AI Agent Memory Standard 'AMP' Released

To solve the Memory Fragmentation problem that occurs during AI agent development, the Agent Memory Protocol (AMP), built on top of MCP (Model Context Protocol), has been released.

Currently, AI agents suffer from poor portability because backends such as Mem0, smriti-memcore, and various vector DBs each use different APIs and schemas. AMP unifies these into a standardized interface, allowing developers to swap between different memory backends without modifying their agent code.

AMP's 6 Core Verbs:

• amp.encode: Store a new memory
• amp.recall: Retrieve relevant memories
• amp.forget: Permanently delete a memory
• amp.consolidate: Trigger backend reorganization and summarization
• amp.pin: Permanently pin an important memory
• amp.stats: Check backend status and usage statistics

The protocol defines Core and Full compliance levels, and comes with a JSON schema, a compliance test suite, and a production-ready reference implementation (amp-server) that wraps smriti-memcore. This allows developers to use various backends such as Chroma, Pinecone, pgvector, and Zep in a standardized way.