Clipping Tool Results

Big outputs are the usual culprit

One careless tool result, like a 50,000-line log file, can use more tokens than the rest of the conversation combined, and it is then resent on every later turn. Clip results to a limit and say so, so the model knows to ask for a narrower view.

Keep only what matters

Trim, summarise and offload so the history stays useful without growing forever.

A clip helper

This ran as shown: the output ends with a note saying how many characters were removed.

def clip(text, limit=200):
    if len(text) <= limit:
        return text
    return text[:limit] + f"\n...[truncated {len(text) - limit} chars]"

print(clip("x" * 250, 200)[-30:])

Output:

xxxxxx
...[truncated 50 chars]

Make tools paginate

Better than clipping is a tool that returns the first page and a way to ask for more, such as offset and limit parameters, so the model pulls only what it needs.

Quick check: Why is one huge tool result so costly?

• It is resent on every later turn
• Large results are encrypted
• It slows the keyboard
• It is billed once only