OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
Agent Memory & RAG

Mnemosyne

Mnemosyne is a TypeScript library that provides persistent memory for AI agents, including a local SQLite engine for storing, correcting and recalling evidence-linked records across agent work.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Mnemosyne — OpenViST project profile in Agent Memory & RAG.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use Mnemosyne when an agent needs persistent, traceable memory and its downstream guidance should respond to corrections. Choose the Local Engine for a self-contained SQLite store, or the Vector Engine when the application uses its service-backed architecture.

Who it is for

  • Developers building AI agents that need memory across sessions.
  • Teams that need to track source evidence and retire dependent guidance after corrections.
  • Developers connecting agent memory through TypeScript, Model Context Protocol, CLI or authenticated HTTP.

Documented uses

  • Store agent facts and procedures with source details, visibility and trust labels.
  • Compile relevant memory into context for an agent while exposing citations and uncertainty.
  • Correct a record and invalidate dependent procedures that are no longer current.
  • Hand off goals, decisions and next steps between agents.
  • Connect an existing memory store through a read-only bridge for gradual migration.
Before you choose

Requirements & limits

The repository identifies an MIT license. Version 2.0.0-rc.9 is a release candidate; the README says stable npm latest remains on 1.x and the 2.0 candidate is available through the next channel.

  • The Local and Vector Engines have separate schemas and do not automatically synchronize or migrate data.
  • The Vector Engine requires configured external services; the Local Engine uses SQLite and does not require a model API key, database server or Docker.
  • Dependencies must be declared by the controller for corrections to invalidate dependent procedures.
  • The observed trust label records that the host recorded a source; it does not certify the source as true.
  • The default context counter measures UTF-8 bytes conservatively, not a model tokenizer.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Run the Local Engine demonstration

Install the version 2.0.0-rc.9 candidate and run its documented demos.

Before you start

  • Node.js 22.16 or later; Node 24 is recommended.
  • An npm project environment.

Steps

  1. Install the exact candidate with the command below.

    Documented command
    npm install mnemosy-ai@2.0.0-rc.9

  2. Run the command below.

    Documented command
    npx --no-install mnemosy demo

  3. Run the command below.

    Documented command
    npx --no-install mnemosy learning-demo

Expected outcome

The first demo shows a handoff, correction, private scope and forgetting; the learning demo exercises capture, scripted observation, skill trials, sharing and correction-driven retirement. The README says these use local APIs and deterministic fixtures without model calls.

Guide limitations

  • This is a release candidate; the README recommends evaluating it on your own workload before production.
  • Unqualified npm install mnemosy-ai selects the stable 1.x channel, not this candidate.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
8 October 2026
Original website title
28naem-del/mnemosyne · Official homepage title · retained in its source language
How we review sourcesSuggest a correction