Unable to connect to proxied model from python api with self-signed SSL certificate #288

New Issue

2026-02-15T16:29:36-05:00

yindo commented

2026-02-15 16:29:36 -05:00

Originally created by @s-andrews on GitHub (Aug 8, 2025).

I'm running an ollama model on a linux system which I'm proxying through an apache server to a subdirectory, so the model will is running on

http://localhost:12345

..is proxied to https://webserver/ABCDEF/

This all works, and I can see the correct responses in the web server api, eg https://webserver/ABCDEF/api/tags returns a valid json document.

I can connect to the command line client OK and get the results from that.

C:\Users\me>set OLLAMA_HOST=https://webserver/ollama/PPKDEZXGHKWEKNOZCWSS/

C:\Users\me>ollama list
NAME           ID              SIZE     MODIFIED
gpt-oss:20b    f2b8351c629c    13 GB    36 minutes ago

C:\Users\me>ollama run gpt-oss:20b
>>> say hi
Thinking...
User just says "say hi". Should respond politely.
...done thinking.

Hello! 👋 How can I help you today?

I can also use the same URL to connect to page assist and that works.

When I try the same thing in the python API I can't get it to work though.

#!python

import ollama

client = ollama.Client(host="https://webserver/ollama/PPKDEZXGHKWEKNOZCWSS/")

response = client.chat(
    model="gpt-oss:20b", 
    messages=[
        {
            "role":"user",
            "content":"Say hi"
        }
    ]
)

print(response.message.content)

...fails with...

Traceback (most recent call last):
  File "C:\Users\andrewss\Desktop\Ollama\ollama_test.py", line 7, in <module>
    response = client.chat(
        model="gpt-oss:20b",
    ...<5 lines>...
        ]
    )
  File "C:\Users\andrewss\Desktop\Ollama\venv\Lib\site-packages\ollama\_client.py", line 342, in chat
    return self._request(
           ~~~~~~~~~~~~~^
      ChatResponse,
      ^^^^^^^^^^^^^
    ...<12 lines>...
      stream=stream,
      ^^^^^^^^^^^^^^
    )
    ^
  File "C:\Users\andrewss\Desktop\Ollama\venv\Lib\site-packages\ollama\_client.py", line 180, in _request
    return cls(**self._request_raw(*args, **kwargs).json())
                 ~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
  File "C:\Users\andrewss\Desktop\Ollama\venv\Lib\site-packages\ollama\_client.py", line 126, in _request_raw
    raise ConnectionError(CONNECTION_ERROR_MESSAGE) from None
ConnectionError: Failed to connect to Ollama. Please check that Ollama is downloaded, running and accessible. https://ollama.com/download

The same script works on the linux host machine if I point to the http://localhost:12345 address, so it's something about the use of the proxy which is causing the failure.

Any suggestions for how to fix or debug this some more would be appreciated. Thanks!

