federated-memory

What problem does it solve?

Federated-memory helps an AI agent retain and recall important information across sessions by storing knowledge in a searchable memory system and enabling read-only sharing across multiple agents without unsafe write sharing.

Core Features & Use Cases

• Query-layer memory federation: lets multiple agents access each other’s stored knowledge via read-only MCP tools while preventing conflicting cross-writes.
• Tiered memory architecture: routes information to working memory, semantic recall over past conversations, and permanent archival knowledge/decisions.
• Vector + optional graph retrieval: uses Qdrant or pgvector for semantic search and adds a graph layer (FalkorDB/Neo4j) when multi-hop relationship queries are required.
• Operational guardrails: includes embedding selection and migration strategy, dedup via content hashing, and retention policies to stop memory from growing unbounded.
• Use Case: When an agent can’t remember prior decisions about “deployment rollback policy,” use semantic search to retrieve those decisions and ensure a consistent answer across future sessions and agent teams.

Quick Start

Ask your AI to design a federated memory setup that uses pgvector for archival recall, deduplicates entries with content hashes, and exposes MCP tools for read-only cross-agent memory search.