Parse requests from Copilot

[gr2m]

Implement a method for parsing incoming requests (from Copilot).

Two reasons

1. Getting type intellisense for the shape of the request
2. Removing Copilot specific keys that cause OpenAI trouble (example)

So maybe it should be two low-level methods. The high-level SDK should do the parsing implicitly and give full type intellisense. And ideally would do the payload clean up transparently as well when passing to a model API other than CAPI.

Example

// `requestBody` can be either a JSON string or an object
const input = parseRequest(requestBody);
// `input` now has full TypeIntellisense

// pass through any input properties safely to OpenAI
const inputCompat = removeCopilotKeysFromRequest(input);
const stream = openai.beta.chat.completions.stream({
  messages: [systemMessage, ...input.data.messages],
  model: "gpt-4-1106-preview",
  stream: true,
});

[alexisabril]

I'm definitely in favor of the ability to type an incoming request shape. I'm agnostic, but not opposed, to the removeCopilotKeysFromRequest method as we may continue to add properties to the incoming request object. As a result, this method may need to have a higher level of upkeep and may be considered coupled (a bit) to OpenAI. For instance, Anthropic may consider the entire shape to be invalid, rendering this translation method moot.

verifyAndParse having a potential avenue across other APIs brings value to replicating this pattern here. For me personally, I'm a fan of two separate methods that can be composed:

verify(parse)(req)

The only thought to keeping the separate I have is if the consumer chooses an architecture where verification happens earlier and separately from any additional request handling.

[gr2m]

good point on making removeCopilotKeysFromRequest specific to OpenAI, I think we should do that.

[gr2m]

I added one more issue

• getUserConfirmation(requestPayload) as a convenience method to get the user confirmation (if there is any) #33

I'm sure a similar function would be useful for function calling, but I get there when I get there. I need to experiment more with

[github-actions]

🎉 This issue has been resolved in version 2.2.0 🎉

The release is available on:

• npm package (@latest dist-tag)
• GitHub release

Your semantic-release bot 📦🚀