Fireworks.ai offers a powerful Responses API that allows for more complex and stateful interactions with models. This guide will walk you through the key features and how to use them.

​
Overview

The Responses API is designed for building conversational applications and complex workflows. It allows you to:

  • Continue conversations: Maintain context across multiple turns without resending the entire history.
  • Use external tools: Integrate with external services and data sources through the Model Context Protocol (MCP).
  • Stream responses: Receive results as they are generated, enabling real-time applications.

​
Basic Usage

You can interact with the Response API using the Fireworks Python SDK or by making direct HTTP requests.

​
Creating a Response

To start a new conversation, you use the client.responses.create method. For a complete example, see the getting started notebook.

from fireworks import LLM

llm = LLM(model="qwen3-235b-a22b", deployment_type="serverless")

response = llm.responses.create(
    input="What is reward-kit and what are its 2 main features? Keep it short Please analyze the fw-ai-external/reward-kit repository.",
    tools=[{"type": "sse", "server_url": "https://gitmcp.io/docs"}]
)

print(response.output[-1].content[0].text.split("</think>")[-1])

​
Continuing a Conversation with previous_response_id

To continue a conversation, you can use the previous_response_id parameter. This tells the API to use the context from a previous response, so you don’t have to send the entire conversation history again. For a complete example, see the previous response ID notebook.

from fireworks import LLM

llm = LLM(model="qwen3-235b-a22b", deployment_type="serverless")

# First, create an initial response
initial_response = llm.responses.create(
    input="What are the key features of reward-kit?",
    tools=[{"type": "sse", "server_url": "https://gitmcp.io/docs"}]
)
initial_response_id = initial_response.id

# Now, continue the conversation
continuation_response = llm.responses.create(
    input="How do I install it?",
    previous_response_id=initial_response_id,
    tools=[{"type": "sse", "server_url": "https://gitmcp.io/docs"}]
)

print(continuation_response.output[-1].content[0].text.split("</think>")[-1])

​
Streaming Responses

For real-time applications, you can stream the response as it’s being generated. For a complete example, see the streaming example notebook.

from fireworks import LLM

llm = LLM(model="qwen3-235b-a22b", deployment_type="serverless")

stream = llm.responses.create(
    input="give me 5 interesting facts on modelcontextprotocol/python-sdk -- keep it short!",
    stream=True,
    tools=[{"type": "mcp", "server_url": "https://mcp.deepwiki.com/mcp"}]
)

for chunk in stream:
    print(chunk)

​
Cookbook Examples

For more in-depth examples, check out the following notebooks:

​
Storing Responses

By default, responses are stored and can be referenced by their ID. You can disable this by setting store=False. If you do this, you will not be able to use the previous_response_id to continue the conversation. For a complete example, see the store=False notebook.

from fireworks import LLM

llm = LLM(model="qwen3-235b-a22b", deployment_type="serverless")

response = llm.responses.create(
    input="give me 5 interesting facts on modelcontextprotocol/python-sdk -- keep it short!",
    store=False,
    tools=[{"type": "mcp", "server_url": "https://mcp.deepwiki.com/mcp"}]
)

# This will fail because the previous response was not stored
try:
    continuation_response = llm.responses.create(
        input="Explain the second fact in more detail.",
        previous_response_id=response.id
    )
except Exception as e:
    print(e)