upLLM

Let's put Ultramemory to work. Here's a complete, runnable example that integrates Ultramemory with your Pydantic AI learning agent powered by DeepSeek. This demonstrates the key production features we've discussed: **learning with versioning, time-aware recall, and automatic conflict resolution**.

### 🧪 Step 1: Setup

First, install the required packages. Ultramemory uses LiteLLM under the hood, so it works seamlessly with DeepSeek.

```bash
pip install ultramemory[local] pydantic-ai openai duckduckgo-search
```

### 💻 Step 2: The Complete Code

Save this as `learning_agent_ultramemory.py`. It's a fully functional agent that learns from web searches and maintains a versioned, time-aware memory.

```python
# learning_agent_ultramemory.py
import asyncio
import os
from dataclasses import dataclass, field
from typing import List, Optional
from datetime import datetime

from pydantic_ai import Agent, RunContext
from pydantic_ai.models.openai import OpenAIModel
from pydantic_ai.common_tools.duckduckgo import duckduckgo_search_tool
from pydantic_ai.messages import ModelMessage

# Import Ultramemory
from ultramemory import UltraMemory

# ============================================================================
# 1. Memory Manager Wrapper (Ultramemory)
# ============================================================================
class UltraMemoryManager:
    """A thin wrapper around Ultramemory for our agent's tools."""
    
    def __init__(self, db_path: str = "./agent_memory.db"):
        # Initialize Ultramemory with DeepSeek as the extraction LLM
        # It will use local embeddings (all-MiniLM-L6-v2) for search by default
        self.memory = UltraMemory(
            db_path=db_path,
            model="deepseek-chat",  # Uses DeepSeek for fact extraction & relation detection
            embedding_model="all-MiniLM-L6-v2",  # Free local embeddings
            api_key=os.getenv("DEEPSEEK_API_KEY"),
            api_base="https://api.deepseek.com/v1"
        )