VLLMChatGenerator

Generate chat responses using models served with vLLM.

Key Features

• Works with any model served by a vLLM server using the OpenAI-compatible API.
• Accepts and returns messages in ChatMessage format.
• Supports streaming responses through a configurable callback.
• Supports tool calling for agentic workflows.
• Supports reasoning models with the extra_body parameter.
• Uses the OpenAI-compatible API, so the same component works with any vLLM-served model.

Configuration

1. Drag the VLLMChatGenerator component onto the canvas from the Component Library.
2. Click on the component to open the configuration panel.
3. On the General tab:
   1. Set the model to the model name served by your vLLM instance (for example, Qwen/Qwen3-4B-Instruct).
   2. Set the api_base_url to your vLLM server address. The default is http://localhost:8000/v1.
4. Go to the Advanced tab to configure generation_kwargs, timeout, max_retries, and tools.

Connections

VLLMChatGenerator receives a list of ChatMessage objects, typically from PromptBuilder or ChatPromptBuilder. It outputs a list of reply ChatMessage objects you can connect to AnswerBuilder or other downstream components.

Source Code

To check this component's source code, open chat_generator.py in the Haystack Core Integrations repository.

Usage Examples

Basic Configuration

  VLLMChatGenerator:
    type: haystack_integrations.components.generators.vllm.VLLMChatGenerator
    init_parameters:
      model: Qwen/Qwen3-4B-Instruct
      api_base_url: http://localhost:8000/v1
      generation_kwargs:
        temperature: 0.7
        max_tokens: 1024

Using the Component in a Pipeline

# haystack-pipeline
components:
  prompt_builder:
    type: haystack.components.builders.chat_prompt_builder.ChatPromptBuilder
    init_parameters:
      required_variables: "*"
      template:
      - role: user
        content: "Answer the following question: {{ question }}"

  llm:
    type: haystack_integrations.components.generators.vllm.VLLMChatGenerator
    init_parameters:
      model: Qwen/Qwen3-4B-Instruct
      api_base_url: http://localhost:8000/v1
      generation_kwargs:
        max_tokens: 1024

  answer_builder:
    type: deepset_cloud_custom_nodes.augmenters.deepset_answer_builder.DeepsetAnswerBuilder
    init_parameters:
      reference_pattern: acm

connections:
- sender: prompt_builder.prompt
  receiver: llm.messages
- sender: llm.replies
  receiver: answer_builder.replies

max_runs_per_component: 100

metadata: {}

inputs:
  query:
  - answer_builder.query
  - prompt_builder.question

outputs:
  answers: answer_builder.answers

Parameters

Inputs

Parameter	Type	Description
messages	List[ChatMessage]	A list of chat messages representing the conversation so far.

Outputs

Parameter	Type	Description
replies	List[ChatMessage]	A list of generated reply messages from the model.

Init Parameters

These are the parameters you can configure in Pipeline Builder:

Parameter	Type	Default	Description
model	str	(required)	The name of the model served by the vLLM instance.
api_key	Optional[Secret]	Secret.from_env_var("VLLM_API_KEY", strict=False)	An API key for authenticated vLLM deployments.
streaming_callback	Optional[Callable]	None	A callback function for streaming responses.
api_base_url	str	http://localhost:8000/v1	The URL of the vLLM server's OpenAI-compatible API.
generation_kwargs	Optional[Dict[str, Any]]	None	Additional generation parameters such as temperature, max_tokens, and top_p.
timeout	Optional[float]	None	Request timeout in seconds.
max_retries	Optional[int]	None	Maximum number of retries on API errors.
tools	Optional[List[Tool]]	None	A list of tools the model can use for tool calling.
http_client_kwargs	Optional[Dict[str, Any]]	None	Additional keyword arguments for the HTTP client.

Run Method Parameters

These are the parameters you can configure for the component's run() method. This means you can pass these parameters at query time through the API, in Playground, or when running a job. For details, see Modify Pipeline Parameters at Query Time.

Parameter	Type	Default	Description
messages	List[ChatMessage]		A list of chat messages representing the conversation.
streaming_callback	Optional[Callable]	None	A callback function to override the init-time streaming callback.
generation_kwargs	Optional[Dict[str, Any]]	None	Generation parameters to override init-time values.
tools	Optional[List[Tool]]	None	Tools to make available to the model.

Related Information

• Pipeline Components