search

karthink / gptel

A simple LLM client for Emacs
GNU General Public License v3.0
1.14k stars 119 forks source link

Just an idea #39

Closed randomwangran closed 10 months ago

randomwangran commented 1 year ago

https://github.com/karthink/gptel/blob/e61180ebc59758a2298ae2441b2fd102d80ec2eb/gptel.el#L376

If TODO -> DONE, things like sectional suggestions can be achieved:

https://youtu.be/MpK9PWo0lUw?t=115

randomwangran commented 1 year ago

15 related

karthink commented 1 year ago

https://youtu.be/MpK9PWo0lUw?t=115

I'm not sure how this feature connects with the TODO item the code. Could you explain in a little more detail?

psionic-k commented 1 year ago

Fundamentally, since we can only edit on one dimension of a request at a time, I think what they mean is to map the user's command over either a list of data or using a list of contexts, prompts, or other request dimensions.

Multiple requests only makes sense when the things being mapped over do not provide context to each other or the goal is to eliminate context leakage from item to item. If the structure of output is difficult to generate correctly, that's another possibility.

If anything is needed, it's probably just the infrastructure to support multiple simultaneous requests. A plural API would be convenient, but really just front-loading some of the plumbing that users can / should do on their own.

karthink commented 1 year ago

If anything is needed, it's probably just the infrastructure to support multiple simultaneous requests.

gptel does not currently enforce any limit on the number of simultaneous requests (in a single buffer or across buffers), so you can fire off as many as you want. I often have 2 requests going at once.

psionic-k commented 1 year ago

infrastructure: the Open AI API calling implementation code. (I didn't look at it yet) plural API: one that handles lists of things instead of a thing