Agent Memory
Store and query persistent working, archival, and entity memory for your AI agents.
Tier 1: Short-Term
Working Memory (`_pb_sessions`)
Fast relational checkpointer storing immediate conversation turns. Automatically summarizes and spills over to archival when the LLM context window reaches capacity.
Latency: <2msRelational SQL
Tier 2: Long-Term
Archival Lake (`_pb_archival`)
Durable vector lake where past interactions and documents are auto-embedded via Pulsbase Neural Embedding Engine and retrieved via Hybrid Reciprocal Rank Fusion.
Latency: <8ms768-dim HNSW Index
Tier 3: Temporal
Entity Graph (`_pb_entities`)
Structured knowledge graph tracking temporal facts about users, user preferences, and evolving state across multi-agent workflows.
Latency: <3msDynamic JSON Extract
Target Cluster:
No Active Agent Memory Pools Indexed
Connect your autonomous AI agent via MCP or universal SDKs to begin indexing working turns and long-term semantic embeddings.