Remote Ollama Server

[revoun]

so has anyone got shell_gpt to work with a remote ollama Server ?
I have a ollama server on a remote machine in my local network, and try to use shell_gpt on my local pc.
But sgpt still trying to connect to localhost ... How can i connect sgpt to a specific Server ?

[will-wright-eng]

check out ollama's docs for setting up a server

[8ar10der]

Hi @will-wright-eng

I can not find the solution in the link you mentioned.
I have the same scenario. I deployed my ollama instance in my homelab (e.g. http://192.168.1.2:11434). How can I let the sgpt to access with the address?

[spennell]

Follow the instructions listed here to install the LiteLLM piece for SGPT - https://github.com/TheR1D/shell_gpt/wiki/Ollama

When editing your ~/.config/shell_gpt/.sgptrc config update the following settings like below, just make sure you replace my URL with yours.

OPENAI_USE_FUNCTIONS=false
API_BASE_URL=http://ai:11434
USE_LITELLM=true
OPENAI_API_KEY=testkey

[8ar10der]

Thanks @spennell . I found I just forget to add the "ollama/" prefix in the DEFAULT_MODEL value. Now all work well. So add one more line in @spennell answer:

DEFAULT_MODEL=ollama/deepseek-coder-v2:latest # Change the model name by you want

OPENAI_USE_FUNCTIONS=false
API_BASE_URL=http://ai:11434
USE_LITELLM=true
OPENAI_API_KEY=testkey

[kov]

Adding the above to the guide would be really useful. Also, litellm can now provide a bearer token for the ollama_chat provider (they still need to add it to the ollama one), so it would be nice to allow passing an api key as well. I got it to work here by just removing the .pop():

diff --git a/sgpt/handlers/handler.py b/sgpt/handlers/handler.py
index a17d802..c630213 100644
--- a/sgpt/handlers/handler.py
+++ b/sgpt/handlers/handler.py
@@ -23,7 +23,6 @@ if use_litellm:
 
     completion = litellm.completion
     litellm.suppress_debug_info = True
-    additional_kwargs.pop("api_key")
 else:
     from openai import OpenAI