Agent Memory — Market

Verified, sourced market claims for Agent Memory. Every fact below traces back to a primary source.

Verified facts

• Major agent platforms increasingly bundle a dedicated memory provider rather than relying on raw vector search — exemplified by AWS selecting Mem0 for its Agent SDK. ↗ (other)
• Mem0 is the exclusive memory provider for the AWS Agent SDK and is among the most-adopted open-source agent-memory layers. ↗ _(historical_event)_
• Mem0 raised ~$24M to build a persistent memory layer for AI agents. ↗ (financial)
• Letta (formerly MemGPT) raised a ~$10M seed at a ~$70M post-money valuation to build OS-style tiered memory for agents. ↗ (financial)
• Zep provides temporal knowledge-graph memory for agents through its open-source Graphiti engine. ↗ _(technical_spec)_
• Agent memory is now benchmarked separately from vector-database retrieval — using suites like LongMemEval and LoCoMo — signaling the market treats stateful memory and RAG as distinct layers. ↗ (other)
• The agent-memory category — temporal knowledge graphs, OS-style memory tiers, and durable cross-session state — emerged as a distinct middleware layer in 2026, separate from vector databases. ↗ (other)
• Mem0, Zep, and Letta are the three most-cited agent-memory frameworks in 2026 developer comparisons. ↗ (other)
• Unlike vector databases that store embeddings for retrieval, agent-memory systems manage evolving, time-aware state about users and tasks across sessions. ↗ (other)
• Letta's MemGPT lineage popularized 'memory tiers' — paging information between a limited context window and external stores like an operating system. ↗ _(technical_spec)_