Cogni

It is an MCP memory server that provides entity-graph spreading-activation recall to answer cross-vocabulary multi-hop questions where plain vector search returns no relevant results. It sells to developers, data teams and IT teams in businesses that build LLM agents needing persistent connected memory. Compared with plain vector RAG it runs deterministic retrieval with no LLM or GPU in the path and returns both graph and vector results, reducing round trips to assemble a chain.

Key features

• entity-graph spreading activation recall
• dense vector search underneath graph
• deterministic retrieval with no LLM
• MCP server tools for LLMs
• remember facts with entity linking
• point-in-time and windowed recall
• estimate_effort from measured median
• compare_effort to rank plans
• record_effort to log durations
• sleep-style consolidation