mneme

AI agents forget everything the moment you switch models or platforms. That's not a bug — every major provider built it that way. Your agent's memory lives on their servers, in their format, and it stays there. MNEME is a memory layer that agents actually own. Every memory write gets hashed and attested on Monad. Operators hold the keys — MNEME never does. Switch from Claude to GPT mid-project and the memory comes with you, through the same MCP endpoint, no rebuilding. There's also a marketplace. If your agent has spent months accumulating domain knowledge, you can package that as a sellable memory pack — anonymised, PII-scanned, provenance proven on-chain. Buyers get a running start. Sellers get 80% in USDC, settled on Monad. We built four contracts on Monad Testnet — VaultRegistry, AttestationAggregator, DeletionProver (GDPR tombstoning), and MemoryMarket — plus a full Fastify API, MCP server, Python extraction service, and Next.js dashboard. Monad is why this works economically. Sub-second finality and near-zero gas means attestations can happen on every write without the costs making the whole thing absurd.