Originally created by @s-andrews on GitHub (Aug 8, 2025). I'm running an ollama model on a linux system which I'm proxying through an apache server to a subdirectory, so the model will is running on ```http://localhost:12345``` ..is proxied to ```https://webserver/ABCDEF/``` This all works, and I can see the correct responses in the web server api, eg ```https://webserver/ABCDEF/api/tags``` returns a valid json document. I can connect to the command line client OK and get the results from that. ``` C:\Users\me>set OLLAMA_HOST=https://webserver/ollama/PPKDEZXGHKWEKNOZCWSS/ C:\Users\me>ollama list NAME ID SIZE MODIFIED gpt-oss:20b f2b8351c629c 13 GB 36 minutes ago C:\Users\me>ollama run gpt-oss:20b >>> say hi Thinking... User just says "say hi". Should respond politely. ...done thinking. Hello! 👋 How can I help you today? ``` I can also use the same URL to connect to page assist and that works. When I try the same thing in the python API I can't get it to work though. ``` #!python import ollama client = ollama.Client(host="https://webserver/ollama/PPKDEZXGHKWEKNOZCWSS/") response = client.chat( model="gpt-oss:20b", messages=[ { "role":"user", "content":"Say hi" } ] ) print(response.message.content) ``` ...fails with... ``` Traceback (most recent call last): File "C:\Users\andrewss\Desktop\Ollama\ollama_test.py", line 7, in <module> response = client.chat( model="gpt-oss:20b", ...<5 lines>... ] ) File "C:\Users\andrewss\Desktop\Ollama\venv\Lib\site-packages\ollama\_client.py", line 342, in chat return self._request( ~~~~~~~~~~~~~^ ChatResponse, ^^^^^^^^^^^^^ ...<12 lines>... stream=stream, ^^^^^^^^^^^^^^ ) ^ File "C:\Users\andrewss\Desktop\Ollama\venv\Lib\site-packages\ollama\_client.py", line 180, in _request return cls(**self._request_raw(*args, **kwargs).json()) ~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^ File "C:\Users\andrewss\Desktop\Ollama\venv\Lib\site-packages\ollama\_client.py", line 126, in _request_raw raise ConnectionError(CONNECTION_ERROR_MESSAGE) from None ConnectionError: Failed to connect to Ollama. Please check that Ollama is downloaded, running and accessible. https://ollama.com/download ``` The same script works on the linux host machine if I point to the http://localhost:12345 address, so it's something about the use of the proxy which is causing the failure. Any suggestions for how to fix or debug this some more would be appreciated. Thanks!

yindo closed this issue

2026-02-15 16:29:36 -05:00

yindo commented

2026-02-15 16:29:37 -05:00

@s-andrews commented on GitHub (Aug 28, 2025):

I figured this out. It turns out it wasn't the proxy which was the issue, it was our site's security system which generates fake SSL certificates to screen https content and then installs a new root certificate so that everything works. The httpx client in ollama doesn't use the site certificate store, so the SSL doesn't verify.

If I add a breakpoint into _client.py I can grab the actual connect error where I see:

ConnectError('[SSL: CERTIFICATE_VERIFY_FAILED] certificate verify failed: unable to get local issuer certificate (_ssl.c:1028)')

The fix for this is in the httpx documentation where it shows how you can force the use of the system certificate store rather than the bundled one.

In my case the fixed code is below, and this works as expected.

#!python

import ollama
import truststore
import ssl

client = ollama.Client(
    host="https://capstone.babraham.ac.uk/ollama/DHIWJCSBEOKIGOAYYNTV/",
    verify=truststore.SSLContext(ssl.PROTOCOL_TLS_CLIENT)
)

response = client.chat(
    model="gpt-oss:20b", 
    messages=[
        {
            "role":"user",
            "content":"Say hi"
        }
    ]
)

print(response.message.content)

@s-andrews commented on GitHub (Aug 28, 2025): I figured this out. It turns out it wasn't the proxy which was the issue, it was our site's security system which generates fake SSL certificates to screen https content and then installs a new root certificate so that everything works. The httpx client in ollama doesn't use the site certificate store, so the SSL doesn't verify. If I add a breakpoint into ```_client.py``` I can grab the actual connect error where I see: ```ConnectError('[SSL: CERTIFICATE_VERIFY_FAILED] certificate verify failed: unable to get local issuer certificate (_ssl.c:1028)')``` The fix for this is in the [httpx documentation](https://www.python-httpx.org/advanced/ssl/) where it shows how you can force the use of the system certificate store rather than the bundled one. In my case the fixed code is below, and this works as expected. ``` #!python import ollama import truststore import ssl client = ollama.Client( host="https://capstone.babraham.ac.uk/ollama/DHIWJCSBEOKIGOAYYNTV/", verify=truststore.SSLContext(ssl.PROTOCOL_TLS_CLIENT) ) response = client.chat( model="gpt-oss:20b", messages=[ { "role":"user", "content":"Say hi" } ] ) print(response.message.content) ```

Sign in to join this conversation.

1 Participants

Notifications

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: ollama/ollama-python#288