Local memory → governed execution → verifiable continuity
I build infrastructure for AI agents that can remember what happened, know what they were authorized to do, emit signed records of what they report doing, and show when their authority ended — with evidence that stays local, portable, and independently verifiable.
whitemagic local-first memory & session continuity (MCP, Rust, MIT)
│
▼
continuity-receipt portable, offline-verifiable evidence (open spec, Apache-2.0)
│
▼
mandalaos-gate-lite bounded agent execution that emits receipts (review snapshot)
│
▼
whitemagic-plugins client integrations & adapters
Each repository stands on its own — WhiteMagic does not require MandalaOS, and Continuity Receipt does not require WhiteMagic.
| Want to… | Start at |
|---|---|
| Use it — memory and continuity for your coding agent | whitemagic |
| Review a protocol — spec, vectors, three verifier implementations (Python, Rust, TypeScript) | continuity-receipt |
| Attack the architecture — containment review snapshot | mandalaos-gate-lite |
Working notes and docs: whitemagic.dev
Stable Gen2 9.3.4 remains the default installation line; newer prereleases are an explicit preview. Start with the setup guide or the receipt consumer quickstart. Signatures bind signed bytes; source truth, real-world authority, completeness and successful execution require the consumer's own policy and evidence.


