[!WARNING] This is the experimental TypeScript version of MeshOS. For the stable Python implementation, please visit Props-Labs/mesh-os.

The TypeScript version is under active development and APIs may change frequently. For production use, we recommend using the Python version.

MeshOS

The Memory & Knowledge Engine for Multi-Agent Systems

MeshOS is a developer-first framework for building multi-agent AI-driven operations with structured memory, knowledge retrieval, and real-time collaboration. Unlike generic memory stores, MeshOS is purpose-built for:

• Autonomous Agents & Teams – Agents and humans evolve a shared memory over time.
• Graph-Based Memory – Track relationships, dependencies, and evolving knowledge.
• Fast Semantic Search – Vector-based retrieval with pgvector.
• Event-Driven Execution – Automate workflows based on evolving context.
• Versioned Knowledge – Track updates, past decisions, and historical context.
• Open & Portable – Runs on PostgreSQL + Hasura with no vendor lock-in.

Why MeshOS?

Most frameworks give you a blob of memories—MeshOS gives you structured, evolving intelligence with deep relationships and versioning.

Who is MeshOS for?

✅ Builders of AI-powered operations – Structured memory and decision-making for AI-driven systems.
✅ Multi-agent system developers – AI agents that need to store, process, and evolve shared knowledge.
✅ Developers & engineers – Wanting an open-source, PostgreSQL-powered framework with no lock-in.

flowchart LR
    %% Main System
    subgraph MeshOS[MeshOS System]
        direction LR

        %% Taxonomy Details
        subgraph Taxonomy[Memory Classification]
            direction TB
            
            subgraph DataTypes[Data Types]
                direction LR
                knowledge[Knowledge Type]
                activity[Activity Type]
                decision[Decision Type]
                media[Media Type]
            end

            subgraph Subtypes[Example Subtypes]
                direction LR
                k_types[Research/Mission/Vision]
                a_types[Conversations/Logs/Events]
                d_types[Policies/Strategies]
                m_types[Documents/Images]

                knowledge --> k_types
                activity --> a_types
                decision --> d_types
                media --> m_types
            end

            subgraph Relations[Edge Types]
                direction LR
                basic[related_to/version_of]
                semantic[influences/depends_on]
                temporal[follows_up/precedes]
            end
        end

        %% Memory Operations
        subgraph MemoryEngine[Memory Operations]
            direction LR
            rememberAction[Store/Remember]
            recallAction[Search/Recall]
            linkAction[Link Memories]
            versioning[Version Control]

            rememberAction --> recallAction
            recallAction --> linkAction
            linkAction --> versioning
        end
    end

    %% Organization & Agents
    subgraph Organization[Organization & Agents]
        direction TB

        %% Company Memory
        subgraph CompanyMemory[Company-Wide Memory]
            direction LR
            corpVision[Company Vision]
            corpMission[Company Mission]
            corpData[Knowledge Base]
        end

        %% Agents
        subgraph Agent1[Research Agent]
            a1Mem[Research Memories]
        end

        subgraph Agent2[Service Agent]
            a2Mem[Service Memories]
        end
    end

    %% System Connections
    Taxonomy --> MemoryEngine
    MemoryEngine --> Organization

    %% Memory Connections
    corpVision -.->|influences| a1Mem
    corpMission -.->|guides| a2Mem
    a1Mem -.->|shares| a2Mem
    a2Mem -.->|feedback| corpData
    a1Mem -.->|versions| corpData

    %% Styling
    classDef system fill:#dfeff9,stroke:#333,stroke-width:1.5px
    classDef engine fill:#fcf8e3,stroke:#333
    classDef taxonomy fill:#e7f5e9,stroke:#333
    classDef types fill:#f8f4ff,stroke:#333
    classDef org fill:#f4f4f4,stroke:#333

    class MeshOS system
    class MemoryEngine engine
    class Taxonomy,DataTypes,Subtypes,Relations taxonomy
    class Organization org

Getting Started

Important Note: The JavaScript client can only connect to an existing MeshOS instance. To launch a new MeshOS system, please use the Python version which includes deployment capabilities.

Usage

import { MeshOS } from '@props-labs/mesh-os';
import dotenv from 'dotenv';

// Load environment variables
dotenv.config();

// Connect To MeshOS
const client = new MeshOS({
  url: process.env.HASURA_URL || 'http://localhost:8080',
  apiKey: process.env.HASURA_ADMIN_SECRET || 'meshos',
  openaiApiKey: process.env.OPENAI_API_KEY
});

// Register an agent with a slug
const agent = await client.registerAgent(
  'AI_Explorer',
  'An agent for exploring data',
  { role: 'explorer' },
  'ai-explorer'
);

// Store structured knowledge
const memory = await client.remember(
  'The Moon has water ice.',
  agent.id,
  {
    type: 'knowledge',
    subtype: 'fact',
    tags: ['astronomy', 'moon'],
    version: 1
  }
);

// Retrieve similar knowledge
const results = await client.recall('Tell me about the Moon.', {
  agentId: agent.id,
  limit: 5,
  threshold: 0.7
});

🏗️ Core Features

✅ Memory for Multi-Agent Systems – Let agents store, retrieve, and link structured knowledge.
✅ Fast Semantic Search – pgvector-powered similarity matching across all memories.
✅ Graph-Based Knowledge – Build evolving relationships between facts, ideas, and actions.
✅ Versioning Built-In – Track updates, past decisions, and context shifts.
✅ Event-Driven Execution – Automate workflows based on new knowledge.
✅ Open & Portable – Runs anywhere PostgreSQL does. No black-box infrastructure.

🔗 Structured Taxonomy & Memory Graph

MeshOS enforces structured knowledge with memory classification and versioning:

Memories evolve over time, with full versioning and relationship tracking.

🛠️ Development & Configuration

Configuration

# Required
OPENAI_API_KEY=your_api_key_here

# Optional (defaults shown)
POSTGRES_PASSWORD=mysecretpassword
HASURA_ADMIN_SECRET=meshos
POSTGRES_PORT=5432
HASURA_PORT=8080
HASURA_ENABLE_CONSOLE=true

Development

git clone https://github.com/props-labs/mesh-os.git
cd mesh-os
pnpm install
pnpm build
pnpm test

Contributing

Contributions are welcome! Please submit a Pull Request.

⚖️ License

This project is licensed under the MIT License – see LICENSE for details.