We rebuilt Sculptor from the ground up, thanks to your feedback. Now you can run multiple Claude Code agents in safe containers and instantly see their changes in action.
Download Scuptor for MacOS (Apple Silicon)
Download Sculptor for MacOS (Intel) [experimental]
Why we built Sculptor
We released Sculptor back in April as a research preview for catching and fixing issues in agent code. As our community started using Claude Code, they told us what they needed most: a way to run multiple agents safely in parallel, save context between sessions, and quickly verify agent changes.
So we rebuilt Sculptor to solve those problems first. Now you can spin up parallel agents in safe containers, stay in flow across sessions, and instantly bring their changes straight into your IDE for testing and review. Catching and fixing issues is where we’re headed next, starting with Suggestions (in beta) and more capabilities to come.
How Sculptor works
Run agents in parallel
With Sculptor, you can spin up new agents the moment you think of something: kick off a refactor while another agent is building out a feature, or launch an agent to check a hunch while you’re debugging, all without breaking your flow.
Every agent runs in its own container, so they can all execute code safely in parallel. Unlike git worktrees—which share your local environment and require reinstalling dependencies for every agent—Sculptor containers keep your machine safe and let you move between agents seamlessly, so you can keep building without interruption.
Instantly test agent changes
Running agents in parallel is powerful—but only if you can quickly see their changes in action. That’s where Pairing Mode comes in. With one click, Sculptor brings an agent’s work from its container into your local repo, keeping your files and git state synced so you can collaborate directly from your IDE.
Test the agent’s work instantly in your own dev environment and commit changes you like—all without losing your flow. Spin up multiple agents to explore different approaches, and jump between them effortlessly. And because the sync works both ways, the agent sees your edits and comments in real time, making iteration feel seamless.
Merge without conflicts
When you’re happy with an agent’s work, Sculptor makes it simple to merge in their changes. Keep what you want, drop what you don’t, and Sculptor will flag potential merge conflicts automatically.
If conflicts do come up, you can hand them back to the agent to resolve. And with our Suggestions feature, Sculptor will even point out fixes to the agent’s code so you can merge with more confidence.
Save context across sessions
Sculptor saves every agent session with its plans, chats, tool calls, and code changes all intact.
You can reopen a past session anytime to pick up where you left off, or revisit an old experiment without re-prompting or starting from scratch. Sculptor keeps your agent history organized in one place—so context is never lost, and progress is always preserved.
What’s next
We’re just getting started. Here are a few things on the horizon:
- Forking agents: spin off a new agent from any point in your session history, so you can reuse context that’s working without reloading from scratch.
- Custom MCPs: add your own MCP servers so agents can call them directly.
- Smarter containers: use your own Dockerfile, or let Sculptor build one from your repo so the agent’s dependencies match yours.
- GPT-5 support: Claude Code is supported today, and GPT-5 is next, giving you the option to choose the right model for the job.
- Instruction audits: write rules in plain English (“don’t use
evalanywhere in the codebase”, “always use NumPy”), and Sculptor will review the agent’s work to catch violations. - More powerful Suggestions: we’re adding new issue types, from flagging misleading behavior (“tests passed” without real tests) to instruction-file following (
claude.md) and beyond.
Come sculpt with us!
A huge thanks to everyone who stuck with us through the research preview :). Your feedback shaped everything you see today.
Sculptor is now available on Mac (Apple Silicon) and Linux, with Mac (Intel) and Windows support on the way.
We’d love to hear what you think! Drop a comment on our launch posts and we’ll highlight your work. And if you want to go deeper, join our Discord—it’s the best place to chat directly with the team and other builders, and share workflow tips.
Can’t wait to see what you build!