Summary

The DeepSeek BYOK connector in the UI only offers the legacy models deepseek-chat and deepseek-r1. DeepSeek's API now exposes the V4 model family (deepseek-v4-pro, deepseek-v4-flash), which cannot be selected because the model picker for DeepSeek is a fixed <select> populated solely from provider.models.

Where

internal/staticserve/static/ui-connectors/providers.js

{
  id: 'deepseek',
  name: 'DeepSeek',
  defaultModel: 'deepseek-chat',
  models: ['deepseek-chat', 'deepseek-r1'],
  ...
}

For non-Ollama providers, ConnectorFormPage.js renders a <select> whose options come only from provider.models, so users cannot type a custom model id — the V4 models are simply unreachable from the UI.

Expected

DeepSeek's current models per the official API docs (https://api-docs.deepseek.com/zh-cn/api/create-chat-completion) should be selectable:

• deepseek-v4-pro
• deepseek-v4-flash

Verification

• Direct call to https://api.deepseek.com/chat/completions with model: "deepseek-v4-pro" → HTTP 200, response echoes "model":"deepseek-v4-pro".
• Through the real lrc path (lrc ui → /api/ui/connectors/validate-key → LiveReview cloud → DeepSeek): deepseek-v4-pro → {"valid":true}, while a bogus model id → {"valid":false}, confirming the cloud actually probes the model, not just the key.

Proposed fix

Add the two V4 models to the DeepSeek models list (keeping the existing two for backward compatibility) and default to deepseek-v4-flash. PR to follow.