<aside> ⚠️

Outdated — see Documentation for the current ArchHub user guide. The feature list and status below describe v1.6.3.

</aside>

Graph-first AI workspace for AEC. Every host, conversation, document, skill, and tool call is a typed node on a canvas. Wire them together with typed bridges that carry real data. Save the canvas as a Skill — composable, shareable, your firm owns it.

The public face of ArchHub. Source code lives on GitHub; this Notion is the pitch + docs + pricing.

Quick links

• 📥 Download installer (Windows): latest release
• 💾 Source code (MIT): github.com/Fargaly/ArchHub
• 📖 User guide: USER_GUIDE.md
• 💬 Issues / requests: GitHub issues
• 🔒 Security: SECURITY.md

What ArchHub does

• One canvas, every tool. 18 host families as of v1.4: Revit, AutoCAD, 3ds Max, Blender, Rhino, Speckle, Outlook, Teams, Notion, LM Studio, Antigravity, Photoshop, Illustrator, InDesign, Word, Excel, PowerPoint, Dropbox — all live on the same graph as drag-and-drop nodes.
• Typed wires carry real data. A Revit view socket plugs into a list_walls node. The wall list flows into a filter, then a dimension placer. Source-field selectors pluck sub-properties out of the payload. Cycles refused at edit time. Wire engine: 28 px snap, hover preview (green ok / red refused), drop-on-node-body auto-port-pick, refusal toast with reason.
• Every node is its own MCP server. Other nodes — or external agents — call its tools by JSON-RPC. The graph is a fully addressable runtime.
• Houdini-style cooking. Lazy + dirty + cached. Freeze any node to lock its output while iterating downstream.
• Subgraph = Skill. Select nodes → Cmd-G → composite node. Save it. Drag it back onto any canvas as a single block.
• Session = canvas. Each saved session is its own graph blob. New session = blank canvas. Switching sessions remounts the whole tree.
• LLM-agnostic. OpenRouter, Anthropic, OpenAI, Google, local LM Studio. Per-task routing. Your keys, your data. Native PyQt SettingsDialog → 5 tabs: Providers / Memory / Profile / Storage / Shortcuts.
• Local-first. Runs on your machine. The cloud backend is live and opt-in: sign-in, session sync, team seats and per-user brain replicas.