Roadmap
githug makes agent work accountable, end to end: every agent gets a name, every commit gets a human, anyone can check, and your repo decides what agents may do. Here's what's live and what's coming. No dates; this is built in the open, a piece at a time. Updated 10 October 2026.
Identity · live
Every agent gets its own identity on GitHub instead of borrowing yours.
- Its own bot. Each agent can get its own GitHub App in two clicks, so its commits and PRs show as
<agent>-githug[bot], with its own avatar and rate limits. - Keys that expire. Agents trade their githug key for a GitHub token that works on one repo and dies in an hour.
- A human on record. Every agent names the person accountable for it. PRs can open with you as author, and the agent and run go in the footer.
- Sub-agents. An agent can create workers under it, with a subset of its repos and permissions. Disable the parent and every worker stops.
- Teams. A GitHub org can own its agents as one account, with each agent still naming its member.
Get started: quickstart.
Proof · live
Anyone can check who made a commit and who answers for it, without trusting the commit message.
- Verify page. Paste any public commit or PR at githug.ai/v to see the agent, the accountable human, and whether githug issued the key.
- Attested pushes.
hugreports every commit it pushes with the agent's own key, so "verified" means the agent's key vouched for that exact commit. Commits that only claim an agent show as consistent, not verified. - Verify Action. A free GitHub Action for maintainers that fails a PR when a commit claims an AI agent nobody can vouch for. Human commits pass untouched. Docs.
- Runs. Every key and commit in one agent run shares a run id, so you can follow a task end to end.
Self-hosted · live
The same service, open source under AGPL-3.0, on your own infrastructure.
- One container with SQLite, or your own Cloudflare account with one script.
- Every feature included, single-tenant, no billing. Your GitHub App, created in one click.
- Self-hosting guide · source at github.com/githug-ai/githug.
Rules · next
Your repo decides what agents may do, enforced when they push, not after.
- Require the chain. Agent commits must carry an accountable human, or the check fails.
- Human sign-off on sensitive paths. Changes to paths you name (CI, auth, billing) need a person's approval.
- Tests must pass before an agent's PR can merge.
- Activity limits. Cap an agent's open PRs, pushes per hour and PR size, per agent or per repo. Over the limit, githug stops issuing keys and the check fails.
Rules post a check your branch protection can require, so they work with the GitHub settings you already have.
Everywhere · later
The same chain on every forge, and evidence you can hand to someone else.
- GitLab and Gitea / Forgejo, with agent identities created through their APIs.
- Evidence export. The full chain for a period, from human to agent to commit to rule result, as a signed file for audits and insurers.
- Agent Merge Index. Public numbers on whose agent PRs get merged, reverted or abandoned, from verified commits only.
How it's built
The service is open source (AGPL-3.0) at github.com/githug-ai/githug. The hug CLI and the verify Action are open source too (Apache-2.0): github.com/githug-ai/hug, github.com/githug-ai/verify. The service is free during the open beta; see pricing.