Pensyve Documentation
Pensyve is a universal memory runtime for AI agents. Store, recall, and manage episodic, semantic, and procedural memories with vector + BM25 + graph retrieval, automatic decay, and multi-agent access control.
Quickstarts
Pick your integration and get running in under 5 minutes.
- Python — PyO3 native module, fastest path
- TypeScript — HTTP client SDK with retry and timeout
- REST API — FastAPI server, language-agnostic
- Go — HTTP client with
context.Context - MCP — Stdio transport for Claude, Cursor, and other MCP hosts
- CLI —
pensyve recall,pensyve rememberfrom your terminal
Key Concepts
Namespaces provide top-level isolation. Each namespace gets its own storage, embeddings, and graph — use them to separate projects, environments, or tenants.
Entities represent the actors in your system: agents, users, teams, or tools. Every memory belongs to an entity.
Episodes are bounded interaction sequences (like a conversation turn or tool call chain). Messages within an episode share context and have a configurable TTL.
Memories come in three types: episodic (what happened), semantic (subject-predicate-object knowledge triples with temporal validity), and procedural (Bayesian action-outcome reliability tracking).