Ядро Памяти (Memory Core)

Ядро Памяти (Memory Core) is a memory engine designed for AI bots that interact with users, aiming to address the problem of bots forgetting information between sessions. The platform provides persistent memory by storing user data across three distinct layers: a hot layer using Redis for instant access to recent facts and active context, a warm layer utilizing Qdrant for semantic search via vector embeddings, and a cold layer based on Neo4j for knowledge graph storage, capturing relationships and long-term insights. This architecture enables bots to recall information from the entire conversation history, personalize responses, and automatically connect facts through a knowledge graph.

The service offers semantic search capabilities, allowing queries to be matched by meaning rather than keywords. It features automatic classification and routing of content to the appropriate storage layer through DRUMS classification, and supports knowledge graph construction for uncovering non-obvious relationships between facts. Multi-tenancy is supported, with strict data isolation by API key, and all data is stored on servers in Saint Petersburg, Russia, in compliance with Russian personal data regulations (FZ-152). An experimental "Wiki Mode" automatically compiles a bot's memory into Notion-like pages and keeps them updated in the background.

Memory Core is accessible via a unified API, eliminating the need for users to set up Redis, Qdrant, or Neo4j themselves. Developers can connect their bots with just a few lines of code and obtain an API key within minutes. The platform targets any bot that communicates with users, including applications in hospitality, e-commerce, healthcare, and education, where remembering user preferences, histories, and interactions enhances personalization and service quality.

Pricing includes a free tier with basic support, limited to one bot, 500 records, and 100 requests per day, covering the hot and warm storage layers. Paid plans offer increased limits, access to all three storage layers including the knowledge graph, priority support, and features like Wiki Mode. 9% availability, while an enterprise option allows for dedicated instances and installation on the client's own server. А.