agrafix / openai-hs

Unofficial OpenAI Haskell Bindings

Unofficial OpenAI Haskell Bindings

Unofficial OpenAI SDK/client for Haskell. It's generated via servant-client from openai-servant with a small amount of hand-written code. Contributions are welcome!

See openai-hs for more details.

openai-hs's Issues

caching best practices?

anyone got thoughts on the best way to do caching of responses? i've been trying to do it at the HTTP level with squid, but it really starts to get fragile and baroque when you're dealing with POSTs.

Version housekeeping

I think you should bump the version with the significant Mar 23 update and push it to hackage.
Currently Hackage and this repo are both 0.2.2.0 (https://hackage.haskell.org/package/openai-hs-0.2.2.0) where the current state should probably be called 0.3?

`arguments` field of `ChatFunctionCall` should be `Text` rather than `Value`

The arguments field of a function call object is not guaranteed to be valid JSON, here's the relevant quote from the API reference

The arguments to call the function with, as generated by the model in JSON format. Note that the model does not always generate valid JSON, and may hallucinate parameters not defined by your function schema. Validate the arguments in your code before calling your function.

Currently if there is a problem in arguments we get a DecodeFailure ... :: ClientError from completeChat, which makes it harder to recover from, because we do not have access to the other parts of the response, which are likely valid.

Add to list of community maintained libraries?

Hi there, this library looks pretty cool. Have you thought about adding it to the list of community maintained libraries? How "complete" is the current state of the library?

Adding your API client to the Community Libraries page

Best regards

Add `function_call` to `ChatCompletionMessage`

The documentation is a bit hairy about it, but apparently a ChatCompletionMessage can take a function_call parameter to force calling a particular function specified in the functions list, see for example:

• https://platform.openai.com/docs/api-reference/chat/create#function_call
• https://platform.openai.com/docs/guides/gpt/function-calling (see the code example)
• https://github.com/openai/openai-cookbook/blob/main/examples/How_to_call_functions_with_chat_models.ipynb (Section "Forcing the use of specific functions or no function")

From a passage of the first link:

If you want to force the model to call a specific function you can do so by setting function_call: {"name": ""}. You can also force the model to generate a user-facing message by setting function_call: "none". Note that the default behavior (function_call: "auto") is for the model to decide on its own whether to call a function and if so which function to call.

According to the documentation, null is a valid value, so this could be modelled as a Maybe ChatFunction or similar.

Rate limiting support.

When doing requests in parallel it's not too hard to hit rate limits:
https://platform.openai.com/docs/guides/rate-limits

Would be great to have rate limiting support, e.g. across all clients generated from the same Manager.

add cpphs to build-depends?

tehres a cabal field for build tool deps, then it wouldn't fail for users who dont have cpphs installed...

Add to Stackage

This is a very helpful package and I think it would be cool to include in Stackage!

https://github.com/commercialhaskell/stackage/blob/master/MAINTAINERS.md#adding-a-package