// the find
0-AI-UG/cate
An infinite zoomable canvas for coding. Editor, terminal, and browser panels in a spatial workspace.
Cate is an Electron desktop IDE where all panels — editors, terminals, browsers, AI agents — live on an infinite zoomable canvas instead of tabs. You drag panels around, dock them into splits, and the layout persists per workspace. Aimed at developers who find VS Code's tab model limiting and want spatial organization of their work.
The companion daemon architecture is the right call — filesystem ops, terminals (node-pty), ripgrep search, and git run in a separate process from the Electron main process with a typed RPC protocol. Security boundary is real, not theater: filesystem access is scoped to registered workspace roots, browser panels have node integration disabled. The test suite is substantial for a project this size: vitest unit tests co-located throughout, Playwright e2e tests covering drag scenarios, cross-window behavior, and canvas stress tests. Git integration goes beyond a status bar — worktrees, staging, history, inline diffs, and a GL-rendered territory visualization that maps worktrees onto canvas regions is a genuinely clever use of the spatial metaphor. Multi-provider AI (Anthropic, OpenAI, Copilot, Gemini, OpenRouter, Groq, Mistral, DeepSeek) with per-chat model selection rather than one locked-in choice.
The core coding agent depends on `@earendil-works/pi`, an opaque third-party package. The product's main differentiator is a black box controlled by someone else — if Pi changes its API or goes unmaintained, the agent feature breaks and there's nothing in the repo to fix. The skills/extension ecosystem is a stub: the registry has one skill (`cate-theme`) and the marketplace infrastructure exists but is empty. The README itself says 'Don't build from source for daily use,' which signals the build process is brittle enough that they'd rather ship binaries than let contributors run it. Electron 41 + Monaco + xterm.js is a heavy stack; the RAM footprint will land in the familiar 400–600MB range, which is a hard sell against native alternatives like Zed or even VS Code for developers who care about resource usage.