What It Does
Collaborator is an Electron desktop application that provides an infinite pan-and-zoom canvas where developers can arrange terminal tiles, markdown notes, code editors, and image viewers as free-floating panels. The stated use case is running AI coding agents (Claude Code, Codex, Gemini CLI, or any terminal-based agent) in terminal tiles while keeping relevant context files, notes, and code visible alongside them on the canvas, without switching windows or tabs.
The tool is local-first: all canvas state, workspace configurations, and tile positions persist as JSON files in ~/.collaborator/ with no cloud sync, no account, and no telemetry described in the README. A companion repository (collaborator-ai/collab-plugins) provides two Claude Code slash commands — /collaborator:initiative and /collaborator:ontology — that scan folders of markdown files to produce goal hierarchies and entity-relation graphs via Claude’s native reasoning.
Key Features
- Infinite canvas: Pan-and-zoom workspace; tiles snap to a grid; canvas viewport position persists between sessions.
- Terminal tiles: Full PTY emulation via xterm.js and node-pty sidecar; each terminal runs an independent persistent session; working directory set to the active workspace path.
- File-tree navigator: Hierarchical navigator sidebar; drag files onto canvas to open as tiles; multiple workspace support with quick switching.
- Markdown editor tiles: Inline editing with live rendering; created by dragging
.mdfiles onto the canvas. - Code editor tiles: Monaco Editor with syntax highlighting and language detection for non-markdown files.
- Image viewer tiles: Read-only display for
.png,.jpg,.jpeg,.gif,.svg,.webp. - Windows (PowerShell + WSL2) support: Since v0.6.0 (March 31, 2026).
- v0.8.0 chat interface: The April 2026 release replaced the agent terminal tile model with a full chat interface with tool-call cards, markdown rendering, and session persistence — a significant paradigm shift.
- collab-plugins: MIT-licensed Claude Code skill commands for initiative analysis and ontology extraction over markdown folders.
- No account required: All data local in
~/.collaborator/as JSON.
Use Cases
- Solo developer context management: A single developer running one or two AI agents in terminal tiles with context markdown files visible alongside — useful for keeping a CLAUDE.md, a scratchpad, and an agent terminal on one screen without alt-tabbing.
- Spatial task mapping: Arranging multiple terminal sessions (one per task or agent) with associated markdown notes on the canvas as a visual map of in-progress work.
- Markdown knowledge pipeline (collab-plugins): Scanning a folder of engineering notes, decisions, or roadmap files through the initiative/ontology Claude Code commands to extract structured insight.
Adoption Level Analysis
Small teams (<20 engineers): Marginal fit at this stage. The tool works for a solo developer seeking a spatial terminal manager. However, it provides no agent-specific features (no git worktree isolation, no diff review, no issue-tracker integration, no multi-agent coordination) that would distinguish it from a tiling terminal emulator. The v0.8.0 architecture shift (terminal → chat) signals the core experience is still being defined.
Medium orgs (20–200 engineers): Does not currently fit. No multi-user features, no shared workspaces, no access control, no audit logging, no CI integration. The Electron binary and local-only storage model are inherently single-user. Teams at this scale have better options in Emdash, Vibe Kanban, or OpenHands.
Enterprise (200+ engineers): Does not fit. No compliance story, no team disclosed, no licensing clarity (license field is not populated in the repository), no enterprise features, no vendor relationship possible.
Alternatives
| Alternative | Key Difference | Prefer when… |
|---|---|---|
| Emdash | Git worktree isolation per agent, 23+ agent providers, SSH remote dev, diff review, PR lifecycle, YC-backed | You need a real agent orchestration workbench with isolation and workflow integration |
| Vibe Kanban | Kanban-style agent management, MCP integration, 23.4k stars, MIT | You want a lightweight proven multi-agent coordinator with broader community |
| tmux / terminal multiplexer | Zero overhead, CLI-native, keyboard-driven, composable | You are comfortable in the terminal and don’t need a GUI canvas |
| Claude Code (direct) | Best single-agent terminal coding experience, Anthropic ecosystem | You run one agent at a time and want the most capable coding agent |
| Tauri-based alternatives | 96% smaller binary than Electron, same web frontend | You need a desktop app but want lower resource overhead than Electron |
Evidence & Sources
- GitHub repository — collaborator-ai/collab-public, 2.4k stars
- collab-plugins repository — MIT, Claude Code skills
- collaborator.bot — minimal landing page
- Haystack IDE HN thread — infinite canvas IDE precedent and failure modes
- Collaborator v0.8.0 release notes — interface paradigm change
Notes & Caveats
- Unknown license: The GitHub repository does not include a LICENSE file at time of review. The collab-plugins repository is MIT, but the license for the main application is undisclosed. This is a meaningful concern before adopting open-source tooling; it creates legal ambiguity for redistribution or embedding.
- Anonymous team: No team members, company affiliation, funding, or backers are disclosed anywhere. The
collaborator.botlanding page is a placeholder. This is not inherently disqualifying for a hobby/community project, but it eliminates any vendor accountability or support pathway. - Fundamental architecture instability: The v0.8.0 release (April 16, 2026) replaced the primary agent interaction model — the terminal tile for agents — with a chat interface. A six-week-old project changing its core interaction paradigm is a signal that the team has not yet found product-market fit or a stable design direction.
- Electron overhead: The application uses Electron 40 with a multi-webview architecture. Electron apps carry significant memory and binary size overhead vs. native or Tauri-based alternatives. This is not unusual in the developer tool space, but noteworthy when lighter alternatives exist.
- No sync story: Local-only JSON storage means canvas layouts do not transfer across machines. Developers working on multiple devices must manage this manually, which becomes a friction point quickly.
- collab-plugins dependency on Claude: The plugins rely on Claude’s reasoning directly; they add no independent intelligence. Any Claude Code user could achieve similar results with a well-crafted prompt. The plugins provide convenience packaging, not proprietary capability.
- No independent validation: As of April 2026, no HN threads, no independent blog posts, no post-mortems, and no production usage reports have been found. The 2.4k stars may reflect early explorer interest rather than sustained use.