What are the rate limits for using multiple AI coding agents on a single GitHub repository?

Verdict: The rate limit does not depend on the number of agents. It depends on the credential each agent uses, so separate credentials give separate budgets.

Limits that apply to a single repository

A single-repository installation does not scale up. It stays at 5,000 requests per hour.

Secondary limits apply to all agents together

These limits protect the whole platform. They are not a per-agent budget, and the docs do not say exactly how they are scoped.

• No more than 100 concurrent requests. This limit is shared across REST and GraphQL.
• No more than 900 points per minute for REST endpoints. Most POST, PATCH, PUT, and DELETE requests cost 5 points. Most GET requests cost 1 point.
• No more than 80 content-generating requests per minute, and 500 per hour. Comments, issues, and pull requests count toward this limit.

Content creation is the limit most likely to affect several agents. Many agents that comment or open pull requests at once can reach it first.

Recommendation

Use one GitHub App installation for each agent, not one shared token. This gives each agent its own 5,000 requests per hour. It also makes each agent's activity easy to separate in the logs. Keep the secondary content limits in mind. Stagger agent writes, and retry with backoff when GitHub returns 403 or 429.

• Choose a shared installation if your agents run at low volume.
• Choose separate installations if agents make many API calls each hour.
• Choose a GitHub Enterprise Cloud organization if you need higher primary limits. Its apps get 15,000 requests per hour.

Each agent can also have its own key. That key mints one-hour tokens scoped to one repository. githug sets this up for agents that run git, and its free beta also gives each agent its own GitHub App.

This is technical information from GitHub's public docs, not legal advice.