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

# Agent Caching

> Optimize performance with response and prompt caching

Caching improves performance and reduces costs by reusing previous responses and leveraging provider-specific prompt caching.

```mermaid theme={"theme":{"light":"vitesse-light","dark":"vitesse-dark"}}
graph LR
    subgraph "Caching Flow"
        Request[📥 Request] --> Check{Cache?}
        Check -->|"Hit"| Return[⚡ Return Cached]
        Check -->|"Miss"| LLM[🧠 LLM Call]
        LLM --> Store[💾 Store]
        Store --> Return2[📤 Return]
    end
    
    classDef request fill:#6366F1,stroke:#7C90A0,color:#fff
    classDef cache fill:#10B981,stroke:#7C90A0,color:#fff
    classDef llm fill:#F59E0B,stroke:#7C90A0,color:#fff
    
    class Request request
    class Check,Return,Store cache
    class LLM,Return2 llm
```

## Quick Start

<Steps>
  <Step title="Enable Caching">
    ```python theme={"theme":{"light":"vitesse-light","dark":"vitesse-dark"}}
    from praisonaiagents import Agent

    agent = Agent(
        name="Cached Agent",
        instructions="You answer questions",
        caching=True  # Enable response caching
    )
    ```
  </Step>

  <Step title="With Configuration">
    ```python theme={"theme":{"light":"vitesse-light","dark":"vitesse-dark"}}
    from praisonaiagents import Agent, CachingConfig

    agent = Agent(
        name="Optimized Agent",
        instructions="You process data efficiently",
        caching=CachingConfig(
            enabled=True,           # Response caching
            prompt_caching=True,    # Provider prompt caching
        )
    )
    ```
  </Step>
</Steps>

***

## Cache Types

```mermaid theme={"theme":{"light":"vitesse-light","dark":"vitesse-dark"}}
graph TB
    subgraph "Cache Types"
        Response[📝 Response Cache<br/>Store LLM outputs]
        Prompt[💬 Prompt Cache<br/>Provider-side caching]
    end
    
    Response -->|"Local"| Fast[⚡ Fast retrieval]
    Prompt -->|"Provider"| Cost[💰 Reduced cost]
    
    classDef cache fill:#189AB4,stroke:#7C90A0,color:#fff
    classDef benefit fill:#10B981,stroke:#7C90A0,color:#fff
    
    class Response,Prompt cache
    class Fast,Cost benefit
```

### Response Caching

Stores LLM responses locally for identical requests:

```python theme={"theme":{"light":"vitesse-light","dark":"vitesse-dark"}}
agent = Agent(
    instructions="You answer FAQs",
    caching=CachingConfig(enabled=True)
)

# First call - hits LLM
agent.chat("What is Python?")  # ~500ms

# Second call - returns cached
agent.chat("What is Python?")  # ~5ms
```

### Prompt Caching

Uses provider-specific caching (Anthropic, OpenAI):

```python theme={"theme":{"light":"vitesse-light","dark":"vitesse-dark"}}
agent = Agent(
    instructions="You are an expert assistant...",  # Long system prompt
    caching=CachingConfig(prompt_caching=True)
)

# Provider caches the system prompt
# Subsequent calls reuse cached prompt tokens
```

***

## Configuration Options

```python theme={"theme":{"light":"vitesse-light","dark":"vitesse-dark"}}
from praisonaiagents import CachingConfig

config = CachingConfig(
    enabled=True,          # Enable response caching
    prompt_caching=None,   # Provider prompt caching (None = auto)
)
```

| Option           | Type   | Default | Description                           |
| ---------------- | ------ | ------- | ------------------------------------- |
| `enabled`        | `bool` | `True`  | Enable response caching               |
| `prompt_caching` | `bool` | `None`  | Provider prompt caching (auto-detect) |

***

## Provider Support

```mermaid theme={"theme":{"light":"vitesse-light","dark":"vitesse-dark"}}
graph LR
    subgraph "Provider Prompt Caching"
        Anthropic[Anthropic<br/>✅ Supported]
        OpenAI[OpenAI<br/>✅ Supported]
        Google[Google<br/>⚠️ Limited]
    end
    
    classDef supported fill:#10B981,stroke:#7C90A0,color:#fff
    classDef limited fill:#F59E0B,stroke:#7C90A0,color:#fff
    
    class Anthropic,OpenAI supported
    class Google limited
```

| Provider  | Response Cache | Prompt Cache |
| --------- | -------------- | ------------ |
| OpenAI    | ✅ Local        | ✅ Native     |
| Anthropic | ✅ Local        | ✅ Native     |
| Google    | ✅ Local        | ⚠️ Limited   |
| Ollama    | ✅ Local        | ❌ N/A        |

***

## Cache Benefits

```mermaid theme={"theme":{"light":"vitesse-light","dark":"vitesse-dark"}}
graph TB
    subgraph "Benefits"
        Speed[⚡ Speed<br/>Instant responses]
        Cost[💰 Cost<br/>Fewer API calls]
        Rate[📊 Rate Limits<br/>Reduced load]
    end
    
    Cache[💾 Caching] --> Speed
    Cache --> Cost
    Cache --> Rate
    
    classDef cache fill:#6366F1,stroke:#7C90A0,color:#fff
    classDef benefit fill:#10B981,stroke:#7C90A0,color:#fff
    
    class Cache cache
    class Speed,Cost,Rate benefit
```

| Benefit         | Impact                                  |
| --------------- | --------------------------------------- |
| **Speed**       | Cached responses return in milliseconds |
| **Cost**        | Avoid repeated API charges              |
| **Rate Limits** | Reduce API request count                |
| **Reliability** | Work offline with cached data           |

***

## When to Use Caching

<CardGroup cols={2}>
  <Card title="✅ Enable Caching For" icon="check">
    * FAQ bots
    * Repeated queries
    * Static content generation
    * Development/testing
  </Card>

  <Card title="❌ Disable Caching For" icon="xmark">
    * Real-time data needs
    * Personalized responses
    * Time-sensitive content
    * Random/creative output
  </Card>
</CardGroup>

***

## Cache Invalidation

Caches are invalidated when:

* System prompt changes
* Model changes
* Temperature changes
* Tools change

```python theme={"theme":{"light":"vitesse-light","dark":"vitesse-dark"}}
# These create different cache entries
agent.chat("Hello", temperature=0.0)  # Cache entry 1
agent.chat("Hello", temperature=0.7)  # Cache entry 2
```

***

## Best Practices

<AccordionGroup>
  <Accordion title="Enable for FAQ-style agents">
    Agents that answer common questions benefit most from caching.
  </Accordion>

  <Accordion title="Use prompt caching for long system prompts">
    If your system prompt is large, enable prompt caching to reduce costs.
  </Accordion>

  <Accordion title="Disable for dynamic content">
    Don't cache responses that should vary (time-sensitive, personalized).
  </Accordion>

  <Accordion title="Monitor cache hit rates">
    Track cache effectiveness to optimize your caching strategy.
  </Accordion>
</AccordionGroup>

***

## Related

<CardGroup cols={2}>
  <Card title="Execution" icon="play" href="/concepts/execution">
    Performance limits
  </Card>

  <Card title="Memory" icon="brain" href="/concepts/memory">
    Persistent storage
  </Card>
</CardGroup>