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

# Haystack SDK

> Use Haystack pipelines with Keywords AI gateway for automatic logging, fallbacks, and cost optimization

<Note> This integration is for the **Keywords AI gateway**. For workflow tracing, see [Haystack Tracing](/integration/development-frameworks/tracing/haystack). </Note>

## Overview

Haystack is an open-source framework for building LLM applications with composable pipelines. The Keywords AI gateway integration routes your LLM calls through Keywords AI for automatic logging, fallbacks, load balancing, and cost optimization.

<Frame className="rounded-md">
  <img src="https://keywordsai-static.s3.us-east-1.amazonaws.com/docs/Integrations/haystack/haystack_gateway.png" alt="Haystack gateway integration" />
</Frame>

## Installation

```bash theme={"system"}
pip install keywordsai-exporter-haystack
```

## Quickstart

### Step 1: Set Environment Variables

```bash theme={"system"}
export KEYWORDSAI_API_KEY="your-keywords-ai-key"
```

### Step 2: Replace OpenAIGenerator with KeywordsAIGenerator

<CodeGroup>
  ```python Basic Usage theme={"system"}
  import os
  from haystack import Pipeline
  from haystack.components.builders import PromptBuilder
  from keywordsai_exporter_haystack import KeywordsAIGenerator

  # Create pipeline
  pipeline = Pipeline()
  pipeline.add_component("prompt", PromptBuilder(template="Tell me about {{topic}}."))
  pipeline.add_component("llm", KeywordsAIGenerator(
      model="gpt-4o-mini",
      api_key=os.getenv("KEYWORDSAI_API_KEY")
  ))
  pipeline.connect("prompt", "llm")

  # Run
  result = pipeline.run({"prompt": {"topic": "machine learning"}})
  print(result["llm"]["replies"][0])
  ```

  ```python With Metadata theme={"system"}
  from keywordsai_exporter_haystack import KeywordsAIGenerator

  pipeline.add_component("llm", KeywordsAIGenerator(
      model="gpt-4o-mini",
      api_key=os.getenv("KEYWORDSAI_API_KEY"),
      generation_kwargs={
          "customer_identifier": "user_123",
          "metadata": {"session_id": "abc123"}
      }
  ))
  ```
</CodeGroup>

That's it! All LLM calls are now automatically logged to Keywords AI.

## Prompt Management

Use platform-managed prompts for centralized control:

```python theme={"system"}
import os
from haystack import Pipeline
from keywordsai_exporter_haystack import KeywordsAIGenerator

# Create pipeline with platform prompt
# No model needed - it comes from the platform
pipeline = Pipeline()
pipeline.add_component("llm", KeywordsAIGenerator(
    prompt_id="your-prompt-id",  # Get from platform
    api_key=os.getenv("KEYWORDSAI_API_KEY")
))

# Run with prompt variables
result = pipeline.run({
    "llm": {
        "prompt_variables": {
            "user_input": "your text here"
        }
    }
})
```

**Benefits:**

* Update prompts without code changes
* Model configuration managed on platform
* Version control & rollback
* A/B testing

<Note>
  Create prompts at: [platform.keywordsai.co/platform/prompts](https://platform.keywordsai.co/platform/prompts)
</Note>

## Supported Parameters

### OpenAI Parameters

All [OpenAI parameters](/api-endpoints/develop/gateway/chat-completions#openai-compatible-parameters) are supported:

```python theme={"system"}
pipeline.add_component("llm", KeywordsAIGenerator(
    model="gpt-4o-mini",
    api_key=os.getenv("KEYWORDSAI_API_KEY"),
    generation_kwargs={
        "temperature": 0.7,      # Control randomness
        "max_tokens": 1000,      # Limit response length
    }
))
```

### Keywords AI Parameters

Use [Keywords AI parameters](/api-endpoints/develop/gateway/chat-completions#keywords-ai-parameters) for advanced features:

```python theme={"system"}
pipeline.add_component("llm", KeywordsAIGenerator(
    model="gpt-4o-mini",
    api_key=os.getenv("KEYWORDSAI_API_KEY"),
    generation_kwargs={
        "customer_identifier": "user_123",           # Track users
        "fallback_models": ["gpt-3.5-turbo"],       # Auto fallbacks
        "metadata": {"session_id": "abc123"},        # Custom metadata
        "thread_identifier": "conversation_456",     # Group messages
        "group_identifier": "team_alpha",           # Organize by groups
    }
))
```

## Workflow Tracing

For complete visibility of your pipeline execution, add workflow tracing to see how data flows through each component.

<Card title="Haystack Tracing Integration" href="/integration/development-frameworks/tracing/haystack" icon="link">
  Learn how to trace your entire Haystack pipeline
</Card>