AI review agents that QA a website against your own checklist, then route the findings for client approval.
Summary
Cursor's Git forge for the agentic era — host repos, review pull requests and let coding agents work in the same place.
Screenshots
Description
Cursor Origin is a Git-compatible code hosting platform built by Cursor (Anysphere) for teams whose codebase is being written as much by AI agents as by people. Instead of bolting an agent onto an existing forge, Origin puts repositories, pull requests, code browsing and agent sessions in one interface, so a developer can read a review comment and hand it straight to an agent that revises the PR without leaving the window.
What it does
- Origin Repos — create and manage repositories from a Codebase tab inside Cursor.
- Two-way GitHub sync — keep existing GitHub repositories alongside Cursor-hosted ones, with pull requests, timelines, commits and check results syncing in both directions. Adopting Origin does not require migrating off GitHub.
- Pull requests built for volume — full PR timelines, commits and status checks, designed for a workflow where dozens of agents branch, rebase and open diffs in parallel.
- Agents in the loop — agents read the code, push changes and update pull requests from the same surface the review happens on.
- App extensions — Vercel spins up a preview deployment per pull request, while Depot and Buildkite run CI, including existing GitHub Actions workflows.
Who it is for
Engineering teams already running Cursor at scale, particularly those hitting the friction of a forge designed around human reviewers when much of the diff volume now comes from automated agents.
Availability and pricing
Origin entered early beta on 17 August 2026 and is available on all paid Cursor plans; there is no separate Origin subscription. Because it is an early beta, feature coverage is narrower than a mature forge — issues, wikis and fine-grained permissions are not part of the initial release.
Reviews
Similar App Suggestions
Turns backlog tickets into reviewed pull requests — background triage, a task board wired to live agent sessions, and MCP control from your own terminal.
One API for agent harnesses — run Codex, Claude Code or Hermes behind your product with shared sessions, streaming, files and cost controls, and swap harness with a config change.
Cloud coding agents that work in isolated sandboxes, run your tests, preview the app in a browser and open a pull request you can review like any other.
Ad-supported coding agent suite — CLI, parallel desktop workspaces, cloud IDE and a prompt-to-deploy web builder, with no subscription or API key.