Support OpenAI-compatible API endpoints for all LLM providers

[Chungzter]

We should unify LLM interaction in commizard to use OpenAI compatibleAPI format instead of provider-specific endpoints.

This will allow the same code path to work for:

• Local Ollama (which supports OpenAI compatibility)
• OpenAI itself with popular GPT models
• Future providers that offer OpenAI-compatible APIs (Grok, Anthropic, Together, Fireworks, etc.)

Why not separate paths

• Reduces code duplication
• Easier to maintain and extend
• Ollama's OpenAI compatibility is stable enough for our use case (simple prompt to commit message generation)

Tasks

• Create a new, minimal LLM client abstraction

  • a system that produces:
    • base_url (default: http://localhost:11434/v1 for Ollama)
    • api_key (optional)
    • model name

• Refactor core generation function

  • Replace current Ollama /api/generate or other api/* calls
  • Map our prompt to messages list (single user message for now)

• Add configuration / switching

  • Store in config.py states: base_url, api_key,
  • Environment variable fallback (OPENAI_API_KEY, OLLAMA_HOST, etc.)

• Update README.md with changes

  • This OpenAI support from Ollama change will need a specific version of Ollama, so this requirement should be visibly stated.

later good features to keep in mind:

• Automatic fallback if one provider fails
• Support for /v1/completions (non-chat) as fallback

[Chungzter]

Update: After several hours wasted on reading up on unavailable compatible v1/responses endpoint, I found out that the support for this endpoint in Ollama has just been added 3 weeks ago! So the only way is to stick with the older chat/completions endpoint for now and possibly switch to v1/responses once chat/completions possibly gets deprecated in favor of v1/responses. For now though, chat/completions is a solid choice, even though anthropic is pushing it's own version, since our use case is very limited in scope.

Here's the link for docs on chat/completions: https://platform.openai.com/docs/api-reference/chat?lang=curl