fix(kimi): adopt live kimi-code token instead of OAuth refresh - #120
Open
yesme wants to merge 32 commits into
Open
fix(kimi): adopt live kimi-code token instead of OAuth refresh#120yesme wants to merge 32 commits into
yesme wants to merge 32 commits into
Conversation
Add Settings → Providers controls to drag-reorder header tabs and set custom display names (emoji allowed). Preferences persist in SQLite and apply on the next dashboard load. Co-authored-by: Grok 4.5 (high effort) <302482056+a-grok-build-bot[bot]@users.noreply.github.com>
Pin header actions so custom/long tab labels scroll inside the tab strip instead of squeezing out the refresh timestamp. Co-authored-by: Grok 4.5 (high effort) <302482056+a-grok-build-bot[bot]@users.noreply.github.com>
Register Grok (and Kimi) in the provider catalog so Settings can list and toggle them. When a saved tab order omits a newly available provider, insert it before the All / API Integrations tabs. Co-authored-by: Grok 4.5 (high effort) <302482056+a-grok-build-bot[bot]@users.noreply.github.com>
Catalog only lists providers implemented on this branch; Kimi stays in the separate Kimi PR. Default label map still knows "kimi" for forks that merge both. Co-authored-by: Grok 4.5 (high effort) <302482056+a-grok-build-bot[bot]@users.noreply.github.com>
Fork main integration for local runtime testing.
…order-and-labels) Fork main integration for local runtime testing. Keeps Kimi Code and Grok catalog entries alongside tab order/labels. Co-authored-by: Grok 4.5 (high effort) <302482056+a-grok-build-bot[bot]@users.noreply.github.com>
Co-authored-by: Grok 4.5 (high effort) <302482056+a-grok-build-bot[bot]@users.noreply.github.com>
Only list providers with Dashboard visibility on in the tab reorder list, and make the custom Tab name field obvious with a pencil label. Co-authored-by: Grok 4.5 (high effort) <302482056+a-grok-build-bot[bot]@users.noreply.github.com>
Co-authored-by: Grok 4.5 (high effort) <302482056+a-grok-build-bot[bot]@users.noreply.github.com>
Co-authored-by: Grok 4.5 (high effort) <302482056+a-grok-build-bot[bot]@users.noreply.github.com>
… 401 kimi-code access tokens last ~15 minutes. Auto-detect was copying the access token into NewKimiClient as staticToken, which skipped disk re-read and OAuth refresh for the whole daemon lifetime — after expiry every poll returned unauthorized even while ~/.kimi-code credentials stayed valid. - Pass empty token for KimiAutoToken so the client uses the kimi-code store - On usages 401, re-read disk then force-refresh (access can be dead before expires_at); still never touch kimi-cli paths - Regression test for unexpired-but-401 force refresh
Anthropic /api/oauth/usage now returns top-level companion fields such as limits (array), spend (unrelated object), and booleans alongside quota buckets. Decoding the whole payload as map[string]*AnthropicQuotaEntry failed with: cannot unmarshal array into Go value of type api.AnthropicQuotaEntry Add a custom UnmarshalJSON that keeps null/object quota entries and ignores non-quota metadata. Enhances onllm-dev#84 by also dropping non-quota objects (spend) and skipping per-key decode errors so one bad companion field cannot take down the entire Anthropic poll. Fixes onllm-dev#82. Related: onllm-dev#84.
Grok and Kimi fell through to the OpenAI icon class in the menubar popover (providerIconClass default). Grok's static asset was also a placeholder plus-circle, and Kimi had no asset at all. Replace grok.svg with the brand monochrome mark, add kimi.svg, and wire both into menubar CSS masks + providerIconClass.
Tray title used selected_quotas click order, so configuring provider list as Codex → Anthropic → Grok still showed 17%·15%·7% (Codex·Grok·Anthropic). Re-sort multi_provider selected_quotas by providers_order during Settings.Normalize so tray title matches the menubar list order. Co-authored-by: Grok 4.5 (high effort) <302482056+a-grok-build-bot[bot]@users.noreply.github.com>
Web Settings > Menubar saved providers_order/visible_providers as "codex"
for single-account setups, while menubar cards always use account-scoped
IDs ("codex:1"). Exact-match filtering then dropped Codex after reorder/save.
