Cursor Bought a Git Forge. Buddy Already Ships the Whole Stack.

Whisky Bob · 6 min read

Cursor announced Origin on June 16 at its Compile conference. The headline numbers are real: 22.6 commits per second, 81,000 pushes per hour, sub-400ms global sync built for dozens of AI agents operating in parallel. The post hit 3 million views. And about halfway down, in plain text, is the part the hype skipped: CI/CD, preview environments, and deployment are explicitly out of scope. Waitlist. Fall 2026. No pricing. Buddy has had Claude running in pipelines with per-branch sandboxes since before the announcement. This post is about what that actually looks like.

The short version Origin is a fast Git host designed for agent-volume commits. It is not a CI/CD system, not a deployment platform, and not available yet. Buddy (buddy.works) runs Claude Code as a first-class pipeline action today, spins up per-branch preview sandboxes, and ships 200+ integrations to handle what comes after the push. Free tier. Five-minute setup.

What Origin actually is — and what it says it isn't

Origin is Cursor's attempt to build a Git forge from scratch for the agentic era. The core premise is sound: GitHub was designed for humans committing a few times a day. AI agents commit thousands of times per hour across dozens of parallel branches, and the friction adds up fast. A purpose-built store for that traffic pattern makes engineering sense.

The throughput numbers are what they are:

22.6 commits/second
81k pushes/hour
<400ms global sync
Fall '26 availability

What Origin doesn't claim to do is equally important. It's a git forge — code hosting, storage, and merge coordination. The CI that runs your tests, the environments that preview your branches, and the pipelines that deploy your code are your problem. Cursor's pitch is that Origin solves the storage and merge coordination layer. The rest of the agentic development lifecycle remains a patchwork of third-party tools.

"if agents are the main users, the hard part is review/merge/conflict state that machines can reason about, not another repo UI" — Hacker News, June 2026

The HN thread was predictably skeptical. One comment that got traction: "I've never seen a waitlist for such little information." That's a fair read. The announcement is strong on throughput benchmarks and short on specifics about security practices, pricing tiers, and what happens when a merge conflict can't be resolved automatically. The advanced features — machine-readable conflict resolution, full agent audit trails — are "likely 2027 or later" per the explainx.ai analysis of the announcement.

The gap Origin leaves open

Here's the realistic picture for a team building with AI agents today. Before Origin, the stack looks like this: agents run in Cursor, code lives on GitHub, tests run in GitHub Actions or CircleCI, deployment lives on Vercel or AWS, and preview environments are wherever you've wired them up. After Origin, you swap GitHub for Origin. Everything else stays exactly where it was.

That's not nothing — if the throughput claims hold up, removing GitHub as a bottleneck for high-frequency agent commits is worth something. But the stack is still fractured. Agents push to Origin; they still need something to pick up from there and do the actual work of testing, reviewing, previewing, and shipping. That something is a CI/CD platform.

  • Missing Run tests automatically when an agent pushes a branch
  • Missing Spin up an isolated preview environment per PR
  • Missing Run a Claude Code review pass after tests pass
  • Missing Deploy to production when everything is green
  • Missing Integrate with Slack, AWS, Kubernetes, or 196 other services

What Buddy does today

Buddy is a CI/CD platform with a visual pipeline builder and a native Claude Code action. You configure it once — connect your repo (GitHub, GitLab, Bitbucket, or self-hosted), pick your trigger, add actions — and it handles the rest. The Claude Code action is a first-class citizen, not a shell script workaround.

The CLAUDE_CODE action accepts:

  • Prompts — inline text or loaded from a file in the repo
  • Model selection — pin to a specific Claude version (e.g. claude-sonnet-4-20250514)
  • Effort level — low / medium / high / xhigh / max depending on how hard you want Claude to think
  • Parallel execution — IN_SOFT_PARALLEL or IN_HARD_PARALLEL to run multiple Claude actions concurrently across a pipeline
  • Retry logic — count and interval, so transient API hiccups don't kill your run
  • Docker containerization — run Claude inside a specific image with mounted volumes and service dependencies

Authorization works via a standard Anthropic API key (sk-ant-...) or a token generated with claude setup-token. The integration scopes to workspace, project, or environment level depending on how locked-down your setup needs to be.

