Dionysus

Dionysus is a voice-first, evidence-grounded laboratory for building a human-reviewed portrait of a person over time.

The repository does not contain a person. It contains the open protocols, schemas, and projection formats needed to conduct reflective interviews, retain the provenance of candidate claims, expose disagreement, and produce small context views for deliberate use.

Current state

This is a skeleton, not a finished interview system.

• interviews/ defines the first interview family and the conversation contract.
• schemas/ separates an interview session from a portrait claim.
• portrait/ contains disposable projection templates, never an authoritative dossier.
• vault/ documents the private boundary. Its contents are ignored by Git.
• legacy/seed-garden/ preserves the former Dionysus seed-garden repository intact.

No voice transport, transcription provider, agent runtime, or personal dataset is selected yet.

Core rule

Conversation is evidence collection, not automatic truth production:

voice or text conversation
  -> transcript and evidence slices
  -> candidate claims
  -> human accept / edit / contest / retire
  -> purpose-bounded portrait views

Claims must remain traceable to evidence and review. Derived portraits must be small enough to inspect, must cite claim IDs, and must never silently become a generic memory authority.

Start here

1. Read DESIGN.md.
2. Review docs/PRIVACY.md.
3. Inspect the interview family in interviews/catalog.toml.
4. Run python scripts/validate_skeleton.py.

The former seed garden remains available at legacy/seed-garden/README.md and at the pre-archive-2026-07-23 Git tag.

License

Apache-2.0. See LICENSE.