Compare ·
Scape vs Sculptor
Sculptor, from Imbue, is a desktop app for running coding agents in parallel. Its README calls it an experimental research preview.[1] Here is how it compares with Scape, with a source for every claim about Sculptor.
| Feature | Scape | Sculptor |
|---|---|---|
| Agents | ||
| Agents | Claude Code, Codex, Copilot, Cursor, OpenCode, pi, omp and Kimi Code, on your own subscriptions. | Claude Code and the Pi harness built in, and it runs any terminal-based agent.[3] |
| Manager agent (Argus) | Argus reads your mission, starts child agents in their own worktrees, checks on them, and keeps a log table. You choose which agent and model its children run on, and can restart it on another agent without losing them. | Several agents per workspace, and a task dependency graph.[1] |
| Agent rooms | Chat rooms where agents and people work one problem together, such as Claude and Codex debating an approach. | No. |
| Messages between sessions | One agent can message or direct another session. | No. |
| Agent review | Adversarial review: a second agent reviews the first one's diff (/scape-codex-review ships with Scape). | PR creation and status.[1] |
| Watchdogs | Supervisors that spot stuck sessions and auto-answer questions and permission gates (Claude Code). | Claude Code runs sandboxed with tool permissions auto-approved.[6] |
| Automation | ||
| Playbooks | Step-by-step automations that chain shell commands, agent prompts, HTTP calls and Scape tools. Lock one and it runs as a fixed script. | No. |
| Scheduler | A per-project calendar that runs playbooks and messages Argus, once or daily, weekly or monthly, while Scape is open. | No. |
| Triggers | Run a playbook or prompt Argus when a webhook arrives, an integration record changes, a table query matches, or a calendar event happens. You arm each one. | CI Babysitter (experimental, off by default): when a GitHub PR's CI fails or it hits a merge conflict, an agent is started to fix it.[7] |
| Working with people | ||
| Flight Plans | One shared, end-to-end encrypted plan for any number of people and their agents. Every open question is a thread; answers land in the doc as decisions and tasks, and you can hand a task straight to an agent. Share it by link and approve each person who asks to join. | No. |
| Encrypted chat | End-to-end encrypted group and one-to-one rooms. Messages are never stored on a server. | No. |
| Sharing | Drag notes, playbooks and tables into an encrypted chat to share them. | No. |
| Workspace | ||
| Notes | Rich-text notes with version history, embedded charts and full-text search, synced through your iCloud. | No. |
| Tables | SQLite-backed tables your agents and playbooks read and write, viewed as a table, kanban or gantt. | No. |
| Charts | Bar, line, area, scatter, pie and KPI charts built from your tables, embeddable in notes. | No. |
| Custom Apps | Small apps built from a JSON spec, kept in your project. | Extensions: JavaScript modules that add panels, header widgets, home views and overlays; an agent can build one.[8] |
| Integrations | Jira, GitHub, Slack, Notion, Linear, ServiceNow, Bugsnag, Datadog, or any HTTP source, in your sidebar. | Bundled Linear extension shows issues linked to the branch and PR; GitHub pull requests are opened and tracked through the GitHub CLI.[8] |
| Dev tools | ||
| Worktrees | A git worktree per session, with a post-create hook you set per repo or globally. | A workspace is a git worktree by default; clone and container backends are experimental.[4][5] |
| Code editor | Monaco, with 60+ languages, a diff view, Vim mode and LSP. | No. |
| Browser | A built-in browser your agents can drive: screenshots, clicks, forms and scraping. | No. |
| Dev servers | A live browser preview paired with each server. Cmd+click jumps to the source, and each worktree gets its own port. | No. |
| Spectator mode | Every file an agent edits opens live in the editor, so you can follow along. | Live diff of the agent's changes for Claude Code and Pi agents.[6] |
| The basics | ||
| Status | In active development. | Latest release v0.49.0 on 2026-10-21. Described as an experimental research preview.[9] |
| Where agents run | On your Mac. Claude Code sessions can also run on a saved SSH host. No hosted agents. | On your machine, with optional Docker, SSH or VM backends (experimental).[5] |
| Platform | macOS. Claude Code sessions can also be driven from the Claude mobile app through Remote Control, opt-in per project. There is no Scape mobile app yet. | macOS (Apple Silicon), Linux and Linux ARM64.[1] |
| Price | 14-day free trial, no account or card. Then Pro at $9.99 a month or $99 a year; you bring your own agent subscriptions. | No price listed. |
| Open source | No. | Yes, MIT.[2] |
Agents
- Agents
Scape: Claude Code, Codex, Copilot, Cursor, OpenCode, pi, omp and Kimi Code, on your own subscriptions.
Sculptor: Claude Code and the Pi harness built in, and it runs any terminal-based agent.[3]
- Manager agent (Argus)
Scape: Argus reads your mission, starts child agents in their own worktrees, checks on them, and keeps a log table. You choose which agent and model its children run on, and can restart it on another agent without losing them.
Sculptor: Several agents per workspace, and a task dependency graph.[1]
- Agent rooms
Scape: Chat rooms where agents and people work one problem together, such as Claude and Codex debating an approach.
Sculptor: No.
- Messages between sessions
Scape: One agent can message or direct another session.
Sculptor: No.
- Agent review
Scape: Adversarial review: a second agent reviews the first one's diff (/scape-codex-review ships with Scape).
Sculptor: PR creation and status.[1]
- Watchdogs
Scape: Supervisors that spot stuck sessions and auto-answer questions and permission gates (Claude Code).
