#what ip are you using?
wooden marten
i am using same ip since it's on same server
just different port
again if i run a manual curl it works fine
curl -s http://192.168.25.15:8000/v1/models -H "Authorization: Bearer KEY" | jq '.data[].id' - this works:
"Qwen3.6-35B-A3B-8bit"
"gemma-4-26B-A4B-it-MLX-8bit"
narrow wadi
what ip do you have set in the config, does it have the /v1 on the end?
And you selected to add ollama as a custom provider right?
This is how I have mine setup currently:
custom_providers:
- base_url: http://127.0.0.1:11434/v1
model: glm-5.1:cloud
name: Ollama Gemma4 Cloud - base_url: http://127.0.0.1:11434/v1
model: glm-5.1:cloud
name: Ollama glm-5.1 Cloud
wooden marten
yes i double checked as taht bit me last time
i'm using local ollama
and it works with port 11434 - not when i changed it to omlx port 8000
narrow wadi
Should be the same for local but the model is just different
wooden marten
yeah, that is what I thought but when i try to chat i get error:
⚠️ API call failed (attempt 1/3): AuthenticationError [HTTP 401]
🔌 Provider: custom Model: gemma-4-26B-A4B-it-MLX-8bit
🌐 Endpoint: http://192.168.25.15:8000/v1
📝 Error: HTTP 401: Invalid API key
📋 Details: {'message': 'Invalid API key', 'type': 'authentication_error', 'param': None, 'code': None}
⚠️ Non-retryable error (HTTP 401) — trying fallback...
❌ Non-retryable error (HTTP 401): HTTP 401: Invalid API key
❌ Non-retryable client error (HTTP 401). Aborting.
🔌 Provider: custom Model: gemma-4-26B-A4B-it-MLX-8bit
🌐 Endpoint: http://192.168.25.15:8000/v1
💡 Your API key was rejected by the provider. Check:
• Is the key valid? Run: hermes setup
• Does your account have access to gemma-4-26B-A4B-it-MLX-8bit?
but if i use curl manually it works fine
narrow wadi
Oh, with https://omlx.ai/ ?
narrow wadi
Hmm, I havent used that before. But it seems to be complaining about an API key. Is the api key present in the config?
Would also be worth checking the models tab in MLX to make sure the model is present
and when you curl it manually you are trying the MLX endpoint or ollama?
wooden marten
that was the result in my curl command it showed the available models
mlx
i shutdown ollama - not needed - this thing runs a server
that becomes the endpoint and curl works
narrow wadi
Hmm, well it's an OpenAI compatable API so I would think it should just work
narrow wadi
I am really not entirely sure tbh, I don't have any mac devices to try it myself either. I would think though if you can do it manually and then set the custom provider it would just work the same as ollama
wooden marten
only difference is ollama did not require an api key - omlx forces it
narrow wadi
Hmm but you said the API key is set in the config now?
narrow wadi
So it should be passing it but that is also the one thing it is mad about
wooden marten
agreed - and tried looking at it through dashboard but that was even more confusing...lol
narrow wadi
Like the web dashboard?
wooden marten
yeah the new one
narrow wadi
Hmm
Does MLX have any logs on it's side?
To see if hermes is even reaching it?
and is the key in the .env under "OPENAI_API_KEY"?
wooden marten
it's rejecting the call - i was stumbling a bit this morning then finally got the syntax right becuase their web page is incorrect for curl testing and once i got curl going to validate it was working - then i ran 'hermes setup' --> Custom --> Other
and setup the provider but when i pasted in the api key it stayed blank
narrow wadi
Yeah, when you paste the key in the terminal it wont show even if it is present
wooden marten
figured as much, security thing, i get it - just wasnt sure if that was normal
hmm looked at .env
narrow wadi
Yeah, the .env is the only way to make sure if they is is pasted correctly
wooden marten
it's under OLLAMA_API_KEY
narrow wadi
That might be the issue
wooden marten
maybe that issue
narrow wadi
I would guess it's looking for the open ai one
I think ollama api key is for if you specifically select ollama as a model provider and not custom
wooden marten
trying to read through documenation but it's not very helpful
narrow wadi
if you do "hermes model"
it will bring you right to the model config setup
then do custom endpoint
wooden marten
let me do it again
narrow wadi
Ok
wooden marten
oh crap - so i removed then re-added and this time it worked..THANK YOU SO MUCH - i think i got bit by the pasting of the key
narrow wadi
Haha, nice! Hopefully hermes works better after this pain point
wooden marten
thank you - truly appreciate you taking the time to walk me through this
narrow wadi
Yeah np haha
wooden marten
i would have banged my head forever.....lol
narrow wadi
I am only 4 days into this myself and I have delt with lots of setup pains
It's super cool once settled in though