ReActChatFormatter class. This class formats the chat history into a list of ChatMessage objects. Here's a basic example of how you can use it:from llama_index.core.agent import ReActChatFormatter
from llama_index.core.llms import ChatMessage
# Initialize the chat formatter
chat_formatter = ReActChatFormatter()
# Define your tools and chat history
tools = [...] # Your tools go here
chat_history = [
ChatMessage(
content="Your message content here",
role="user",
)
]
# Format the chat history
input_msgs = chat_formatter.format(tools, chat_history)input_msgs is the list of ChatMessage objects that you can pass to the LLM. Each ChatMessage object represents a message in the conversation and has a content attribute (the message content) and a role attribute (the role of the message sender, which can be "user" or "assistant").PromptTemplate class and the agent.update_prompts() method:from llama_index.core import PromptTemplate
# Define your system prompt
react_system_prompt = PromptTemplate("Your system prompt here")
# Update the system prompt
agent.update_prompts({"agent_worker:system_prompt": react_system_prompt})"Your system prompt here" should be replaced with your actual system prompt.[...] and "Your message content here" with your actual tools and message content.agent.get_prompts() method, and you can update the prompts using the agent.update_prompts() method.# Get the current prompts
prompt_dict = agent.get_prompts()
for k, v in prompt_dict.items():
print(f"Prompt: {k}\n\nValue: {v.template}")
# Update the prompts
react_system_prompt = PromptTemplate("Your system prompt here")
agent.update_prompts({"agent_worker:system_prompt": react_system_prompt})"Your system prompt here" should be replaced with your actual system prompt.agent.chat() method.agent.chat() method ()ChatMessage objects and passing that list to the ReActChatFormatter's format() method. Each ChatMessage object represents a message in the conversation.from llama_index.core.agent import ReActChatFormatter
from llama_index.core.llms import ChatMessage
# Initialize the chat formatter
chat_formatter = ReActChatFormatter()
# Define your tools and chat history
tools = [...] # Your tools go here
chat_history = [
ChatMessage(
content="Your first message content here",
role="user",
),
ChatMessage(
content="Your second message content here",
role="assistant",
),
# Add more messages as needed
]
# Format the chat history
input_msgs = chat_formatter.format(tools, chat_history)input_msgs is the list of ChatMessage objects that you can pass to the LLM. Each ChatMessage object has a content attribute (the message content) and a role attribute (the role of the message sender, which can be "user" or "assistant").[...] and "Your first message content here" and "Your second message content here" with your actual tools and message content.ReActOutputParser class. This class parses the output of the ReAct agent into a structured format.from llama_index.core.agent.react.output_parser import ReActOutputParser
from llama_index.core.llms import ChatResponse
from llama_index.core.agent.types import Task
def parse_react_output_fn(
task: Task, state: Dict[str, Any], chat_response: ChatResponse
):
"""Parse ReAct output into a reasoning step."""
output_parser = ReActOutputParser()
reasoning_step = output_parser.parse(chat_response.message.content)
return {"done": reasoning_step.is_done, "reasoning_step": reasoning_step}
parse_react_output = AgentFnComponent(fn=parse_react_output_fn)parse_react_output_fn is a function that takes a task, a state, and a chat response, and returns a dictionary with two keys: "done" (a boolean indicating whether the reasoning step is done) and "reasoning_step" (the parsed reasoning step).Task, Dict[str, Any], and ChatResponse with your actual task, state, and chat response.PydanticOutputParser with a ReAct agent, you would need to initialize the PydanticOutputParser with your desired output class and then use it to parse the output from the ReAct agent. However, the provided knowledge sources do not provide a direct example of using PydanticOutputParser with a ReAct agent.PydanticOutputParser:from llama_index.core.output_parsers import PydanticOutputParser
# Define your output class
class MyOutputClass(BaseModel):
# Define your fields here
...
# Initialize the output parser
output_parser = PydanticOutputParser(output_cls=MyOutputClass)
# Use the output parser to parse the output
parsed_output = output_parser.parse(output_text)MyOutputClass should be replaced with your actual output class, and output_text should be replaced with the text output from the ReAct agent.