Collin Dutter comments

Results 87 comments of


                                            Collin Dutter

Add `RunnableMixin` for pre/post processing on arbitrary run execution

I'd like to hear more about what kind of pre/post processing you're envisioning. Is it mostly for data coercion of LLM output?

Allow passing k/v pairs to a custom J2 template

This should be implemented through the following: 1. Add `task: self` to `BaseTask.full_context`. 2. Add `context: self.full_context` as arguments to all Task uses of `J2.render`.

Allow passing k/v pairs to a custom J2 template

@vachillo I know this is quite old but can you expand on the ask here? Can you provide a code snippet that shows what you're trying to do?

if `client` is passed as a parameter, other fields should be optional

This is potentially a problem for any co-dependent public fields. One option is to make all fields optional and then use validators to ensure a valid combination of fields. We...

`Workflow.run()` will never return if a `Task` raises an uncaught exception

@vachillo can you provide a reproducible example?

`Workflow.run()` will never return if a `Task` raises an uncaught exception

@vachillo bump

Support OpenAi Realtime Endpoints

Blocked on #1280

feature request: return to previous state

@fgmacedo could somehow `source` be used here? Something like: ```python def on_ask_if_exit(self, source: State) -> None: user_input = input("Answer: ") if user_input == "exit": self.send("exit") else: # Return to the...

feat: add support for gpt-image-1

> as good as it can get given the state of the image gen API. approved after 3.9 issues are fixed I want to do another pass of manual testing...

Use underlying SDKs to automatically fill in model options when creating drivers

Blocked on #1587. Need to move imports out of `TYPE_CHECKING` to accomplish this.