#75 Add built-in presets for common OpenAI-compatible providers

[Anonymous]

Originally created by: sosidudku1

Summary

Only openrouter and aimlapi ship as named providers. Everything else goes through the generic openai-compatible option, where the user types a base URL, an API key and an exact model id by hand.

Users are already doing this successfully, so this is not about capability, it is about setup friction and typos.

User feedback (Discord)

"I did put it through oai compatible, yes I want more presets"

"Ironically I use nous instead of openrouter since you get more bang for your buck for literally the same service"

Proposal

Ship presets for popular OpenAI-compatible endpoints so the user picks a name instead of typing a URL. Each preset pre-fills baseUrl, sensible defaults and, where the endpoint supports it, a model list fetched from /v1/models.

Candidates: Nous, DeepSeek (direct), Groq, Together, Fireworks, Ollama Cloud, LM Studio (local).

The generic openai-compatible option stays for anything not on the list.

Where

src/llm/provider/registry/register-built-in-providers.ts already registers llama-server, openai-compatible, openrouter and aimlapi. Presets could be a data table consumed by the setup flow rather than new provider kinds.

Related

Auto-fetching /v1/models for openai-compatible endpoints (#31) pairs naturally with this: the preset picks the endpoint, the model list populates itself.

[Anonymous]

Originally posted by: sosidudku1

Duplicate of [#69], which covers the same request. Closing this one to keep the discussion in a single place.

[Anonymous]

Ticket changed by: sosidudku1

• status: open --> closed