Mem0 Integration

Mem0 provides semantic memory capabilities — auto-learning from conversations and retrieving relevant memories via vector search.

How It Works

1. Auto-learn: After each agent response, the conversation is sent to Mem0 in a background task
2. Memory extraction: Mem0’s LLM extracts key facts and preferences
3. Embedding: Facts are embedded and stored in a vector database
4. Retrieval: When building context for new messages, relevant memories are retrieved via semantic search

Setup

# Install PocketPaw
curl -fsSL https://pocketpaw.xyz/install.sh | sh

# Or add the memory extra manually
pip install pocketpaw[memory]

# Configure Mem0
export POCKETCLAW_MEM0_AUTO_LEARN=true
export POCKETCLAW_MEM0_LLM_PROVIDER="ollama"
export POCKETCLAW_MEM0_LLM_MODEL="llama3.2"
export POCKETCLAW_MEM0_EMBEDDER_PROVIDER="ollama"
export POCKETCLAW_MEM0_EMBEDDER_MODEL="nomic-embed-text"
export POCKETCLAW_MEM0_VECTOR_STORE="qdrant"

Provider Configuration

LLM Provider

The LLM is used by Mem0 to extract facts from conversations:

Embedder Provider

The embedder converts text to vectors for semantic search:

Vector Store

Embedding Dimensions

Embedding dimensions must match between the model and the vector store collection. PocketPaw includes a built-in dimensions lookup and auto-detection for Ollama models:

_EMBEDDING_DIMS = {
    "nomic-embed-text": 768,
    "mxbai-embed-large": 1024,
    "text-embedding-3-small": 1536,
    "text-embedding-3-large": 3072,
}

For unknown Ollama models, PocketPaw queries the model to auto-detect dimensions.

Dashboard Configuration

The web dashboard provides a Memory settings panel where you can:

• Enable/disable auto-learning
• Select LLM and embedder providers
• Choose models
• Test memory retrieval

Fully Local Setup

For a completely local memory system with no external API calls:

# Install Ollama
curl -fsSL https://ollama.com/install.sh | sh

# Pull models
ollama pull llama3.2
ollama pull nomic-embed-text

# Configure
export POCKETCLAW_MEM0_AUTO_LEARN=true
export POCKETCLAW_MEM0_LLM_PROVIDER="ollama"
export POCKETCLAW_MEM0_LLM_MODEL="llama3.2"
export POCKETCLAW_MEM0_EMBEDDER_PROVIDER="ollama"
export POCKETCLAW_MEM0_EMBEDDER_MODEL="nomic-embed-text"
export POCKETCLAW_MEM0_VECTOR_STORE="qdrant"

This runs everything locally — LLM inference, embedding, and vector storage.