Personal Twin

Open, AI-friendly source repository for a personal digital twin.

The project keeps canonical measurements separate from editor-specific 3D files and web delivery assets.

• Canonical data: JSON validated by JSON Schema
• Space authoring: Sweet Home 3D (.sh3d)
• Body authoring: MakeHuman / MPFB + Blender
• Scene composition: Blender (.blend)
• Web delivery: glTF / GLB
• AI access: structured JSON first; editor MCP integrations for scene operations

Principles

1. Data is the source of truth. 3D meshes are views of canonical measurements.
2. Prefer open, portable formats.
3. Keep schemas, units and provenance explicit for AI/tooling.
4. This repo is public; real personal measurements and detailed home geometry are private by default.
5. Public derivatives should be portable to a static website.

Repository layout

• schemas/ - canonical JSON Schema contracts
• examples/ - sanitized examples committed to Git
• data/private/ - real personal measurements (gitignored)
• source/ - authoring conventions and intentionally public source assets
• exports/ - web-facing derived assets
• scripts/ - validation/export automation
• docs/ - architecture and measurement conventions

First vertical slices

Space

1. Measure one real room.
2. Build it in Sweet Home 3D.
3. Record canonical dimensions in a private JSON file.
4. Export a public GLB only when desired.
5. Use it for a real purchase/layout decision.

Body

1. Measure a minimal body profile.
2. Store measurements with provenance and date.
3. Build an approximate avatar with MakeHuman/MPFB.
4. Compose/export through Blender.
5. Use structured measurements for fit decisions and the mesh for visualization.

V1 web workbench

V1 keeps mature editors but avoids streaming the room editor as video:

• SweetHome3DJS 7.5.2 runs natively in the browser with WebGL for normal room/furniture editing.
• Sweet Home 3D desktop remains headless/maintenance-side and exposes Sweet Home 3D MCP for AI scene operations.
• Blender + MPFB remains in the maintenance Webtop for body/avatar generation and advanced 3D work.
• Blender MCP exposes body/general 3D scene operations to AI.
• Personal Twin Memory MCP stores durable AI notes and exposes canonical JSON reads.
• Tailscale Serve provides private browser/MCP access.

Start with ./scripts/workbench-up.sh and verify with ./scripts/workbench-smoke.sh. Startup deterministically regenerates the local/tailnet MCP registry; smoke verification includes reversible Sweet Home 3D and Blender write-path checks plus Memory MCP CRUD. See docs/workbench.md for ports, runtime layout and operating rules.

Validate examples

python3 -m pip install -r requirements-dev.txt
python3 scripts/validate_json.py

Privacy

Do not commit precise personal measurements, detailed residential geometry, private photos, scans or body scans by default. Put local records under data/private/ and private 3D sources under source/private/.

License

Code, schemas and documentation are MIT unless a file states otherwise. Third-party assets retain their original licenses.