# Dollhouse Deviants

- **Event:** [Meta Global AI Developer Hackathon](https://cerebralvalley.ai/e/meta-ai-developer-hackathon)
- **When:** Oct 26, 12:00 PM – Nov 4, 9:00 PM (EST)
- **Where:** Online
- **Team:** [Sonja Kelley](https://cerebralvalley.ai/u/Sonja-On-Fire)
- **GitHub:** https://github.com/musclehed03/hive-mind-mcp
- **Demo video:** https://rumble.com/v7hueg0-hive-mind-your-ai-finally-remembers-never-restate-yourself-hive-mind-mcp-de.html
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1575

Hive-Mind is a provider-neutral persistent memory layer for every AI you use — "Never Restate Yourself." Every AI assistant starts from zero and forgets what the last one knew, so people restate themselves to each one, every time. Hive-Mind ends that: one local-first memory store (SQLite + JSONL + Markdown vault) that any MCP client can read and write, so a fact captured once is available to every assistant, forever, without being retyped or re-explained.
Every capture passes a Burn-After-Reading scrubber before anything is stored or sent — API keys, passwords, and secrets are trapped and counted, never saved. Cross-Hatch verification checks each fact for consistency before it is trusted, and contradictions never overwrite silently: they wait for an explicit human yes, with the change written down. A Caveman shorthand layer compresses recall to save tokens, and a living nebula graph grows from a dust cloud to a Radiant Sun as memories arrive and get verified.
Accessibility and security are version-one features, not add-ons. The interface is built for neurodivergent users as well as everyone else: one-flick presets that change font, color, and spacing together for common dyslexic and ADHD needs without a settings maze, with finer controls underneath for those who want them — including dyslexia letter-spacing kept moderate and capped, because spacing pulled too wide breaks letter grouping. Typed color-blind palettes cover every type, a pure-grayscale achromatopsia mode verified by a grayscale-only user keeps every meaning readable without color, and fast tooltips explain every control in plain words. On security: storage is local-first, and the snapshot backup is a private, secure backend the user sets up themselves — nothing is hosted publicly, and no snapshot ever leaves the machine to a cloud the user did not configure with their own keys. Ultimate access always stays with the end user: the user's own keys and approvals gate every external connection, and nothing is sent anywhere without the scrubber and the user's own consent.

## More from Meta Global AI Developer Hackathon

- [TailorApps](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1572)
- [Letlatsa Pheane](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1573)
- [Hồi Sau](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1574)
- [Rimsha Tariq](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1576)
- [am solo builder](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1577)
- [Alrevion](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1578)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1575. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt