Issue with Ollama

[LStromann]

=<@1069659243810144386> please move your problem here

This discussion was automatically pulled from Discord.

[LStromann]

tiagoriosrocha — 13/12/25, 1:42 AM
Did using Open WebUI work? I am trying but I don't have success yet....
I'm using Cognee to connect directly to Ollama. It's working fine. But I need to
connect to Ollama through Open WebUI, however I don't know how to configure the environment
variables and the compatible model for this. For now, here are the environment variables I'm
using:

LLM_API_KEY="sk-key"
LLM_MODEL="llama3:8b"
LLM_PROVIDER="ollama"
LLM_ENDPOINT="http://localhost:3000/api/chat/completions"
EMBEDDING_PROVIDER="ollama"
EMBEDDING_MODEL="avr/sfr-embedding-mistral:latest"
EMBEDDING_ENDPOINT="http://localhost:3000/ollama/api/embed"
EMBEDDING_DIMENSIONS=4096
HUGGINGFACE_TOKENIZER="Salesforce/SFR-Embedding-Mistral"

ERROR:

[2m2025-12-15T17:21:17.412371[0m [[31m[1merror [0m] [1m<failed_attempts>

Error code: 405 - {'detail': 'Method Not Allowed'} None Error code: 405 - {'detail': 'Method Not Allowed'} None Error code: 405 - {'detail': 'Method Not Allowed'} None Error code: 405 - {'detail': 'Method Not Allowed'} None Error code: 405 - {'detail': 'Method Not Allowed'} None

</failed_attempts>

<last_exception>
Error code: 405 - {'detail': 'Method Not Allowed'}
</last_exception>[0m [[0m[1m[34mcognee.shared.logging_utils[0m][0m

[2m2025-12-15T17:21:17.413564[0m [[31m[1merror [0m] [1mConnection to LLM could not be
established.

[LStromann]

tiagoriosrocha — 11:28 PM
My endpoints is working when I trying access directly, I tested using this codes: #############################
#CHAT OPEN WEBUI:

import requests

url = "http://localhost:3000/api/chat/completions"
headers = { "Authorization": f"Bearer sk-key" }
data = {
"model": "llama3:8b",
"messages": [
{
"role": "user",
"content": "Explain about Neural Networks."
}
]
}

response = requests.post(
url,
json=data,
headers=headers
)

if response.status_code != 200:
print("Erro:", response.status_code, response.text)
exit(1)

result = response.json()

answer = result["choices"][0]["message"]["content"]

print("Response:")
print(answer)

############################
#EMBEDDING OPEN WEBUI:

import requests

url = "http://localhost:3000/ollama/api/embed"
headers = {"Authorization": "Bearer sk-key"}
data = {
"model": "avr/sfr-embedding-mistral:latest",
"input": "what is the capital of Brazil?"
}

response = requests.post(url, json=data, headers=headers)

embedding = response.json()['embeddings'][0]

print(f"Size: {len(embedding)}")
print(f"vector start: {embedding[:5]}")
but don't work with cognee

[LStromann]

LLM API KEY="sk-737983c62cea4440b586867d7e85b270"
LLM MODEL="llama3:8b"
LLM_PROVIDER="custom"
LLM_ENDPOINT="http://localhost:3000/api/"
EMBEDDING PROVIDER="ollama"
EMBEDDING_MODEL="avr/sfr-embedding-mistral:latest"
EMBEDDING_ENDPOINT="http://localhost:3000/api/embeddings"
EMBEDDING DIMENSIONS=4096
HUGGINGFACE_TOKENIZER="Salesforce/SFR-Embedding-Mistral"

[LStromann]

can you try this once what error we are getting please share

[LStromann]

ok. here is the error for this environmental variables.

[LStromann]

LLM_MODEL="openai/llama3:8b"

[LStromann]

please try with this

[LStromann]

ok, follow the error, it seems to be the embedding model for now..

[LStromann]

when I change embedding environment variables the error is...
EMBEDDING_PROVIDER="custom"
EMBEDDING_MODEL="openai/vr/sfr-embedding-mistral:latest"

[LStromann]

please try this

LLM_API_KEY="sk-737983c62cea4440b586867d7e85b270"
LLM_MODEL="openai/llama3:8b"
LLM_PROVIDER="custom"
LLM_ENDPOINT="http://localhost:3000/api/"

EMBEDDING_PROVIDER="openai"
EMBEDDING_MODEL="avr/sfr-embedding-mistral:latest"
EMBEDDING_ENDPOINT="http://localhost:3000/api/"
EMBEDDING_API_KEY="sk-737983c62cea4440b586867d7e85b270"
EMBEDDING_DIMENSIONS=4096

[LStromann]

Ok. The error was huge! I can't capture all.. I used this variables:
LLM_API_KEY="sk-737983c62cea4440b586867d7e85b270"
LLM_MODEL="openai/llama3:8b"
LLM_PROVIDER="custom"
LLM_ENDPOINT="http://localhost:3000/api/"

EMBEDDING_API_KEY="sk-737983c62cea4440b586867d7e85b270"
EMBEDDING_MODEL="avr/sfr-embedding-mistral:latest"
EMBEDDING_PROVIDER="openai"
EMBEDDING_ENDPOINT="http://localhost:3000/api/"
EMBEDDING_DIMENSIONS=4096

[LStromann]

can you please try this one too

LLM_API_KEY="sk-737983c62cea4440b586867d7e85b270"
LLM_MODEL="llama3:8b"
LLM_PROVIDER="ollama"
LLM_ENDPOINT="http://localhost:3000/ollama/v1/"

EMBEDDING_API_KEY="sk-737983c62cea4440b586867d7e85b270"
EMBEDDING_MODEL="avr/sfr-embedding-mistral:latest"
EMBEDDING_PROVIDER="openai"
EMBEDDING_ENDPOINT="http://localhost:3000/api/"
EMBEDDING_DIMENSIONS=4096

[LStromann]

follow the error (without HUGGINGFACE_TOKENIZER)

[LStromann]

when I put the HUGGINFACE_TOKENIZER, the error is:

[LStromann]

when I play the python script, I run the command: "ollama ps". I can see the model llam3:8b running.. In this way, I understood that it is connecting and running... the problem probably is the embedding connection for now...

[LStromann]

If I am using EMBEDDING_PROVIDER="openai", I need remove HUGGINGFACE_TOKENIZER="Salesforce/SFR-Embedding-Mistral"? I am confusing about it.

[LStromann]

I don't know if the EMBEDDING_MODEL="avr/sfr-embedding-mistral:latest" is compatible with openai...

[LStromann]

May you change the topic name for: Issue with cognee and Ollama connection through Open WebUI.

[LStromann]

Issue with cognee and Ollama connection through Open WebUI

[LStromann]

I have tried using ollama with cognee i didn't try it with open web ui

[LStromann]

But I will suggest use BAML

[LStromann]

Instead of instructor

[LStromann]

It works well