Post

Who is going to be the first ATProto dev to build an LLM based feature that uses the auth'ed user to provision an API key and do the inference on behalf of the user?

Bailey Townsend 🦀@pds.dad

Oh yeah and @cocore.dev does for api keys (and other things) Probably enough but if you need more let me know and I’ll keep thinking cocore.dev/docs/api

cocore.devAppView API · co/coreRead-only XRPC queries over the co/core AppView index — receipts, jobs, providers, and social graph state.

25/6/26, 2:09 pm0 reposts33 likes

mary🐇

@mary.my.id

27 June

why not skip the whole api key and allow hitting the inference endpoint with an atproto service token 🤔

We're on it!

What I really want is the ability to request custom models

QUIET DAMMIT

Oh now *that* is a good idea! Damn you and your nersniping 😂

Vika 🏳️‍⚧️

@vikanezrimaya.xyz

27 June

@xan.lol hey, doesn't witchsky need an LLM API for alt text?…

xan

@xan.lol

27 June

yes, although idk if any of the models on cocore support reading images? should see to make cocore work if it does

co/core

@cocore.dev

27 June

We just added image support! It should work on 0.9.31 (which of course means to work right we need the ability to target machines on a given minimum version number, argh opening another PR), which should be all we need for this!

co/core

@cocore.dev

27 June

@xan.lol I'll (@devingaffney.com here) get you set up here in a few minutes and provision two machines and DM you with enough details for how to test it!

co/core

@cocore.dev

27 June

Many such cases