Feature Request: When the user has defined an LLM ...
# proj-agent-canvas
c
Feature Request: When the user has defined an LLM Profile, from e.g. Anthropic, then the next time they define another Anthropic LLM we could recognize the key and use it. Otherwise, well, the user has to go chase for where that key was... 😅
acknowledged 2
nod 2
👍 1
To be honest, this isn't really a canvas issue, I think maybe it's an SDK issue, to define / read LLM keys not per LLM, and not primarily per profile, rather: • per provider first (and mostly) • fallback per profile (for the case when the user really wants to use multiple keys per provider, e.g. our litellm proxies case)
h
In a similar line of thinking I'd like to get export/importing for profiles as a feature too. So I can easily setup multiple profiles and move them between installs or make it easier to re-set up if I clear out my .openhands folder to start fresh
nod 1
c
In theory, that should be simple because they’re .json files in a directory, just copy should do well… except for keys
Right now, when used via agent-server the keys are encrypted (I think)
h
What if you could save the LLM API key and then re-use them. Maybe an LLM API view then each LLM profile uses that key
c
One thing I’d like to get to, is to store them in Apple keyring (and the ~equivalents on other systems).
Once done, keys are there, the rest of an LLM Profile is just parameters
and they’re in .json
h
That makes sense
Is there anything like that for windows/linux? I don't know of anything
c
Pretty sure yes, let me double check quick
• GNOME Keyring (ships with GNOME/Ubuntu) • KWallet (KDE) • Windows Credential Manager (wincred) Python
keyring
should work at least on Linux and Mac, and according to Claude also on Windows
(dunno if I fully trust it 😅)
h
TIL. I've never heard of wincred
nod 1