Changelog

You can now upload images to agents in Sculptor and agents will be able to see them (in full resolution)!

Thanks to @Bryden, and everyone on Discord who requested this feature @sleepingsheep @gloomy @justuseapen @CMyNerd :)

We’ve also added support for Haiku 4.5, so you can write code at 1/3 the cost and 2x the speed 🔥 @Sam E

Improvements & Fixes

• Fixed frontend UI flickers 🙏🏻 @Guinness
• Improved tracking for .gitignored files to speed up Pairing Mode start ✨ @Maciek
• Fixed task crashes when forking 🎉 @Guinness
• Reduced volume of log uploads to Sentry for bug reporting 💎 @Maciek

We’ve sped up the Sculptor app startup time! Sculptor should now just take a few seconds to start. If you’re downloading an update, it’ll take a bit longer.

Improvements & Fixes

• More reliable Docker detection and warnings when Docker is disconnected — Sculptor should no longer crash when your Docker is disconnected.
• Fixed shutdown edge cases (threads, tmux session, nginx handle) so quitting Sculptor doesn’t leave orphan processes.

You can now fork Claude Code agents with Sculptor!

Spin off a new agent from any message in your chat — it keeps all prior context.

Forking lets you share feature context across multiple agents, evaluate multiple libraries for the same task, and save money by reusing context.

<include image / video>

Our favorite Forking workflows:

• Curate clean context in one agent, then branch off new ones that reuse it.
• Try the same task multiple ways. Then, pick your favorite and merge!
• Fork to recover an agent when context gets messy — like when debugging derails your feature work.

You can now remove repositories from the Sculptor app — just go to Settings → Repositories and remove the repo.

Sculptor now works on Linux! 🎉🎉🎉

Download it here.

You can now see when Pairing Mode is on for an agent, no matter which agent you’re looking at.

Now you can start a new task using CMD+N, and search with CMD+P.

Sculptor is now a desktop app, so you can easily alt-tab to it!

New Features

• Added bundled Mutagen for improved file synchronization
• Added external link handling capabilities
• Added comprehensive desktop application documentation

Improvements

• Enhanced configuration system with per-user settings support
• Enhanced local sync with safer initialization and error handling
• Improved local sync testing infrastructure for better reliability
• Removed repository state caching for improved performance
• Eliminated force fetching to prevent unnecessary git operations
• Improved sync button organization with better data flow
• Enhanced commit notifications for better collaboration

Fixes

• Fixed critical snapshot functionality to prevent agent state corruption
• Fixed git reflog handling preventing repository corruption
• Fixed desktop application title bar display issue
• Fixed header display issues in UI

Now you can selectively compact an agent’s context when you’re running low.

What’s new

• You can one-click copy your own messages from chat to reuse prompts without retyping
• Empty states for Plan/Suggestions make it more clear what to expect before results arrive
• The system prompt now tells Claude it runs inside a container so it won’t install dependencies and think they’re on your machine

Fixes

• A regression that broke @-mentions is fixed

You can now @mention files and folders in your repo from chat, so the agent knows to look at them!

Improvements

• You can now create a new task directly from the chat view
• Pairing Mode now pauses when you don’t have the agent’s branch checked out locally, so you can avoid weird local git states

Fixes

• Resolved a rare crash condition due to heavy file operations

Now you can open huge diffs without waiting—content streams in progressively so you can begin reviewing instantly.

Improvements

• Links in the Chat view now open in new tabs
• Added dialog to show you what data we collect when you give feedback on tasks for increased transparency
• Task startup time reduced by ~2 seconds with smarter Docker caching

Fixes

• Tasks now start reliably even if your repo is empty

You can now write Markdown when chatting with the agent. Paste in Markdown or code and it’ll stay readable.

Improvements

• You can now give a quick 👍 or 👎 on tasks to share feedback on the agent’s work
• Docker errors are now classified and shown more clearly, so you know what went wrong with the container

Fixes

• Welcome messages now display properly when you rtask list is empty
• UI rendering bug that caused freezing is fixed

You can now spin up agents off any local branch—not just main—using the branch selector when starting a task.

Highlights

• The chat layout is reordered to Plan → Changes → Suggestions → Log for a more natural workflow
• Your draft messages are now saved even when you switch away from the task
• You can now retry agent tasks when they fail

Fixes

• Script execution is sandboxed more robustly so failures are surfaced without breaking the container
• Fixed bugs that prevented you from stopping an agent’s work mid-stream