Mnemo

Mnemo offers a robust memory infrastructure designed for advanced agent applications. Key features include: * Event-sourced, append-only records for data integrity * Citation-backed retrieval, tracing results to original sources * Hybrid semantic and lexical search with consistent performance * Model-agnostic architecture for flexible integration * US data residency and enterprise-ready capabilities This platform ensures that agents have access to a rich, auditable memory, moving beyond simple key-value storage. It supports detailed ingestion of conversations, documents, and transcripts, performing atomic fact extraction to decompose content into granular, provable facts. The advanced retrieval system combines seven signals (vector, BM25, fact, temporal, entity-graph, concept-boost, semantic-bridge) fused via Reciprocal Rank Fusion (RRF) for highly accurate and contextual recall. Mnemo is built for agents that demand sophisticated memory functions, including temporal reasoning ('last week', 'before the migration') and an agentic refinement loop that re-queries on low-confidence results. Its BYO-Model architecture allows integration with various language models like OpenAI and Anthropic, preventing vendor lock-in. The system boasts an 85.2% score on LongMemEval-S, demonstrating its effectiveness in complex memory tasks. Ideal for developers and engineering teams building sophisticated agent applications that require reliable, verifiable, and performant long-term memory. It's available through TypeScript and Python SDKs, with multiple framework adapters, enabling seamless integration into existing developer workflows and agent frameworks.