~ $ inspect projects/pensyve

Pensyve Agent Memory Runtime

Open source, offline first, framework agnostic, protocol native

Pensyve is Major 7 Apps' universal memory runtime for AI agents. It gives applications episodic, semantic, and procedural memory; ranks recall with an 8-signal retrieval pipeline; models forgetting with FSRS; and runs locally by default while exposing Python, TypeScript, Go, MCP, REST, and CLI interfaces.

Updated .

At a glance

  • Publisher: Major 7 Apps, founded by Seth Hobson
  • Memory types: episodic, semantic, and procedural
  • Retrieval: 8-signal fusion with FSRS-based forgetting
  • Storage: local SQLite and ONNX by default, with optional PostgreSQL
  • Interfaces: Python, TypeScript, Go, MCP, REST, and CLI
  • License: Apache 2.0

Why an agent memory runtime

Agent applications often lose useful context between sessions or couple memory to a single framework. Pensyve separates storage, retrieval, reinforcement, and forgetting into a reusable runtime that can serve different agent stacks through standard interfaces.

The offline-first path keeps local applications independent of a hosted service. Teams that need shared or larger deployments can use PostgreSQL while retaining the same protocol-facing model.

Memory model

MemoryRole
EpisodicRecords events and experiences that an agent may need to recall later
SemanticStores durable facts and concepts independently of one conversation
ProceduralPreserves learned approaches and outcomes for repeated work
Retrieval and forgettingCombines eight ranking signals and FSRS scheduling to surface useful memories over time

Use from an existing stack

Pensyve can be integrated through a language SDK or connected as an MCP server. The canonical repository documents local and PostgreSQL-backed deployment, REST and CLI access, and installation for each supported language:

pip install pensyve
npm install @pensyve/sdk
go get github.com/major7apps/pensyve/pensyve-go/v3@latest

See the source repository and Pensyve MCP setup guide for current details.