Experimental galactic wizard #15860
Experimental galactic wizard #15860
Conversation
|
This is awesome - can definitely see some version of this being a huge asset (to new users especially). What was the response time roughly? |
|
@neoformit It's pretty fast in my hands. Admittedly without specific measurements, and it does seem to fluctuate with load, but the response time is generally <1s. |
client/src/schema/schema.ts
Outdated
| "/api/chat": { | ||
| /** | ||
| * Query | ||
| * @description We're off to ask the wizard |
dannon
force-pushed
the
galactic-wizardry
branch
2 times, most recently
from
764e420
to
bfefa20
Compare
dannon
force-pushed
the
galactic-wizardry
branch
2 times, most recently
from
b35a2d4
to
234558d
Compare
dannon
changed the title
dannon
force-pushed
the
galactic-wizardry
branch
from
234558d
to
41be831
Compare
|
This needs another run of |
Most of this was from a branch that didn't share a common base with the ChatGXY PR and didn't cleanly cherry-pick unfortunately, hence the manual additions and massaging.
Don't need the model manager; keep it simpler.
Co-authored-by: John Davis <[email protected]>
dannon
force-pushed
the
galactic-wizardry
branch
from
00fe1a4
to
c98216d
Compare
… query for shared/errored datasets, etc)
dannon
force-pushed
the
galactic-wizardry
branch
from
a13c2dc
to
4de179a
Compare
|
This is really cool stuff, thanks! We're off to ask the wizard.... |
|
Thanks for the review and approval @jdavcs -- since the tests passed on the last run I'm going to go ahead and save CI time after making those last minor requested tweaks and sanity checking locally. I'm working on a list of follow-ups, but I'm going to merge it so we can get this into the release and continue to iterate on it. |
|
This PR was merged without a "kind/" label, please correct. |
Talked with several folks who were interested in this last week so I'm sharing this basic implementation of a galactic wizard / chat assistant that I was playing with. Not currently linked into the UI, just go to
<instance>/wizard.This interface could be a chat box (though responses are generally longer than one would want to see there), or whatever else.
Tinkering with the prompt (should probably be made configurable) is something folks can explore and for a real deployment we'd almost certainly want to pursue something similar to https://supabase.com/blog/chatgpt-supabase-docs to contextualize answers using the contents of the GTN, GHelp, or the hub.
Right now, this also (intentionally) doesn't allow follow-up questions to avoid having to track/store context, but it would be straightforward to add.
Anyway, this could go in any number of directions -- just wanted to toss it out here for folks to play with and think about.
How to test the changes?
(Select all options that apply)
License