Core concepts

The fundamental storage primitive for multi-agent systems. One datapack per application, or one project. It is the isolation unit for agents working together. Each datapack bundles a database, the knowledge base, agent memories, collective memories, shared knowledge, MCP server, and credentials.

Extracted memories from conversations are specific pieces of information or preferences that are explicitly stored persistently for the long term. Memories are stored as entity and relationship, powered by Meko's graph storage, also called MAGE. Agent memories can be promoted to a shared knowledge base, so all agents can learn from each other's learnings.

Information repository that acts as a single source of truth for agents, grounding decisions, reducing hallucinations, and ensuring consistent responses. Uses semantic search and RAG to understand query intent, not just keywords. Documents are chunked, embedded, and indexed for retrieval.

Meko provides end-to-end visibility into every conversation, from initial prompt through LLM chain-of-thought, including memory and knowledge additions and search, the associated API and SQL calls, the queries executed, and the results retrieved.