- Match bare base keys to account-scoped cards in filter/sort/options
- Always emit codex:<id> from the settings provider list when profiles exist
- Tray selected_quotas ordering also accepts bare "codex" in providers_order
Co-authored-by: Grok 4.5 (high effort) <302482056+a-grok-build-bot[bot]@users.noreply.github.com>
Grok and Kimi current payloads only set snake_case resets_at and never computed timeUntilReset. Menubar normalizeQuotas only reads camelCase keys (resetsAt/renewsAt + timeUntilReset), so reset countdown lines were empty for both providers. Align with Anthropic/Codex/Cursor: set resetsAt, timeUntilReset, and timeUntilResetSeconds (keep resets_at for dashboard JS compatibility). Co-authored-by: Grok 4.5 (high effort) <302482056+a-grok-build-bot[bot]@users.noreply.github.com>
Menubar hard-coded labels like "Kimi Code" / "Codex - default" ignored dashboard_provider_labels (Settings tab rename). Load the same label map and resolveProviderTabLabel for every menubar card; multi-account Codex/MiniMax only append " - <account>" when more than one account. Also show renamed titles in the web Menubar provider-order list. Co-authored-by: Grok 4.5 (high effort) <302482056+a-grok-build-bot[bot]@users.noreply.github.com>
Add a monochrome brand mark left of each provider tab label using CSS mask-image so the icon follows tab text color (active/hover). Includes OpenRouter, Cursor, API Integrations, and All icons alongside existing provider SVGs. Co-authored-by: Grok 4.5 (high effort) <302482056+a-grok-build-bot[bot]@users.noreply.github.com>
CSS mask-image icons disappeared when browsers kept an immutable style.css?v=… cache from before the icon rules existed. Use real <img> tags with absolute /static/icons paths so brand marks always paint next to tab labels. Co-authored-by: Grok 4.5 (high effort) <302482056+a-grok-build-bot[bot]@users.noreply.github.com>
Paint provider tab icons with CSS mask + currentColor so they track tab text color across light/dark tokens, hover, and active accents. Use absolute icon URLs on each tab via --provider-tab-icon; SVG fills are solid black for reliable mask alpha. Co-authored-by: Grok 4.5 (high effort) <302482056+a-grok-build-bot[bot]@users.noreply.github.com>
Replace verbose \"Last updated: HH:MM:SS Asia/Shanghai\" with \"Updated HH:MM:SS UTC+8\". Full IANA zone stays in the tooltip. Reset labels use the same short offset for consistency. Co-authored-by: Grok 4.5 (high effort) <302482056+a-grok-build-bot[bot]@users.noreply.github.com>
Opening the tray popover reloaded /menubar every time, so WKWebView showed a blank shell then re-painted. Preload the page at menubar start, skip full navigation when the URL is already loaded, and soft- refresh snapshot data via JS when reopening. Co-authored-by: Grok 4.5 (high effort) <302482056+a-grok-build-bot[bot]@users.noreply.github.com>
Allow users to disable OAuth refresh-and-write of coding-harness credentials so onWatch does not compete with Claude/Kimi/Codex CLIs for one-time refresh tokens. Defaults to enabled for existing behavior. Co-authored-by: Grok 4.5 (high effort) <302482056+a-grok-build-bot[bot]@users.noreply.github.com>
Add a GNOME Shell extension that reuses the macOS /menubar Web UI for Linux: top-bar multi-provider % with monochrome icons, hover/click WebKitGTK panel host, panel-aware icon tinting, system dark passthrough for WebView, and a native close bridge so X no longer navigates to dashboard/login. Co-authored-by: Grok 4.5 (high effort) <302482056+a-grok-build-bot[bot]@users.noreply.github.com>
Safari often fails CSS mask-image when the URL is supplied via a custom property, so provider tabs rendered without glyphs. Switch to real <img> tags (always paint) and invert monochrome SVGs in dark theme. Normalize icon SVGs to 24x24 black fills for reliable sizing. Co-authored-by: Grok 4.5 (high effort) <302482056+a-grok-build-bot[bot]@users.noreply.github.com>
Dashboard CLI strip preferred shorter windows first, since they are usually exhausted sooner: 1. Gemini 5h 2. Gemini Weekly 3. Claude + GPT 5h 4. Claude + GPT Weekly Co-authored-by: Grok 4.5 (high effort) <302482056+a-grok-build-bot[bot]@users.noreply.github.com>
IDE GetUserStatus and RetrieveUserQuotaSummary both treat Gemini Pro and Flash as one shared quota. Splitting them into two cards showed identical usage and looked like a bug. Group all Gemini models under a single "Gemini Quota" pool; keep legacy pro/flash keys as aliases. Co-authored-by: Grok <[email protected]>
fix(antigravity): merge Pro and Flash into shared Gemini pool
A running kimi-code process owns the refresh-token chain. Calling POST /api/oauth/token from onWatch rotates that token and invalidates the live CLI session. When kimi-code is detected, skip OAuth, re-read kimi-code.json (including mtime/size changes), and use the access token the CLI already wrote. OAuth refresh stays for the idle-CLI path when Auto refresh tokens is enabled. Co-authored-by: Grok 4.6 <281844019+a-claude-code-bot[bot]@users.noreply.github.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
When Kimi Code is running, onWatch was still calling
POST /api/oauth/token(or skipping refresh entirely if Auto refresh tokens was off). Refresh-token rotation invalidates the live CLI session, and an idle/expired disk token then 401s forever even though the CLI is alive.If a
kimi-codeprocess is detected, onWatch now adopts the access token the CLI already wrote to~/.kimi-code/credentials/kimi-code.jsonand does not OAuth-refresh. Credential cache also invalidates when the file mtime/size changes, so a CLI persist is picked up on the next poll.OAuth refresh is unchanged when kimi-code is not running and Settings → Auto refresh tokens is on.
Changes
internal/api/kimi_process.go: detect a livekimi-codeprocess (same idea as skipping Anthropic refresh while Claude Code is running)internal/api/kimi_client.go: while the CLI is live, use the disk access token; re-read on 401; never rotate the refresh tokeninternal/api/kimi_credentials.go: reload credentials when the file mtime/size changes instead of waiting out the 30s cachedocs/KIMI_SETUP.md: document the live-CLI vs idle-CLI refresh policyTest plan
go test -race ./internal/api/ -run 'TestKimi|TestLoadKimi|TestIsKimi|TestDetectKimi'kimi-code.json, with noPOST /api/oauth/tokenlive kimi-code owns the refresh token/adopting access token)