# mem0

Long-term memory layer for AI agents with REST API support.

## Description

Mem0 provides a sophisticated memory management system for AI applications and agents. It enables AI assistants to maintain persistent memory across conversations and sessions using vector storage.

## Features

- 💾 **Long-term Memory**: Persistent memory across conversations
- 🔍 **Semantic Search**: Vector-based similarity search
- 🤖 **Multiple LLM Support**: OpenAI, Anthropic, Groq, Ollama, etc.
- 📊 **Vector Stores**: Qdrant, Chroma, Pinecone, pgvector, etc.
- 🌐 **REST API**: Easy integration via HTTP endpoints
- 🎯 **Multi-modal Support**: Text and image memories
- 🔧 **Configurable**: Flexible LLM and embedding models

## Installation

### Via Overlay

```nix
{pkgs, ...}: {
  environment.systemPackages = with pkgs; [
    mem0
  ];
}
```

### Direct Reference

```nix
{pkgs, ...}: {
  environment.systemPackages = with pkgs; [
    inputs.m3ta-nixpkgs.packages.${pkgs.system}.mem0
  ];
}
```

### Run Directly

```bash
nix run git+https://code.m3ta.dev/m3tam3re/nixpkgs#mem0
```

## Usage

### Command Line

```bash
# Start mem0 server
mem0-server

# With custom port
MEM0_PORT=8080 mem0-server

# With LLM provider
MEM0_LLM_PROVIDER=openai OPENAI_API_KEY=sk-xxx mem0-server
```

### Python Library

```python
from mem0 import Memory

# Initialize with OpenAI
memory = Memory(
    llm_provider="openai",
    llm_model="gpt-4o-mini",
    vector_store="qdrant",
    openai_api_key="sk-xxx"
)

# Add a memory
result = memory.add(
    "I prefer coffee over tea",
    metadata={"user_id": "123"}
)

# Search memories
memories = memory.search("What does the user prefer?")
```

### REST API

```bash
# Start server
mem0-server

# Add memory
curl -X POST http://localhost:8000/v1/memories \
  -H "Content-Type: application/json" \
  -d '{"content": "User likes Python"}'

# Search memories
curl http://localhost:8000/v1/memories/search?q=Python
```

## Configuration

### Environment Variables

```bash
# LLM Configuration
export MEM0_LLM_PROVIDER=openai
export MEM0_LLM_MODEL=gpt-4o-mini
export MEM0_LLM_TEMPERATURE=0.7
export OPENAI_API_KEY=sk-xxx

# Vector Store
export MEM0_VECTOR_PROVIDER=qdrant
export QDRANT_HOST=localhost
export QDRANT_PORT=6333

# Server
export MEM0_HOST=0.0.0.0
export MEM0_PORT=8000
export MEM0_WORKERS=4
export MEM0_LOG_LEVEL=info
```

### NixOS Module (Recommended)

Use the NixOS module for production:

```nix
{config, ...}: {
  imports = [m3ta-nixpkgs.nixosModules.mem0];

  m3ta.mem0 = {
    enable = true;
    port = 8000;
    llm = {
      provider = "openai";
      apiKeyFile = "/run/secrets/openai-api-key";
      model = "gpt-4o-mini";
    };
    vectorStore = {
      provider = "qdrant";
      config = {
        host = "localhost";
        port = 6333;
      };
    };
  };
}
```

See [mem0 Module](../modules/nixos/mem0.md) for full module documentation.

## Supported LLM Providers

| Provider | Model Examples | Notes |
|----------|---------------|-------|
| `openai` | gpt-4, gpt-3.5-turbo | Most tested |
| `anthropic` | claude-3-opus, claude-3-sonnet | Requires key |
| `groq` | mixtral-8x7b-32768 | Fast inference |
| `ollama` | llama2, mistral | Local only |
| `together` | llama-2-70b | API access |

## Supported Vector Stores

| Provider | Requirements | Notes |
|----------|--------------|-------|
| `qdrant` | qdrant server | Recommended |
| `chroma` | chroma server | Simple setup |
| `pgvector` | PostgreSQL + pgvector | SQL-based |
| `pinecone` | Pinecone API | Cloud only |
| `redis` | Redis stack | Fast |
| `elasticsearch` | ES cluster | Scalable |

## Use Cases

### AI Chatbot with Memory

```python
from mem0 import Memory

memory = Memory()

# During conversation
memory.add("User works at Google as a software engineer")

# Later conversations
memories = memory.search("Where does the user work?")
# Returns: ["User works at Google as a software engineer"]
```

### Personal Assistant

```python
memory = Memory()

# Store preferences
memory.add("User prefers dark mode", metadata={"category": "preferences"})
memory.add("User is vegetarian", metadata={"category": "diet"})

# Retrieve relevant info
memories = memory.search("What should I cook for the user?")
```

### Code Assistant

```python
memory = Memory()

# Store project context
memory.add("This is a NixOS project with custom packages")

# Later analysis
memories = memory.search("What kind of project is this?")
```

## Requirements

### Vector Store

You need a vector store running. Example with Qdrant:

```nix
services.qdrant = {
  enable = true;
  port = 6333;
};
```

### LLM Provider

You need API keys or local models:

```nix
# OpenAI
services.mem0.llm.apiKeyFile = "/run/secrets/openai-api-key";

# Or use agenix
age.secrets.openai-api-key.file = ./secrets/openai.age;
```

## Build Information

- **Version**: 1.0.0
- **Language**: Python
- **License**: Apache-2.0
- **Source**: [GitHub](https://github.com/mem0ai/mem0)

## Python Dependencies

- `litellm` - Multi-LLM support
- `qdrant-client` - Qdrant client
- `pydantic` - Data validation
- `openai` - OpenAI client
- `fastapi` - REST API
- `uvicorn` - ASGI server
- `sqlalchemy` - Database ORM

## Platform Support

- Linux (primary)
- macOS (may work)
- Windows (not tested)

## Related

- [mem0 Module](../modules/nixos/mem0.md) - NixOS module documentation
- [Adding Packages](../guides/adding-packages.md) - How to add new packages
- [Port Management](../guides/port-management.md) - Managing service ports
- [Quick Start](../QUICKSTART.md) - Getting started guide