Agent memory / Web3
MNEME
Portable memory for AI agents: persistent context across conversations and models, with a verifiable deletion proof on Monad.
MNEME is a memory layer that gives AI agents persistent context across conversations and models. I co-built it in a three-person team and led full-stack and blockchain development. It won Monad Blitz Pune in July 2026.
I built the Fastify API, pgvector semantic recall, the TypeScript integration and the Monad EVM components. The architecture includes batch attestations, deletion tombstones and a memory-pack market. The hosted MVP runs on Render and Aiven PostgreSQL with local embeddings, and I added a remote MCP endpoint, a Python SDK and the Context Router Chrome extension. By 8 October the project had 133 tests across packages.
The working hosted MVP does not mean every roadmap feature is deployed. Hosted encryption is at rest, not end-to-end, and the sample demo memories are not live blockchain transactions. Next: faster, smaller retrieval, more integrations and model-independent benchmarks.