it's better to support ollama

[YinYongHongYork]

Do you have a plan to support ollama? thanks in advance.

[mattzcarey]

hey as mentioned here, I will not be putting time in to support multiple model providers. These will have to come from the community and they will be use at your own discretion :)

We have had successful contributions for Azure OpenAI and an ongoing one for Deepseek. Ollama has an added complexity as you would mostly likely need GPUs to run reasonable models in which case using Mac runners with MLX backend or external Nvidia (which i cannot test) is prefered.

[kamilhawdziejuk]

Hi, I believe this is related. I was able to download qwen2.5-coder LLM model on locally running ollama and if this was the question than.. configuration actually allows to support that setup. I was successfully able to change args.ts configuration to make it work!

.option('modelString', {
description:
'The model to use for generating the review. Defaults to "openai:gpt-4o-mini".',
type: 'string',
//default: 'openai:gpt-4.1-mini',
default: 'openai:qwen2.5-coder'
})
.option('reviewLanguage', {
description: 'Specifies the target natural language for translation',
type: 'string',
default: 'English',
})
.option('platform', {
description: 'Platform type',
choices: ['github', 'gitlab', 'azdev', 'local'],
type: 'string',
default: 'local',
})
.option('maxSteps', {
description: 'Maximum number of agentic steps to take',
type: 'number',
default: 25,
})
.option('baseUrl', {
description: 'Base URL for the platform',
type: 'string',
default: 'http://localhost:11434/v1'
})

[mattzcarey]

This makes me so happy. Thanks for sharing this :)

[mattzcarey]

To add to this that means you should be able to make it work with.

npx shippie review --modelString=openai:qwen2.5-coder --baseUrl=http://localhost:11434/v1