Zekra

Zekra provides a persistent memory service for computational agents, ensuring learned information is retained across sessions. Key capabilities include: * Hybrid recall combining vector similarity and BM25 * Hot and cold memory tiers for efficient access * Reconsolidation of facts upon recall * Universal client connectivity via a single protocol This system allows agents to build a cumulative knowledge base, where facts are updated rather than duplicated, and previous learnings are recalled with citations. It intelligently fuses different search methods to ensure comprehensive and accurate information retrieval, whether for exact phrases or conceptual understanding. The service is designed to function like a biological memory system, with a fast-access 'hot tier' for recent or critical information and a 'cold tier' for long-term storage. A salience gate manages the flow of information between these tiers, consolidating important learnings. Recall is an active process; retrieving a memory not only strengthens it but also allows for corrections and updates, preventing the perpetuation of stale data. Zekra is ideal for developers creating agents that require continuous learning and long-term knowledge retention. It ensures that any agent, regardless of its client environment (e.g., Claude Desktop, Codex, Gemini), can access and contribute to a shared, evolving memory over a common communication protocol. This enables a more consistent and informed agent experience across diverse applications.