We study the primitives of agent cognition: how should an agent decide what to remember, when a memory has become stale, how to retrieve the right context, and how to plan and reason about multi-step goals.
Our research informs our products directly — every pipeline stage in Recall exists because we could not find a satisfactory answer in the existing literature.
Typed, temporal-aware memory layer. Seven-stage write pipeline rejects noise before storage. Five-retriever hybrid search fuses semantic, keyword, graph, temporal, and type signals via RRF.
Goal-directed execution layer. Decomposes goals into DAGs of steps with subgoals, expected tools, and risk types. Maps steps to MCP tool calls, infers parameters from Recall context, replans on failures.
Structured, policy-aware reasoning for multi-step workflows. Combines chain-of-thought, self-consistency, Recall retrieval, and MCP tool coordination. Grounds answers in memories and tool outputs.