tapps-brain-memory

Persistent cross-session memory for AI agents is achieved by capturing interaction facts and recalling them automatically into model prompts using BM25 ranking and exponential decay, ensuring context survives session boundaries.

Multi-agent Hive memory sharing is handled through federation features that allow cross-agent and cross-project context recall, consolidation, and profile-driven workflows for collaborative memory management.

BM25 ranking scores stored memories by relevance to the current prompt, while exponential decay gradually reduces the weight of older memories, ensuring the most contextually important facts are assembled first.

Configurable profiles allow you to customize scoring, exponential decay rates, and sharing rules, tailoring how memories are captured, consolidated, and recalled for specific project or agent requirements.

Pre-compaction flush preserves important context by triggering a lifecycle hook that saves critical memories to durable indexing before memory pruning or disposal occurs during the compact phase.

MCP tools and resources provide programmatic control and diagnostics for persistent memory, exposing lifecycle hooks like bootstrap, ingest, assemble, compact, and dispose for automated agent context management.