AI model can't connect

[D1st3f]

List of model from providers not working + custom model always 404

[stefan-ernst]

Hey, thanks for reporting. Is networking / egress working? Do you see anything in the log?

[D1st3f]

It was csrf prolem. I cant connect to local ip(

[stefan-ernst]

I see - I will add an allowLocalLLM config flag - the default is secure but too strict

[kovak1]

I can't use the local AI model either (LM Studio hosted)
windshift-v0.7.2-windows-amd64

Before running the exe file, I configured the following:
set LLM_ALLOWED_PRIVATE_CIDRS=127.0.0.1/32,::1/128

2026/06/08 12:57:28 WARN LLM connection test failed connection_id=5 error="failed to connect to LLM service: failed to connect to LLM service: Post "http://localhost:11434/v1/chat/completions\": connection refused"

[stefan-ernst]

Hey @kovak1 - thanks for reporting. Are you on the latest main tag? Because that flag has not been shipped yet

Current main is a bit unstable, I will have to test everything well then put out 0.7.3

[kovak1]

Okay, I get it now!
So it won't work at all with 0.7.2, but the allowLocalLLM flag will work starting with 0.7.3.

This is a great project—it’s very useful, and I’m learning a lot from it. Keep up the good work!

[stefan-ernst]

Thank you for your kind words! Yes, its a bit "in progress" at the moment. Hopefully I can wrap up release 0.7.3 soon!

In the meantime, if you select the :main tag or the latest commit id (:9c3ff69) you can pull the docker image from the latest commits. Hope this enables the override parameters

[stefan-ernst]

Hey guys, the new 0.8 release now adds: ALLOW_LOCAL_CONNECTIONS=true - simple, allows all local connections for all endpoints instead of relying on separate allowlists. so the previous LLM_ALLOWED_PRIVATE_CIDRS is not needed anymore