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

# Gateway

> The central orchestrator

## What is Gateway?

Gateway is **the** central orchestrator. Every LLM call goes through it.

```python theme={null}
from splinter import Gateway, ExecutionLimits

gateway = Gateway(
    limits=ExecutionLimits(max_budget=10.0, max_steps=100)
)
gateway.configure_provider("openai", api_key="sk-...")
```

## Configuration

### Limits

```python theme={null}
gateway = Gateway(
    limits=ExecutionLimits(
        max_budget=10.0,      # Stop at $10
        max_steps=100,        # Stop at 100 LLM calls
        max_time_seconds=300, # Stop at 5 minutes
    )
)
```

### Loop Detection

```python theme={null}
from splinter import LoopDetectionConfig

gateway = Gateway(
    loop_detection=LoopDetectionConfig(
        max_repeated_outputs=3,  # Stop if same output 3x
        max_no_state_change=5,   # Stop if state unchanged 5x
    )
)
```

### Multiple Providers

```python theme={null}
gateway.configure_provider("openai", api_key="sk-...")
gateway.configure_provider("anthropic", api_key="sk-ant-...")
gateway.configure_provider("gemini", api_key="...")
```

## Making Calls

### Basic Call

```python theme={null}
response = await gateway.call(
    agent_id="researcher",
    provider="openai",
    model="gpt-4o-mini",
    messages=[
        LLMMessage(role="user", content="Hello")
    ],
)
```

### With Tools

```python theme={null}
response = await gateway.call_with_tools(
    agent_id="researcher",
    provider="openai",
    model="gpt-4o-mini",
    messages=[...],
    tools=[...],
    tool_executor=my_tool_executor,
)
```

## Observability

### Get Metrics

```python theme={null}
metrics = gateway.get_metrics()
# {
#     "total_cost": 0.05,
#     "total_steps": 10,
#     "total_tokens": 5000,
#     "elapsed_seconds": 30.5,
# }
```

### Get Call History

```python theme={null}
history = gateway.get_call_history(agent_id="researcher", limit=10)
for record in history:
    print(f"{record.agent_id}: ${record.cost:.4f}")
```

## Hooks

```python theme={null}
def log_before(agent_id, request):
    print(f"[{agent_id}] Calling LLM...")

def log_after(agent_id, response):
    print(f"[{agent_id}] Got response, cost: ${response.cost:.4f}")

gateway = Gateway(
    on_before_call=log_before,
    on_after_call=log_after,
)
```

## Internal Structure

```python theme={null}
class Gateway:
    def __init__(self):
        self._control = _ControlLayer(...)      # Limits, loops, tools
        self._providers = {}                     # OpenAI, Anthropic, etc.
        self._coordination = _CoordinationLayer(...) # Checkpoints
```

Gateway orchestrates these layers. You don't interact with them directly.