Skip to content

Questions: s08 tool result budget #605

Open

Description

@shenhai-ran

First of all, great work! I’ve learned a lot from it.

I have a question regarding tool_result_budget. My understanding is that it checks the tool results from the previous run, considering both the total size and the size of each result, and then compacts them before the subsequent LLM calls.

Does this mean that large tool results from the previous run are never shown to the LLM in full, and that the LLM only sees a preview?

If so, could this lead to the following issues?

  1. The LLM might miss important information that appears later in the result, for example, near the end of a very large log file.
  2. In some extreme cases, the LLM might try to retrieve the full content by making another tool call in the current run, only for that result to be compacted again in the next run. Could this potentially create a loop?

Thanks for your help and explanation!

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions