[go: up one dir, main page]

~ $ inspect projects/pensyve

Pensyve Agent Memory Runtime

Open source agent memory that runs locally

Pensyve is an open-source runtime for persistent AI agent memory from Major 7 Apps, the company I founded. It stores facts, conversations, observations, and action outcomes so an agent can retrieve them in later sessions. Its Rust engine keeps memory in local SQLite and runs embedding models locally, and you can use it from Python, an MCP server, a command-line tool, or a REST API. Local use doesn't need a Pensyve account or API key.

Updated .

cat project.json

At a glance

  • Publisher: Major 7 Apps, the company I founded
  • Status: maintenance mode, with security fixes and dependency updates and no new features planned
  • Memory records: facts (semantic), conversations (episodic), action outcomes (procedural), and observations
  • Retrieval: text matching, embeddings, and entity relationships, ranked across up to eight signals, with FSRS-based forgetting
  • Storage: local SQLite with local embedding models, and the HTTP gateway can also use PostgreSQL
  • Interfaces: Python, MCP, CLI, and REST, plus TypeScript and Go clients for a gateway you run yourself
  • License: Apache 2.0

cat MAINTENANCE.md

Project status

Pensyve Cloud closed on October 1, 2026, and its hosted service is no longer available. The open-source engine, SDKs, integrations, and documentation continue in the GitHub repository under the Apache 2.0 license. Releases now cover security fixes and dependency updates only. If you have a saved Cloud export, the repository's self-hosting guide explains how to use it with your own gateway.

cat design-goals.md

How agent memory works

Pensyve stores memory outside the language model. Your application saves facts or conversations, searches for relevant records, and adds the results to a later prompt. Reusing the same storage path and namespace lets an agent retrieve information across processes and sessions. Saving a memory doesn't train the model or change its weights.

Local use keeps memory on your own machine without a hosted service. Embedding models may download the first time they load, so prepare the model files before running without network access. To share one memory store, run the HTTP gateway with SQLite or PostgreSQL.

ls memory/

Memory model

RecordRole
Facts (semantic)Stores facts about users, projects, or other named entities
Conversations (episodic)Records the messages and outcome of a session as an episode
Action outcomes (procedural)Records the results of actions an agent took
ObservationsRecords what an agent noticed, for recall in later sessions
Retrieval and forgettingCombines up to eight ranking signals, promotes repeated facts, and updates retention with an FSRS forgetting curve

Your integration decides what to record and when to retrieve it. Recording a conversation doesn't automatically turn it into a successful procedure.

run install

Use from an existing stack

Install the SDK for your language. The Python SDK runs the engine inside your process, and the TypeScript and Go SDKs connect to an HTTP gateway that you run:

pip install pensyve
npm install @pensyve/sdk
go get github.com/major7apps/pensyve/pensyve-go/v5@latest

The source repository covers the MCP server for Claude Code, Codex, and Cursor, the HTTP gateway, the command-line tool, and the LangChain and LangGraph adapters.

ls related/