Agent Executor with Structure Output

[daaffalbari]

can i use Agent Executor with JSONOutputParser()?

For example:

tools = [
    tool_search_youtube,
    tool_get_youtube_description,
    tool_get_transcript,
    tool_summarize_video_transcript,
    tool_generate_question,
]

prompt = ChatPromptTemplate.from_messages(
    [
        ("system", main_prompt),
        ("human", "{input}"),
        # MessagesPlaceholder('chat_history', optional=True),
        MessagesPlaceholder("agent_scratchpad"),
    ]
)

from langchain.agents import AgentExecutor, create_openai_tools_agent

agent = create_openai_tools_agent(llm, tools=tools, prompt=prompt)
agent_executor = AgentExecutor(agent=agent, tools=tools)

[dosubot[bot]]

I found two similar unsolved discussions that might be relevant to your question:

1. create_tool_calling_agent only output tool result in JSON instead of a straightforward answer ^[1].
2. Does AgentExecutor support a JSON dict for tools? ^[2].

Regarding your question, you can use AgentExecutor with JSONOutputParser in LangChain. The JSONOutputParser is designed to parse tool invocations and final answers in JSON format, which can be integrated with AgentExecutor to handle structured outputs. The parse method of JSONOutputParser interprets the output to determine whether it should result in an AgentAction or an AgentFinish ^[3].

To continue talking to Dosu, mention @dosu.

[daaffalbari]

@dosu can you give me example code to give json output using agent?

[dosubot[bot]]

To generate JSON output using an agent in LangChain, you can use the create_json_chat_agent function. Here's an example code snippet demonstrating how to set up and use this function:

from langchain import hub
from langchain_community.chat_models import ChatOpenAI
from langchain.agents import AgentExecutor, create_json_chat_agent
from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder

# Define the system and human prompts
system = '''Assistant is a large language model trained by OpenAI.'''
human = '''TOOLS
------
Assistant can ask the user to use tools to look up information that may be helpful in answering the users original question. The tools the human can use are:

{tools}

RESPONSE FORMAT INSTRUCTIONS
----------------------------

When responding to me, please output a response in one of two formats:

**Option 1:**
Use this if you want the human to use a tool.
Markdown code snippet formatted in the following schema:

```json
{{
    "action": string, \\ The action to take. Must be one of {tool_names}
    "action_input": string \\ The input to the action
}}

Option #2:
Use this if you want to respond directly to the human. Markdown code snippet formatted in the following schema:

{{
    "action": "Final Answer",
    "action_input": string \\ You should put what you want to return to use here
}}

USER'S INPUT

Here is the user's input (remember to respond with a markdown code snippet of a json blob with a single action, and NOTHING else):

{input}'''

Create the prompt template

prompt = ChatPromptTemplate.from_messages(
[
("system", system),
MessagesPlaceholder("chat_history", optional=True),
("human", human),
MessagesPlaceholder("agent_scratchpad"),
]
)

Initialize the language model and tools

model = ChatOpenAI()
tools = [...] # Define your tools here

Create the JSON chat agent

agent = create_json_chat_agent(model, tools, prompt)

Create an agent executor

agent_executor = AgentExecutor(agent=agent, tools=tools)

Invoke the agent with input

response = agent_executor.invoke({"input": "hi"})
print(response)

Using with chat history

response_with_history = agent_executor.invoke(
{
"input": "what's my name?",
"chat_history": [
HumanMessage(content="hi! my name is bob"),
AIMessage(content="Hello Bob! How can I assist you today?"),
],
}
)
print(response_with_history)

This code sets up a JSON chat agent using LangChain, which can generate JSON-formatted responses based on the input and the defined tools <sup>[[1]](https://github.com/langchain-ai/langchain/blob/master/libs/langchain/langchain/agents/json_chat/base.py)</sup>.


<!-- Dosu Comment Footer -->
*To continue talking to [Dosu](https://dosu.dev), mention @dosu.*