mirror of
https://github.com/MoonshotAI/kimi-code.git
synced 2026-08-29 20:12:37 +00:00
c0fede246d
* feat(desktop): add custom provider management in settings Add a Providers tab to the settings dialog where users manage API providers in place (accordion rows; view = edit = add): - Manual add/edit/rename/delete with a unified form (wire protocol, api key with password field + visibility toggle + stored-key prefill, base URL, model rows); renames migrate aliases and default pointers server-side; provider writes never move the global default model. - Import from the server-proxied models.dev directory: searchable catalog with ineligible entries greyed out, minimal import form (name / API key / base URL only when required). - Import from a private api.json registry (URL + optional Bearer key); re-importing the same URL refreshes it. - Old-daemon degradation: when the connected server predates the catalog routes the directory source hides and the flow falls back to manual-only. web-core gains the matching client methods (add/update/delete provider, catalog list/get/import, registry import) with snake_case wire types and envelope parsing hardened for bare 204s and non-envelope errors. Bumps the kimi-code submodule to the provider-write-endpoints branch (MoonshotAI/kimi-code PR #2110) which backs every new endpoint; the pointer moves to the main merge commit once that PR lands. * fix(desktop): refresh config when the providers panel mounts Model counts and the edit form's rows derive from the config models section, which another client may have rewritten since app launch — reload it alongside providers and models so edits never start from a stale snapshot. Exposes loadConfig on the client facade and bumps the kimi-code submodule for the comment-convention cleanup. * fix(desktop): re-probe auth readiness in the send gate before prompting login The cached authReady flag predates readiness changes that never re-ran checkAuth through this client — a provider added in settings, a default model bumped by a composer pick. The gate re-checks GET /auth on a stale negative before concluding sign-in is required, so a user with a configured provider and picked model can send without a login prompt. Also records the fresh-setup default-model seeding rule in the plan doc. * feat(desktop): add a custom-provider entry on the onboarding login step A second card below the Kimi sign-in option completes onboarding (same as skipping login) and deep-links into Settings -> Providers via a new initialTab prop on SettingsDialog, so first-run users with their own API key can configure a provider instead of signing in. |
||
|---|---|---|
| .. | ||
| providers-tab-proposals.html | ||