Cursor Unveils Origin, a Git Forge for AI Agents at 22.6 Commits/Sec

⏱ 5 min read · Last updated 2026-06-18

At its Compile keynote on June 16, 2026, Cursor introduced Origin, a git forge rebuilt from the ground up for the agentic era. In a live demonstration, the team pushed code through Origin at a rate no human-era git host was designed for, 22.6 commits per second inside a single repository. The forge is Git-compatible and API- and MCP-extensible, so AI coding agents can clone, branch, commit, rebase, review, and fix CI failures programmatically, at scale.

Why It Matters

AI-assisted coding has moved from autocomplete to autonomous. GitHub surveyed 2,000 developers in 2024 and found that 92% already used AI coding tools daily or weekly (GitHub Octoverse 2024). A handful of those agents can trivially generate dozens of commits per minute across multiple branches, throughput that chokes traditional git hosts. Standard forges like GitHub and GitLab were architected for a world where a human typed every commit message. Origin starts from a different assumption: the primary contributor is as likely to be an agent as a person, and the host needs to keep up.

What’s New / How It Works

Origin is a full git forge that speaks standard git wire protocols and works with existing CLI tools, CI runners, and IDE integrations. The difference is a programmable control plane exposed through a REST API and the Model Context Protocol (MCP). Agents can trigger repository operations directly, cloning, branching, committing, rebasing, and even resolving merge conflicts and build failures, without human intervention.

Origin treats AI agents as first-class contributors, not tools to be tolerated within a human-era forge.

Crucially, Origin’s internal architecture parallelizes operations that sequential forges serialize. When multiple agents generate concurrent pull requests, Origin queues and auto-resolves conflicts using an embedded AI resolver, then re-runs tests and merges if checks pass. A built-in agentic review layer inspects diffs and surfaces only the changes that require human attention. Together, these features let Cursor demo 22.6 commits per second in a single repo while maintaining a clean, mergeable history.

Origin was launched alongside Cursor iOS, a beta mobile app that brings AI-powered code editing to the phone, and a new frontier model pre-trained from scratch on more than 100,000 GPUs in collaboration with SpaceX. The model announcement ties directly to the scale of compute that a forge like Origin will consume, and to the SpaceX–Cursor deal that is reshaping the landscape.

The Numbers

22.6 commits per second in a single repository demonstrated during the keynote (Origin launch page)
100,000+ GPUs used to pre-train the new frontier model, developed with SpaceX
MCP-native API so agents can drive the forge programmatically, clone, branch, commit, rebase, review, and fix failures
Agentic merge-conflict resolution and auto-remediation of CI/build failures, parallelized at scale
Git-compatible, works with existing git tooling, CI runners, and IDEs
Fall 2026 targeted release, waitlist now open; no public pricing yet

“A git forge for the agentic era.”
Cursor, Origin launch page

What Comes Next

Origin is currently in invitation-only preview, with general availability planned for autumn 2026. Cursor says it will open-source the MCP agentic APIs and publish a reference implementation so other forge operators can follow. Pricing is still under wraps, but the code-forge world is already speculating that Origin will be priced for agent-scale compute, not per-seat human seats. The iOS app beta hints at an even broader agentic future where code review and merge operations happen on any device, at any time.

The new frontier model, trained on SpaceX’s compute grid, will feed directly into Origin’s agentic review and conflict-resolve layers. The merger with SpaceX, covered in our analysis of SpaceX’s $60 billion Cursor acquisition, gives Cursor dedicated AI-scale infrastructure that few rivals can match.

What This Means for You

If you build software, or run a business that depends on it, the arrival of an agent-native forge signals that agent-authored code is about to flood repositories. Teams will need to decide how much autonomy they grant to AI in branching, merging, and CI/CD, and where human oversight remains critical. The speed at which features can move from request to merged PR will compress, but so will the learning curve for tooling that was never designed for this rate of change. Familiarizing your dev team with MCP-based agent workflows and with the concepts Origin pioneers, automated conflict resolution, agentic review, will be a competitive advantage long before the majority of forges catch up.

The Bigger Picture

Cursor Origin is not a GitHub clone with a chatbot; it is a re-think of what source control infrastructure owes its users when half of them are software agents. If the vision holds, developers will stop managing commits and start managing intentions, and that shift rewrites every assumption about how code gets built and maintained.

Frequently Asked Questions

What is Cursor Origin?

Cursor Origin is an agent-native git forge announced at Cursor’s Compile 2026 conference. It layers a programmable control plane on top of standard git so AI coding agents can perform repository operations, cloning, branching, committing, rebasing, and even resolving merge conflicts and CI failures, programmatically and in parallel. It is Git-compatible and works with existing tooling.

How fast can Origin handle commits?

During the launch demo, Origin sustained 22.6 commits per second in a single repository. That throughput was designed specifically for agentic workflows where multiple AI agents generate concurrent changes, far beyond what traditional git hosts were designed to handle.

Is Origin compatible with regular git?

Yes. Origin speaks standard git wire protocols and integrates with existing CLI tools, CI runners, and IDE extensions. Developers can continue to use the git commands they know while agents take advantage of the MCP and REST API extensions.

What is MCP and why does Origin use it?

MCP (Model Context Protocol) is an open protocol that lets AI models interact with external tools. Origin exposes forge operations as MCP tools, so AI coding agents can manipulate repositories natively, cloning, branching, committing, and reviewing code, without needing custom scripts or brittle shell wrappers.

Can Origin resolve merge conflicts automatically?

Yes. Origin includes an agentic merge-conflict resolver that evaluates conflicting changes, generates a resolution, re-runs tests, and merges if checks pass. The review layer surfaces only the changes that require a human’s attention, keeping the pipeline moving.

When will Origin be available and how much does it cost?

Origin is currently on a waitlist with general availability planned for fall 2026. Cursor has not yet published pricing, but early signals suggest it will be structured around agent-scale compute rather than per-user seats.