Add support for function call start_callback #101
Conversation
|
The latest updates on your projects. Learn more about Vercel for Git ↗︎
|
markbackman
force-pushed
the
mb/start-callback
branch
from
6e6657b to
13ef0c6
Compare
| @@ -7,6 +7,37 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0 | |||
|
|
|||
| ## [Unreleased] | |||
|
|
|||
| ### Added | |||
|
|
|||
| - Added support for function `start_callback` to provide immediate feedback | |||
There was a problem hiding this comment.
start_callback feels like an odd name to me, but i see now that it was already established as a name with pipecat 👍
There was a problem hiding this comment.
Yeah, it's a weird name, but it's the same name used in Pipecat...
CHANGELOG.md
Outdated
|
|
||
| ```python | ||
| # Define a start callback | ||
| async def before_check_availability(function_name: str, llm: Any, context: Any) -> None: |
There was a problem hiding this comment.
In context, the args here feel kind of arbitrary to me. I understand that start_callback in pipecat happens to be part of LLMService so the arguments there are scoped to things that an LLMService would know about, but...in the context of defining a flow, it doesn't make much sense why this callback would only have access to the LLM and the LLM context (as well as a function name, which is something we don't do for any other handler/action/what-have-you...)
There was a problem hiding this comment.
If we could choose the args we wanted for this callback rather than just go with what pipecat's LLMService exposes, what would we want? I feel like we might want to provide the FlowManager wholesale, like we decided to do for pre-/post-actions.
| ) -> None: | ||
| if start_callback: | ||
| try: | ||
| # Pass flow_manager instead of llm and context |
There was a problem hiding this comment.
Ok, I was trying to figure out why we had created a wrapper for it. I can see it now. 👍
There was a problem hiding this comment.
Yup, this was in response to Paul's comment. We wanted to avoid using the default args from Pipecat and instead offer the FlowManager as an arg.
Pipecat's LLMServices support a
start_callbackbefore a function call runs. This callback allows you to execute functionality before the LLM function call. It's handy when you want to push a frame or run functionality before the function call takes place.