FAQ
If your question isn't here, open a GitHub Discussion.
How is this different from Cursor or Windsurf?+
Those are single-player AI editors. withvibe is multiplayer: a team shares one env and one agent session. The cross-functional flow — stakeholder + PM + dev + QA in one place — is the point.
How is this different from GitHub Codespaces or Replit?+
Codespaces and Replit give you a remote dev environment. withvibe gives you a remote dev environment plus a shared AI collaborator plus an agent gate before merge. The env is the unit of work for the whole team, not just engineers.
Do I need to use Claude?+
Today, yes — withvibe is built on the Claude Agent SDK. Other models are on the roadmap; the agent abstraction is provider-agnostic in design.
Can it run on my private repos?+
Yes. Set a GITHUB_TOKEN in your deployment (or a per-workspace token) and the API will clone private repos into envs. Code never leaves your infrastructure when self-hosted.
Is the AI prompt history visible to the whole team?+
Inside an env, yes — that's the point. Across envs, no: each env has its own session, and workspace-level memory is opt-in.
What happens if the agent breaks something?+
Three layers of defense: (1) every change is in a branch, never on main; (2) humans review before submitting to the gate; (3) the agent gate runs security / review / tests / policy agents before merge. Mistakes stay in the env.
Can I disable the agent gate for trivial changes?+
You can configure which agents run per repo via withvibe.gate.yml. Disabling the gate entirely is possible but not recommended — it's the main reason this is safe to give to non-engineers.
How do I contribute?+
Open an issue or PR on GitHub. We tag good first issues. The architecture (web / api / db, plus per-env runner sidecars) is documented in the repo's README.
What's the license?+
GNU AGPL-3.0 — open source. You can read, modify, and self-host it freely; contributions welcome (a CLA is required). The hosted tier is offered under a separate commercial license.