@presidio-dev/agent-memory

real long-term memory for ai agents. not rag. not a vector db. self-hosted.

⚠️ About This Package

This is a Presidio fork of OpenMemory (originally by nullure), based on version 1.3.2. We maintain this fork to add custom features and enhancements specific to our use cases while staying up-to-date with upstream improvements.

Original Project: OpenMemory by nullure
License: Apache 2.0
This Fork: https://github.com/kishan0725/OpenMemory

This package provides the same powerful cognitive memory engine as OpenMemory, with additional features and improvements by Presidio.

@presidio-dev/agent-memory is a cognitive memory engine for llms and agents.

• 🧠 real long-term memory (not just embeddings in a table)
• 💾 self-hosted, local-first (sqlite / postgres)
• 🧩 integrations: mcp, claude desktop, cursor, windsurf
• 📥 sources: github, notion, google drive, onedrive, web crawler
• 🔍 explainable traces (see why something was recalled)

your model stays stateless. your app stops being amnesiac.

quick start

npm install @presidio-dev/agent-memory

import { Memory } from "@presidio-dev/agent-memory"

const mem = new Memory()
await mem.add("user likes spicy food", { user_id: "u1" })
const results = await mem.search("food?", { user_id: "u1" })

drop this into:

• node backends
• clis
• local tools
• anything that needs durable memory without running a separate service

that's it. you're now running a fully local cognitive memory engine 🎉

📥 sources (connectors)

ingest data from external sources directly into memory:

const github = await mem.source("github")
await github.connect({ token: "ghp_..." })
await github.ingest_all({ repo: "owner/repo" })

available sources: github, notion, google_drive, google_sheets, google_slides, onedrive, web_crawler

features

✅ local-first - runs entirely on your machine, zero external dependencies
✅ multi-sector memory - episodic, semantic, procedural, emotional, reflective
✅ temporal knowledge graph - time-aware facts with validity periods
✅ memory decay - adaptive forgetting with sector-specific rates
✅ waypoint graph - associative recall paths for better retrieval
✅ explainable traces - see exactly why memories were recalled
✅ zero config - works out of the box with sensible defaults

cognitive sectors

openmemory automatically classifies content into 5 cognitive sectors:

| sector | description | examples | decay rate | |--------|-------------|----------|------------| | episodic | time-bound events & experiences | "yesterday i attended a conference" | medium | | semantic | timeless facts & knowledge | "paris is the capital of france" | very low | | procedural | skills, procedures, how-tos | "to deploy: build, test, push" | low | | emotional | feelings, sentiment, mood | "i'm excited about this project!" | high | | reflective | meta-cognition, insights | "i learn best through practice" | very low |

configuration

environment variables

# database
OM_DB_PATH=./data/om.db              # sqlite file path (default: ./data/openmemory.sqlite)
OM_DB_URL=sqlite://:memory:          # or use in-memory db

# embeddings
OM_EMBEDDINGS=ollama                 # synthetic | openai | gemini | ollama
OM_OLLAMA_URL=http://localhost:11434
OM_OLLAMA_MODEL=embeddinggemma       # or nomic-embed-text, mxbai-embed-large

# openai
OPENAI_API_KEY=sk-...
OM_OPENAI_MODEL=text-embedding-3-small

# gemini
GEMINI_API_KEY=AIza...

# performance tier
OM_TIER=deep                         # fast | smart | deep | hybrid
OM_VEC_DIM=768                       # vector dimension (must match model)

# metadata backend (optional)
OM_METADATA_BACKEND=postgres         # sqlite (default) | postgres
OM_PG_HOST=localhost
OM_PG_PORT=5432
OM_PG_DB=openmemory
OM_PG_USER=postgres
OM_PG_PASSWORD=...

# vector backend (optional)
OM_VECTOR_BACKEND=valkey             # default uses metadata backend
OM_VALKEY_URL=redis://localhost:6379

programmatic usage

import { Memory } from '@presidio-dev/agent-memory';

const mem = new Memory('user-123');  // optional user_id

// add memories
await mem.add(
    "user prefers dark mode",
    {
        tags: ["preference", "ui"],
        created_at: Date.now()
    }
);

// search
const results = await mem.search("user settings", {
    user_id: "user-123",
    limit: 10,
    sectors: ["semantic", "procedural"]
});

// get by id
const memory = await mem.get("uuid-here");

// wipe all data (useful for testing)
await mem.wipe();

performance tiers

• fast - synthetic embeddings (no api calls), instant
• smart - hybrid semantic + synthetic for balanced speed/accuracy
• deep - pure semantic embeddings for maximum accuracy
• hybrid - adaptive based on query complexity

mcp server

@presidio-dev/agent-memory includes an mcp server for integration with claude desktop, cursor, windsurf, and other mcp clients:

npx @presidio-dev/agent-memory serve --port 3000

claude desktop / cursor / windsurf

{
  "mcpServers": {
    "agent-memory": {
      "command": "npx",
      "args": ["@presidio-dev/agent-memory", "serve"]
    }
  }
}

available mcp tools:

• openmemory_query - search memories
• openmemory_store - add new memories
• openmemory_list - list all memories
• openmemory_get - get memory by id
• openmemory_reinforce - reinforce a memory

examples

// multi-user support
const mem = new Memory();
await mem.add("alice likes python", { user_id: "alice" });
await mem.add("bob likes rust", { user_id: "bob" });

const alicePrefs = await mem.search("what does alice like?", { user_id: "alice" });
// returns python results only

// temporal filtering
const recent = await mem.search("user activity", {
    startTime: Date.now() - 86400000,  // last 24 hours
    endTime: Date.now()
});

// sector-specific queries
const facts = await mem.search("company info", { sectors: ["semantic"] });
const howtos = await mem.search("deployment", { sectors: ["procedural"] });

api reference

new Memory(user_id?: string)

create a new memory instance with optional default user_id.

async add(content: string, metadata?: object): Promise<hsg_mem>

store a new memory.

parameters:

• content - text content to store
• metadata - optional metadata object:
  • user_id - user identifier
  • tags - array of tag strings
  • created_at - timestamp
  • any other custom fields

returns: memory object with id, primary_sector, sectors

async search(query: string, options?: object): Promise<hsg_q_result[]>

search for relevant memories.

parameters:

• query - search text
• options:
  • user_id - filter by user
  • limit - max results (default: 10)
  • sectors - array of sectors to search
  • startTime - filter memories after this timestamp
  • endTime - filter memories before this timestamp

returns: array of memory results with id, content, score, sectors, salience, tags, meta

async get(id: string): Promise<memory | null>

retrieve a memory by id.

async wipe(): Promise<void>

⚠️ danger: delete all memories, vectors, and waypoints. useful for testing.

license

apache 2.0

links

• main repository
• python sdk
• vs code extension
• documentation
• discord