Sculptor is a desktop application for running coding agents in parallel. You connect a repository, create a workspace — an isolated copy of the code — and prompt an agent; when it finishes you review the changes and merge them back. Opening another workspace starts a separate project, and adding a second agent to an existing workspace puts two agents on the same one.
Each agent runs in its own container rather than a shared git worktree, so agents execute code without touching the local environment or reinstalling dependencies per agent. Pairing Mode brings an agent's work out of its container into the local repository, keeping files and git state synced in both directions, so changes can be tested in your own development environment and the agent sees your edits as you make them. When work is accepted, Sculptor flags potential merge conflicts during the merge back.
Harness choice is part of the design: integrated support covers the Pi Coding Agent harness alongside Claude Code, and any terminal-based agent can also be run and managed. The application ships a library of workflow skills for agentic development, including a fix-bug workflow that runs a short reproduction interview, writes failing tests, and then makes them pass. It is published as an experimental research preview and is under active development.
Features
- Container per agent: each agent executes in an isolated container instead of a shared local worktree
- Parallel workspaces: several agents run at once on different tasks or on the same workspace
- Pairing Mode: two-way sync of an agent's container work into the local repository and IDE
- Merge review: accepted changes merge back with automatic flagging of potential conflicts
- Harness choice: integrated support for Claude Code and the Pi Coding Agent, plus any terminal-based agent
- Bundled skills: workflow skills such as spec, mocks, and fix-bug, usable across parallel workspaces
- Built-in terminal: a terminal scoped to the workspace, alongside chat, changes, and pull-request views
- Pull requests: opening a GitHub pull request from a workspace and tracking its status
- Command palette: search and navigation across the application
- Container backend: experimental Docker or remote execution
- Platforms: desktop builds for macOS on Apple silicon and for Linux, including ARM64

