Tool calls with empty results can make the assistant stay silent

[bnovik0v]

Problem

When a tool finishes with an empty-looking result like None or {}, Pipecat may treat that as if there was no real result.

That causes two problems:

• the LLM may not run again, so the assistant stays silent
• the tool result stored in context can become "COMPLETED" instead of the real value

Example

A weather tool times out and returns None.

Current behavior:

• the tool call is marked finished
• the assistant is not triggered to speak again
• the user hears nothing

Expected behavior:

• Pipecat should still treat this as a finished tool call
• the LLM should be able to say something like Sorry, that tool timed out

Another example:
A tool returns {} to mean no data found.

Current behavior:

• Pipecat stores "COMPLETED" in context

Expected behavior:

• Pipecat should preserve {} in context so the model can reason about the real result

Scope

This issue is about final tool results that are falsy:

• None
• ""
• 0
• {}
• []