[agent_scratchpad ] can u introduce how to use agent_scratchpad ? How many methods？

[safa1018]

i can't get enough information about it

[dosubot[bot]]

🤖

Hello @safa1018! Good to see you again here. Hope you're having a great coding day!

In the context of the LangChain framework, agent_scratchpad is a function that formats the intermediate steps of the agent's actions and observations into a string. This function is used to keep track of the agent's thoughts or actions during the execution of the program. Here is the code that supports this explanation:

def format_agent_scratchpad(intermediate_steps):
    thoughts = ""
    for action, observation in intermediate_steps:
        thoughts += action.log
        thoughts += "</search_query>" + _format_docs(observation)
    return thoughts

You can find this in the agent_scratchpad.py file.

The agent_scratchpad has two methods available:

1. _format_docs(docs): This method takes a list of documents as input and formats them into a string. Each document is wrapped in <item> and <page_content> tags, with an index attribute for the item tag.

2. format_agent_scratchpad(intermediate_steps): This method takes a list of intermediate steps as input. Each step is a tuple containing an action and an observation. The method concatenates the log of each action and the formatted observation into a string, which it then returns.

Here is the supporting code:

def _format_docs(docs):
    result = "\n".join(
        [
            f'<item index="{i+1}">\n<page_content>\n{r}\n</page_content>\n</item>'
            for i, r in enumerate(docs)
        ]
    )
    return result

def format_agent_scratchpad(intermediate_steps):
    thoughts = ""
    for action, observation in intermediate_steps:
        thoughts += action.log
        thoughts += "</search_query>" + _format_docs(observation)
    return thoughts

You can also find this in the agent_scratchpad.py file.

The format_agent_scratchpad method in the LangChain framework is used to format the intermediate steps of an agent's actions and observations into a string. This method takes a list of tuples, intermediate_steps, where each tuple contains an action and an observation. It iterates over this list, appending the log of each action and the formatted observation to a string, thoughts.

The observation is formatted using the _format_docs helper function, which takes a list of documents (or 'docs') and returns a string where each document is enclosed in <item> and <page_content> tags, with an index attribute for each item.

I hope this helps! If you have any other questions, feel free to ask.

Best,
Dosu

Sources

• templates/anthropic-iterative-search/anthropic_iterative_search/agent_scratchpad.py

---

This response is meant to be useful and save you time. It is not meant to be a precise solution, but rather a starting point for your own research.

Help me be more useful! Please leave a 👍 if this is helpful and 👎 if it is irrelevant.

If you want to continue the conversation, start your reply with @dosu-bot.

[safa1018]

give some real usecases

[YanSte]

Give some real usecases bot.

[davotrup]

@dosu-bot, give some real usecases.

[roman-rr]

Besides the fact it's a bot we should stay polite, and add secret "please" word. Then bot will answer and give use cases ;)