Adding an assistant: endpoint presets and a model list from the provider issue #109
closed: completed andymccall opened this · 0 comments
andymccall
Andy, 2026-10-04, after trying the assistants on a real proposal (not for now):
The add form should offer the common providers with their endpoints filled in (Anthropic, OpenAI, DeepSeek, Mistral, Perplexity, Google, a local Ollama) and an Other for a URL typed by hand; picking one sets the API shape too.
Once an endpoint and token are given, fetch the provider's model list (OpenAI-style /v1/models, Anthropic's /v1/models) and offer it as a drop-down, so the model name cannot be mistyped; keep a free-text fallback for providers without a list.
The connection test on add already exists; it should run against the chosen model.
Also from today: a name typed with a typo ("DeekSeek") gets an identicon rather than the brand mark; the presets would name the assistant correctly by default.
Andy, 2026-10-04, after trying the assistants on a real proposal (not for now):
/v1/models, Anthropic's/v1/models) and offer it as a drop-down, so the model name cannot be mistyped; keep a free-text fallback for providers without a list.