Backend
1. controller/channel.go
• Always hydrate `ChannelInfo` from DB in `UpdateChannel`, keeping `IsMultiKey` true so `MultiKeySize` is recalculated.
2. model/channel.go
• getKeys(): accept both newline-separated keys and JSON array (`[ {...}, {...} ]`).
• Update(): reuse new parser-logic to recalc `MultiKeySize`; prune stale indices in `MultiKeyStatusList`.
Frontend
1. pages/Channel/EditChannel.js
• `handleVertexUploadChange`
– Reset `vertexErroredNames` on every change so the “ignored files” prompt always re-appears.
– In single-key mode keep only the last file; in batch mode keep all valid files.
– Parse files, display “以下文件解析失败，已忽略：…”.
• Batch-toggle checkbox
– When switching from batch→single while multiple files are present, show a confirm dialog and retain only the first file (synchronises state, form and local caches).
• On opening the “new channel” side-sheet, clear `vertexErroredNames` to restore error prompts.
Result
• “已启用 x/x” count updates immediately after editing multi-key channels.
• Vertex-AI key upload works intuitively: proper error feedback, no duplicated files, and safe down-switch from batch to single mode.
Backend
• `model/channel.go`
– On multi-key channel updates, recalculate `MultiKeySize` from the current key list.
– If the request omits `key`, fetch existing keys from DB to avoid resetting the count to `0/0`.
– Remove out-of-range entries from `MultiKeyStatusList` to keep data consistent.
Frontend
• `web/src/pages/Channel/EditChannel.js`
– Vertex AI channels no longer require re-uploading a key file in edit mode.
– When no new key is provided during editing, exclude the `key` field from the payload to prevent overwriting the stored value.
These changes ensure that editing a channel no longer zeroes out the enabled key counter and that Vertex AI channels can be modified without mandatory key re-submission.
1. Recalculate `ChannelInfo.MultiKeySize` based on the current key list when a multi-key channel is updated.
2. Purge any indexes in `MultiKeyStatusList` that exceed the new key count to avoid stale or out-of-range data.
This ensures the front-end display (`enabled x/x`) always reflects the real number of keys after users add or remove keys in edit mode.
This commit refactors the application's error handling mechanism by introducing a new standardized error type, `types.NewAPIError`. It also renames common JSON utility functions for better clarity.
Previously, internal error handling was tightly coupled to the `dto.OpenAIError` format. This change decouples the internal logic from the external API representation.
Key changes:
- A new `types.NewAPIError` struct is introduced to serve as a canonical internal representation for all API errors.
- All relay adapters (OpenAI, Claude, Gemini, etc.) are updated to return `*types.NewAPIError`.
- Controllers now convert the internal `NewAPIError` to the client-facing `OpenAIError` format at the API boundary, ensuring backward compatibility.
- Channel auto-disable/enable logic is updated to use the new standardized error type.
- JSON utility functions are renamed to align with Go's standard library conventions (e.g., `UnmarshalJson` -> `Unmarshal`, `EncodeJson` -> `Marshal`).
Summary
1. **model/channel.go**
• Replaced the pointer-only `Value()` with a value-receiver implementation
• GORM can now marshal both `ChannelInfo` and `*ChannelInfo`, eliminating
`unsupported type model.ChannelInfo` runtime error.
2. **controller/channel.go**
• Refactored `getVertexArrayKeys()` – every element is now
- passed through `json.Marshal` when not already a string
- trimmed & validated before insertion
• Guarantees each service-account key is persisted as a **pure JSON string**
instead of the previous `map[...]` dump.
Result
• Channel creation / update succeeds without SQL driver errors.
• Vertex-AI batch & multi-key uploads are stored in canonical JSON, ready for
downstream SDKs to consume.
Summary
• Migrated all ratio-related sources into `setting/ratio_setting/`
– `model_ratio.go` (renamed from model-ratio.go)
– `cache_ratio.go`
– `group_ratio.go`
• Changed package name to `ratio_setting` and relocated initialization (`ratio_setting.InitRatioSettings()` in main).
• Updated every import & call site:
– Model / cache / completion / image ratio helpers
– Group ratio helpers (`GetGroupRatio*`, `ContainsGroupRatio`, `CheckGroupRatio`, etc.)
– JSON-serialization & update helpers (`*Ratio2JSONString`, `Update*RatioByJSONString`)
• Adjusted controllers, middleware, relay helpers, services and models to reference the new package.
• Removed obsolete `setting` / `operation_setting` imports; added missing `ratio_setting` imports.
• Adopted idiomatic map iteration (`for key := range m`) where value is unused.
• Ran static checks to ensure clean build.
This commit centralises all ratio configuration (model, cache and group) in one cohesive module, simplifying future maintenance and improving code clarity.
feat(api):
• Add optional `type` query param to `/api/channel` endpoint for type-specific pagination
• Return `type_counts` map with counts for each channel type
• Implement `GetChannelsByType`, `CountChannelsByType`, `CountChannelsGroupByType` in `model/channel.go`
feat(frontend):
• Introduce type Tabs in `ChannelsTable` to switch between channel types
• Tabs show dynamic counts using backend `type_counts`; “All” is computed from sum
• Persist active type, reload data on tab change (with proper query params)
perf(frontend):
• Use a request counter (`useRef`) to discard stale responses when tabs switch quickly
• Move all `useMemo` hooks to top level to satisfy React Hook rules
• Remove redundant local type counting fallback when backend data present
ui:
• Remove icons from response-time tags for cleaner look
• Use Semi-UI native arrow controls for Tabs; custom arrow code deleted
chore:
• Minor refactor & comments for clarity
• Ensure ESLint Hook rules pass
Result: Channel list now supports fast, accurate type filtering with correct counts, improved concurrency safety, and cleaner UI.
