review pull requests and merge code. Repositories your agent can push to, with no real repo at risk.
A stateful copy of GitHub, built for testing agents.
It accepts the same API and MCP calls, checks sign-in and input, and follows the same rules, but runs on its own data, so nothing reaches your real account.
- Repositories, branches and commits
- Issues and pull requests
- Reviews, comments and checks
- Labels and milestones
- Permissions and branch rules
Every run starts from a company that already has history. We seed:
- Repositories with history
- Open pull requests with reviews
- Issues linked to tickets
- Teams and permissions
Use GitHub in staging, without using production.
Validate every action
The twin checks what the agent sends the way GitHub would: required fields, permissions, rate limits, malformed input. Mistakes fail here, not in your account.
Start from a realistic state
Describe the company and the situation in plain English. Drizz seeds the twin to match, and resets it to that state before every run.
Replay tasks across tools
The GitHub twin shares its people, customers and history with the other twins, so a task that touches three tools runs the way it would at work.
Swap the address. Your agent code stays the same.
Replace the live GitHub endpoint with the twin’s. Everything your agent already does keeps working, on data that is yours to break.
{
"mcpServers": {
"github": {
"type": "http",
"url": "<twin MCP endpoint>",
"headers": { "Authorization": "Bearer <token>" }
}
}
}REST works on the same data. Both are reset with one call.
Connected to the rest of the company.
A pull request links to the Jira or Linear ticket it closes and the Slack thread where it was reviewed.
Read the guide: How to test AI agents before production →