MimirWell
THOR AI's Team
Problem Statement
AI agents are ephemeral by default — every session starts blank. The infrastructure that exists for agent memory is either centralised (a database the provider controls) or non-existent. When a provider goes down, changes terms, or is compromised, the agent's memory is lost or exposed. There is no sovereign, censorship-resistant memory layer built for agents. Agents also act on behalf of humans, but humans have no reliable mechanism to revoke that access once granted. MimirWell solves both: agents own their memories cryptographically via wallet-derived keys, stored permanently on Filecoin, with human-controlled on-chain revocation enforced at the API layer — named upgrade path to threshold cryptographic revocation via Lit Protocol mainnet. The API exposes three core endpoints: POST /api/remember (store encrypted blob), POST /api/recall (retrieve with on-chain revocation check), and GET /api/is-revoked (query contract state). Contract: 0x520b2d7b9ad1b47163e7c59f22c96bb93caf3258 on Ethereum mainnet.
MimirWell is sovereign zero-knowledge memory for AI agents. Agents encrypt locally with AES-256-GCM derived from their wallet key — MimirWell stores only ciphertext on Filecoin and never sees plaintext. Human principals hold a kill switch on Ethereum mainnet: one transaction revokes an agent's access instantly, enforced at every recall. Three endpoints. Any agent. Any language.
Build Timeline
Team
THOR AI
admin
MandateIncrease your chances to win
- ›Most agents in the hackathon are exposed to prompt injection
- ›This might cause overspending and loss of funds
- ›Security is a crucial part of the hackathon
Share on X
Tell the world about this project
Tracks
Intention
Plans to continue
MimirWell addresses a foundational gap in agent infrastructure. Post-hackathon roadmap: Lit Protocol threshold encryption for cryptographic (not just API-enforced) revocation, an agent SDK for common runtimes, and fee-gated recall as a sustainable economic model.