Install AI agents into any GitHub repo. Give each a task, a schedule, and a dollar budget. They run, commit, and stop when the money runs out. No surprises.
{
"$schema": "https://selfdrivingrepo.com/schema.json",
"agents": [{
"name": "research-digest",
"prompt": "Scan arxiv for new LLM papers, summarize into docs/",
"llm_model": "anthropic/claude-sonnet-4.6",
"frequency_hours": 24,
"budget_cents": 1000,
"budget_period": "month",
"tools": {
"bash": true,
"diff": false,
"threads": false,
"webSearch": true,
"upstreamRepos": []
}
}]
}
Pick a repo — new or existing. One click.
Give each agent a prompt, a schedule, a model, and a budget cap. Config lives in the repo as JSON.
Agents run on schedule and commit to your branch. Budget hits zero → agent stops. Git history is the audit trail.
Set a dollar limit per agent per month. When it's spent, the agent stops. Period. No runaway bills, no overages.
Agents commit to your git repo. No proprietary format. No hosted database. Cancel anytime — the work stays.
selfdrivingrepo.json — version-controlled alongside your content. Edit in the dashboard or in your editor.
A researcher, a writer, a linter. Each has its own model, prompt, schedule, and budget. Compose what you need.
It's just a GitHub App and a JSON file. Remove the app, keep your repo. Everything is plain files in git.
You pay OpenRouter model cost + 20%. No platform fee, no seat fee, no minimum. Your agents only spend what you budget.
| Model | Input / 1M tokens | Output / 1M tokens |
|---|---|---|
| Loading prices… | ||
Prices include the 20% markup. All 300+ OpenRouter models are supported — these are the most popular. Prices update live from OpenRouter.
Connect GitHub. Define agents. Set a budget. That's it.
Continue with GitHub