* backend
- model: add `Remark` field (varchar 255, `json:"remark,omitempty"`); AutoMigrate handles schema change automatically
- controller:
* accept `remark` on user create/update endpoints
* hide remark from regular users (`GetSelf`) by zero-ing the field before JSON marshalling
* clarify inline comment explaining the omitempty behaviour
* frontend (React / Semi UI)
- AddUser.js & EditUser.js: add “Remark” input for admins
- UsersTable.js:
* remove standalone “Remark” column
* show remark as a truncated Tag next to username with Tooltip for full text
* import Tooltip component
- i18n: reuse existing translations where applicable
This commit enables administrators to label users with private notes while ensuring those notes are never exposed to the users themselves.
Backend
- Introduce `setting/console_setting` package that defines `ConsoleSetting` struct with JSON tags and validation rules.
- Register the new module with `config.GlobalConfig` to enable automatic injection/export of configuration values.
- Remove legacy `setting/console.go` and the manual `OptionMap` hooks; clean up `model/option.go`.
- Add `controller/console_migrate.go` providing `/api/option/migrate_console_setting` endpoint for one-off data migration.
- Update controllers (`misc`, `option`, `uptime_kuma`) and router to consume namespaced keys `console_setting.*`.
Frontend
- Refactor dashboard pages (`SettingsAPIInfo`, `SettingsAnnouncements`, `SettingsFAQ`, `SettingsUptimeKuma`) and detail page to read/write the new keys.
- Simplify `DashboardSetting.js` state to only include namespaced options.
BREAKING CHANGE: All console-related option keys are now stored under `console_setting.*`. Run the migration endpoint once after deployment to preserve existing data.
Backend
• Introduced `validateExpiredTime` helper in `controller/redemption.go`; reused in both Add & Update endpoints to enforce “expiry time must not be earlier than now”, eliminating duplicated checks
• Removed `RedemptionCodeStatusExpired` constant and all related references – expiry is now determined exclusively by the `expired_time` field for simpler, safer state management
• Simplified `DeleteInvalidRedemptions`: deletes codes that are `used` / `disabled` or `enabled` but already expired, without relying on extra status codes
• Controller no longer mutates `status` when listing or fetching redemption codes; clients derive expiry status from timestamp
Frontend
• Added reusable `isExpired` helper in `RedemptionsTable.js`; leveraged for:
– status rendering (orange “Expired” tag)
– action-menu enable/disable logic
– row styling
• Removed duplicated inline expiry logic, improving readability and performance
• Adjusted toolbar layout: on small screens the “Clear invalid codes” button now wraps onto its own line, while “Add” & “Copy” remain grouped
Result
The codebase is now more maintainable, secure, and performant with no redundant constants, centralized validation, and cleaner UI behaviour across devices.
Front-end enhancements around “Add custom models”:
• EditChannel.js / EditTagModal.js
– Skip models that already exist instead of blocking the action.
– Collect actually inserted items and display:
• Success toast: “Added N models: model1, model2 …”
• Info toast when no new model detected.
– Keeps UX smooth while preserving deduplication logic.
• i18n
– en.json: added keys
• "已新增 {{count}} 个模型：{{list}}"
• "未发现新增模型"
– Fixed a broken JSON string containing smart quotes to maintain valid syntax.
Result:
Users can bulk-paste model names; duplicates are silently ignored and the UI clearly lists what was incrementally appended. All messages are fully internationalised.
Closes#1218
The previous patch lower-cased `group` and `model` when building the
temporary `abilitySet` used to prevent duplicate inserts.
This merged models that differ only by letter case, e.g.
`GPT-3.5-turbo` vs `gpt-3.5-turbo`, causing them to disappear from the
user’s available-models list and pricing page.
Change:
• ability.go – removed all `strings.ToLower` calls when composing the
deduplication key (`group|model`), so duplicates are checked in a
case-sensitive manner while preserving the original data.
Result:
• `GPT-3.5-turbo` and `gpt-3.5-turbo` are now treated as distinct
models throughout the system.
When importing large model lists (≈700+) an attempt to save a channel
could fail with:
Error 1062 (23000): Duplicate entry 'default-DeepSeek-1' for key 'abilities.PRIMARY'
Root cause: AddAbilities / UpdateAbilities inserted the same
(group, model) pair multiple times if the input list contained
duplicates or case-variants (e.g. `default` vs `Default`).
Changes:
• ability.go
– AddAbilities: introduced `abilitySet` to deduplicate by
lower-cased `group|model` key before batch-inserting.
– UpdateAbilities: applied the same deduplication logic when
rebuilding abilities inside a transaction.
Notes:
• The lower-casing is only for set comparison; the original
`group` and `model` values are preserved when persisting to DB,
so case sensitivity of stored data is unchanged.
• Batch chunking logic (lo.Chunk) and performance characteristics
remain unaffected.
Fixes#1215
- Add IP field to Log model with database index and default empty value
- Implement conditional IP recording based on user setting in RecordConsumeLog and RecordErrorLog
- Add UserSettingRecordIpLog constant and update user settings API to handle record_ip_log field
- Create dedicated "IP记录" tab in personal settings under "其他设置" section
- Add IP column to logs table with help tooltip explaining recording conditions
- Make IP column visible to all users (not admin-only) with proper filtering for consume/error log types
- Restrict display of use_time and retry columns to consume and error log types only
- Update personal settings UI structure: rename "通知设置" to "其他设置" to accommodate new functionality
- Add proper translation support and maintain consistent styling across components
The IP logging feature is disabled by default and only records client IP addresses
for consume (type 2) and error (type 5) logs when explicitly enabled by users
in their personal settings.
