Where context is built.
OrbIDE turns your thinking into structured, queryable context that any AI tool can consume. Write a note. Promote a sentence into a Decision, a paragraph into a Source, a checklist into Tasks. Feed it to Claude, Cursor, Copilot, or any MCP-compatible agent.
Knowledge graph
~2 MB
Native bundle
<100 ms
Cold start
830+
Passing tests
MIT
Open licence
Problem
Why we're building this.
AI loses context every session
Your AI coding agent starts every task from scratch. No memory of your architecture. No clue about past decisions. No context about why you built things the way you did.
CLAUDE.md goes stale immediately
Hand-maintained context files diverge from the codebase within a sprint. By the time you remember to update them, the LLM has already given you the wrong answer.
Notes apps aren't agent-readable
Obsidian and friends are great for humans. Agents see undifferentiated blobs of text and miss the structure you knew was there — decisions, tasks, sources, tests.
Features
What makes it different.
Context-centric, not page-centric
A single line can be a first-class object. Promote any text into a Task, Decision, Meeting, Source, or Playbook without leaving your notes. Each object has its own type, schema, and addressable identity.
AI gets memory, not just prompts
Context packs bundle your architecture decisions, naming rules, project structure, and active tasks into a format AI agents actually understand. Stop hand-maintaining CLAUDE.md.
Write anything
Rich text with Tiptap. Code blocks with CodeMirror. LaTeX compilation. Markdown export. It's a full workspace for articles, code, specs, and research notes — not just a note app.
3D knowledge graph
Visualise your entire workspace as an interactive force-directed 3D graph. See how decisions connect to projects, tasks link to sources, and ideas form clusters. Built on Three.js.
Lighter than VS Code, smarter than Obsidian
~2 MB core bundle. Boots in under 100 ms. Native desktop via Tauri — no Electron bloat. 830+ tests on every commit. Offline-first with Yjs CRDT for collaborative edits.
One-click context export
Generate CLAUDE.md, .cursorrules, AGENTS.md, wiki/index.md, and schema/context.json from your workspace. Your AI tools finally know what you know — without manual maintenance.
Roadmap
Coming in the next releases.
These features are actively in development. Watch the repo to see them ship.
Dynamic Context Engine
The workspace progressively builds structured context from your notes as you write. Architecture decisions, naming conventions, key files, project structure — all extracted automatically and kept in sync across a raw / wiki / schema layout.
3D Force-Directed Knowledge Graph
Interactive Three.js visualisation where every object, document, and code module is a node in a physics-simulated 3D space. Zoom, rotate, click to navigate. Watch your knowledge graph grow in real time.
Codebase scanning & analysis
Open any project folder. OrbIDE scans the directory tree, detects languages, frameworks, entry points, and import / export relationships. The scan builds a code knowledge graph that merges with your note objects.
Context export for AI tools
One-click generation of CLAUDE.md (dense, <500 words), .cursorrules (conventions and patterns), AGENTS.md (skills and decision logs), wiki/index.md, and schema/context.json — every tool gets the artefact that matches its protocol.
Local vector search (WASM)
Client-side semantic search using Voy (75 KB WebAssembly). Find related notes by meaning, not just keywords. No server, no API calls — runs entirely in your browser.
MCP server
OrbIDE as a Model Context Protocol server. Any MCP-compatible AI tool can query your workspace: search notes, read decisions, get context packs, list active tasks — all through a standard protocol.
Stack