> ## Documentation Index
> Fetch the complete documentation index at: https://docs.statebase.org/llms.txt
> Use this file to discover all available pages before exploring further.

# OpenAI Integration

> Optimized prompt engineering with StateBase

## The "Amalthea" Pattern

When using StateBase with OpenAI, we recommend a specific prompting pattern to maximize reliability: **Context Injection**.

Instead of sending the raw chat history array to OpenAI, you let StateBase curate the context.

### Code Example

```python theme={null}
from openai import OpenAI
from statebase import StateBase

client = OpenAI()
sb = StateBase()

# 1. Fetch Curated Context
# StateBase automatically ranks recent turns + relevant memories
context_package = sb.sessions.get_context(
    session_id="sess_123",
    query="How does the user like their coffee?"
)

# 2. Construct Prompt
messages = [
    {
        "role": "system", 
        "content": f"""
        You are a helpful assistant.
        
        # PREVIOUS KNOWLEDGE
        {context_package.memories_str}
        
        # CURRENT STATE
        {context_package.state_str}
        
        # RECENT HISTORY
        {context_package.recent_history_str}
        """
    },
    {"role": "user", "content": "Order me a coffee."}
]

# 3. Generate
completion = client.chat.completions.create(
    model="gpt-4-turbo",
    messages=messages
)
```

## Why this works better?

1. **Token Efficiency**: You don't send 50 turns of history. You send a summarized state + 5 relevant turns.
2. **Accuracy**: "Memories" (long-term facts) are injected explicitly, preventing the model from having to "search" its context window.
3. **Cost**: Significantly reduces input token costs for long-running sessions.