Sculptor: Claude Code runs sandboxed with tool permissions auto-approved.[6]
Automation
- Playbooks
Scape: Step-by-step automations that chain shell commands, agent prompts, HTTP calls and Scape tools. Lock one and it runs as a fixed script.
Sculptor: No.
- Scheduler
Scape: A per-project calendar that runs playbooks and messages Argus, once or daily, weekly or monthly, while Scape is open.
Sculptor: No.
- Triggers
Scape: Run a playbook or prompt Argus when a webhook arrives, an integration record changes, a table query matches, or a calendar event happens. You arm each one.
Sculptor: CI Babysitter (experimental, off by default): when a GitHub PR's CI fails or it hits a merge conflict, an agent is started to fix it.[7]
Working with people
- Flight Plans
Scape: One shared, end-to-end encrypted plan for any number of people and their agents. Every open question is a thread; answers land in the doc as decisions and tasks, and you can hand a task straight to an agent. Share it by link and approve each person who asks to join.
Sculptor: No.
- Encrypted chat
Scape: End-to-end encrypted group and one-to-one rooms. Messages are never stored on a server.
Sculptor: No.
- Sharing
Scape: Drag notes, playbooks and tables into an encrypted chat to share them.
Sculptor: No.
Workspace
- Notes
Scape: Rich-text notes with version history, embedded charts and full-text search, synced through your iCloud.
Sculptor: No.
- Tables
Scape: SQLite-backed tables your agents and playbooks read and write, viewed as a table, kanban or gantt.
Sculptor: No.
- Charts
Scape: Bar, line, area, scatter, pie and KPI charts built from your tables, embeddable in notes.
Sculptor: No.
- Custom Apps
Scape: Small apps built from a JSON spec, kept in your project.
Sculptor: Extensions: JavaScript modules that add panels, header widgets, home views and overlays; an agent can build one.[8]
- Integrations
Scape: Jira, GitHub, Slack, Notion, Linear, ServiceNow, Bugsnag, Datadog, or any HTTP source, in your sidebar.
Sculptor: Bundled Linear extension shows issues linked to the branch and PR; GitHub pull requests are opened and tracked through the GitHub CLI.[8]
Dev tools
- Worktrees
Scape: A git worktree per session, with a post-create hook you set per repo or globally.
Sculptor: A workspace is a git worktree by default; clone and container backends are experimental.[4][5]
- Code editor
Scape: Monaco, with 60+ languages, a diff view, Vim mode and LSP.
Sculptor: No.
- Browser
Scape: A built-in browser your agents can drive: screenshots, clicks, forms and scraping.
Sculptor: No.
- Dev servers
Scape: A live browser preview paired with each server. Cmd+click jumps to the source, and each worktree gets its own port.
Sculptor: No.
- Spectator mode
Scape: Every file an agent edits opens live in the editor, so you can follow along.
Sculptor: Live diff of the agent's changes for Claude Code and Pi agents.[6]
The basics
- Status
Scape: In active development.
Sculptor: Latest release v0.49.0 on 2026-10-21. Described as an experimental research preview.[9]
- Where agents run
Scape: On your Mac. Claude Code sessions can also run on a saved SSH host. No hosted agents.
Sculptor: On your machine, with optional Docker, SSH or VM backends (experimental).[5]
- Platform
Scape: macOS. Claude Code sessions can also be driven from the Claude mobile app through Remote Control, opt-in per project. There is no Scape mobile app yet.
Sculptor: macOS (Apple Silicon), Linux and Linux ARM64.[1]
- Price
Scape: 14-day free trial, no account or card. Then Pro at $9.99 a month or $99 a year; you bring your own agent subscriptions.
Sculptor: No price listed.
- Open source
Scape: No.
Sculptor: Yes, MIT.[2]
“No” means it isn’t in Sculptor’s app or docs: we checked their own site, docs, changelog and repo on , and only count a feature that’s built in and does the same job. If that’s changed, tell us and we’ll update the page.
Why people pick Scape
- Hand an Argus the whole job. Give it the goal and your guardrails. It starts your agents with the routine permissions already granted, checks on them, and only brings you the decisions that are yours to make.
- Solve the problem in front of you. Ask your agents for a Custom App, a playbook or a table that does the job: a morning incident triage, a launch tracker that refreshes itself on a schedule, a report that lands in your notes. Code is only part of the work.
- Let your agents talk to each other. Put Claude and Codex in one room to argue out a fix while you read along, or have one session hand work to another. Any agent, any model, on the subscriptions you already pay for.
- Work with your team, privately. Bring your whole team and everyone's agents into a Flight Plan. Each open question is a thread, the answers land in the doc as decisions and tasks, and a task can go straight to an agent. Talk in end-to-end encrypted rooms: messages aren't stored on a server, and you can run the relay yourself.
- Own your AI setup. Your apps, playbooks and tables live on your Mac, and your notes sync through your own iCloud. Switch the agent or model underneath them whenever you like, and everything you built keeps working.
Where Sculptor is stronger
- Open source.[2]
Sources
Facts about Sculptor were checked on and may have changed since. Something wrong? Email contact@scape.work and we’ll fix it.
- Sculptor on GitHub
- Sculptor: licence
- Sculptor docs: integrated harnesses
- Sculptor docs: workspaces
- Sculptor docs: container backend
- github.com…ue-ai/sculptor/blob/main/docs/help/integrated_harnesses.md
- github.com/imbue-ai/sculptor/blob/main/docs/help/ci_babysitter.md
- github.com/imbue-ai/sculptor/blob/main/docs/help/extensions.md
- github.com/imbue-ai/sculptor/releases/tag/sculptor-v0.49.0