Kyros

Kyros functions as a mediation layer that connects AI agents to vector and graph databases. Agents can store memories through simple API calls that handle embedding generation, hashing for integrity, and assignment of decay weights. Retrieval uses similarity search combined with retention scoring to surface only pertinent context for prompts. The architecture supports both direct SDK usage in Python and TypeScript as well as a proxy mode that intercepts requests to common LLM providers. This allows context injection without modifying existing agent code. Built-in support for the Model Context Protocol further enables integration with various development environments. Framework connectors exist for popular libraries so that memory operations occur automatically during agent execution. All stored items remain auditable through Merkle structures while the system filters out low-relevance entries based on configurable thresholds.