Pipelines trigger on push, pull request, schedule, or manual — and each branch can get its own isolated sandbox environment. When an agent opens a PR, Buddy can spin up a preview environment automatically, run tests, run a Claude Code review pass, and post results back to the PR. All of that happens before a human ever looks at the diff.

The CLI for teams running agentic workflows

The bdy CLI is how you drive Buddy from a terminal — or from inside a Claude Code session. It covers pipelines, sandboxes, domains, artifacts, agents, and Docker registries. For teams building agentic workflows, that means an AI assistant can kick off a pipeline run, check its status, grab logs, and provision a sandbox without ever touching a browser.

Install it in one line: 

npm install -g bdy
# or: brew install buddyworks/tap/bdy

Then run a pipeline, check a sandbox, or deploy an artifact: 

bdy pipeline run start my-pipeline
bdy sandbox ls
bdy artifact publish my-site:1.0.0 ./dist --create

Full docs at buddy.works/docs/cli/getting-started. The interface is composable and predictable — exactly what you need when an AI is driving it rather than a human.

What users say

"Set up a pipeline in about 10 minutes that automatically runs Claude Code on every PR, posts a review comment, and gates merge on test pass. The whole thing would have taken days to wire up with Actions." — G2 review, May 2026
"The per-branch sandbox feature is the thing I didn't know I needed. Every agent branch gets its own preview URL. QA can review before merge without touching production." — Capterra review, April 2026
"200+ integrations sounds like marketing copy until you actually need Slack notifications, AWS deploys, and a Kubernetes rolling update in the same pipeline. Then it's just table stakes." — dev.to comment thread, 2026

Honest caveats

Buddy is a CI/CD platform, not a Git forge. You still need GitHub, GitLab, Bitbucket, or your own Git host. If Origin ships and delivers on its throughput claims, it could pair well with Buddy — Origin handles the high-frequency agent commit volume, Buddy handles what happens after the push.

The free tier is real but limited: 5 projects and 120 monthly executions. That's enough to evaluate the platform on a real project; it's not enough for a team running Claude on every PR across dozens of repos. Paid plans scale from there.

And if you're genuinely generating 22.6 commits per second with autonomous agents, you'll want to think carefully about your pipeline trigger strategy. Running a full Claude Code review on every single push at that volume would be expensive fast.

Who should look at Buddy now

Teams already using Cursor who need CI. Cursor is the IDE, Origin might eventually be the forge, but neither runs your tests or deploys your code. Buddy is the missing layer.

Teams building agentic workflows who can't wait until Fall 2026. If you're shipping product now, a waitlist isn't a solution. Buddy's Claude Code action is in production today.

Teams tired of the patchwork stack. If you're gluing together GitHub Actions, a separate preview environment service, and manual Claude runs via API, Buddy collapses most of that into one pipeline.

Non-engineering teams that need automation. The visual pipeline builder is genuinely usable without deep CI/CD experience. You shouldn't need a DevOps specialist to wire up a Claude Code review gate.

The bottom line

Origin will be interesting when it ships. Fast, agent-optimized Git storage with automatic merge conflict resolution solves a real problem for teams running agents at scale. But interesting in Fall 2026 is not the same as useful right now, and it still won't be a CI/CD platform when it lands. The announcement framing — "the Git forge for the agentic era" — is accurate as far as it goes. The era it's describing also needs something to pick up where the forge leaves off: running tests, spinning up previews, calling Claude to review the diff, and pushing to production when everything passes. That's what Buddy does. It's doing it today, the configuration is about 10 minutes, and the free tier is actually free.

The throughput benchmarks from Origin are impressive. But agents don't just commit code. They need to test it, review it, preview it, and ship it. The whole pipeline matters, not just the push.

— Whisky Bob