Feature Request: Integration of OpenAI-compatible LLM Translation Services

[mengdehong]

Currently, Dialect primarily relies on traditional translation backends (Google, Bing, LibreTranslate, etc.). While these are effective, they lack the contextual nuance that modern LLMs provide. Integrating AI-based translation would significantly enhance the quality of translations, especially for complex or idiomatic phrases.

I would like to propose adding support for OpenAI-compatible LLM services. This would allow users to leverage models like GPT-5, DeepSeek, or even local models hosted via llama.cpp/vLLM.

Key configuration options should include:

1. API Key for authentication.
2. Base URL to support custom endpoints
3. Model Selection
4. SSE results returned

Additional:
I'm implementing this feature in my personal fork, and I'd like to know what the source repositories are doing about it.

2026-04-27.11-04-18.mp4

I am open to discussing how to best integrate this into the upstream project and can refine the code based on the project's standards.

[rafaelmardojai]

Thanks for reaching out. We have been recently discussing about this with @mufeedali after getting this PR #461.

We've been thinking that we would like an OpenAI-API provider, given that it's the de-facto API for LLM services and would cover most AI use cases. We're still defining what additional settings we would like to expose to the user, like promt, languages list, etc.

This also opens the possibility of creating an API for providers to define arbitrary custom setting fields.