@spicyapi/sdk 0.4.0 → 0.5.0

package/contracts/openapi.yaml

@@ -13,6 +13,17 @@ info:
     The `/v1` operations tagged `OpenAI Compatible` are the one documented
     exception: they return OpenAI / Anthropic response and error objects so
     the official SDKs work by changing only `base_url`.
+
+    Error text language: `msg`, task `errorMessage` and the compatible
+    layers' `error.message` are English by default. Send `Accept-Language`
+    (`en`, `zh-Hant`, `ja`, `ko`, `de`, `fr`, `es`, `pt-BR`, `ru`; regional
+    variants such as `de-DE` or `zh-TW` match their language, `zh-CN` and `*`
+    do not) to choose another language for one request, or set the account's
+    API error language in the console; a recognized header takes precedence.
+    Webhooks follow the account setting. Error responses carry
+    `Content-Language`. Codes (`code`, `errorCode`, OpenAI `type` / `code`,
+    Gemini `code` / `status`) never change with the language, so branch on
+    codes, not on text.
 servers:
   - url: https://api.spicyapi.ai
     description: Production
@@ -22,7 +33,13 @@ tags:
   - name: Account
   - name: Media
   - name: OpenAI Compatible
-    description: OpenAI / Anthropic protocol shells under `/v1`. They share authentication, limits, validation and billing with `/api/v1` but use the external protocol's own response and error shapes.
+    description: >-
+      OpenAI / Anthropic protocol shells under `/v1` and the Google Gemini shell under `/v1beta`.
+      They share API keys, limits, validation and billing with `/api/v1` but use the external
+      protocol's own request, response and error shapes. On these paths the API key may be sent as
+      `Authorization: Bearer <key>`, `x-api-key: <key>` (Anthropic SDKs) or
+      `x-goog-api-key: <key>` (Google GenAI SDKs); `/api/v1` accepts only `Authorization: Bearer`.
+      Keys in the URL query string are never accepted.
 security:
   - bearerAuth: []
 paths:
@@ -81,6 +98,7 @@ paths:
         private/loopback destinations before acceptance.
       parameters:
         - $ref: '#/components/parameters/IdempotencyKey'
+        - $ref: '#/components/parameters/SpicyRetention'
       requestBody:
         required: true
         content:
@@ -202,7 +220,7 @@ paths:
         as a Server-Sent Events stream. Each event is a `data:` line carrying one
         JSON chunk; the stream ends with `data: [DONE]`.
 
-        The hold is settled when the stream completes. If the upstream call fails
+        The hold is settled when the stream completes. If generation fails
         before the first event, the response is a normal JSON error and the hold is
         released. If it fails mid-stream the connection is terminated with an
         `error` event and the hold is still released — a truncated answer is never
@@ -213,6 +231,7 @@ paths:
         returns `503`.
       parameters:
         - $ref: '#/components/parameters/IdempotencyKey'
+        - $ref: '#/components/parameters/SpicyRetention'
       requestBody:
         required: true
         content:
@@ -242,7 +261,7 @@ paths:
       summary: Create a new task from a failed or expired task
       description: |
         The source terminal task is never modified. Current model schema,
-        pricing, permissions, balance, and routing are evaluated again. An
+        pricing, permissions, balance, and availability are evaluated again. An
         optional Idempotency-Key is scoped to the source task and retry action.
         The source task must have been created by the same API key; unknown,
         cross-account, and same-account cross-key task IDs all return 404.
@@ -269,6 +288,57 @@ paths:
         '429': { $ref: '#/components/responses/RateLimited' }
         '500': { $ref: '#/components/responses/ServerError' }
         '503': { $ref: '#/components/responses/Unavailable' }
+  /api/v1/jobs/purge:
+    post:
+      tags: [Tasks]
+      operationId: purgeTask
+      summary: Destroy the stored content of a finished task
+      description: |
+        Removes the task's generated media objects, its result payload, and the
+        stored request text including the prompt.
+
+        **Billing evidence is never touched.** The ledger, the charged amount,
+        the model identifier, the state, the timestamps and the request_id all
+        remain, so the task stays fully auditable in billing. What is destroyed
+        is the content, not the spend record; the response repeats this as
+        `billingRetained`.
+
+        Only a finished task can be destroyed; a task that is still queued or
+        running returns 400. Its outputs have not landed yet, and removing half
+        of them is worse than removing none. An accepted generation task cannot
+        be stopped, so the only thing to do is wait for it to reach a terminal
+        state and call this endpoint then.
+
+        The call is idempotent by `taskId`: destroying an already destroyed task
+        returns 200 with the original `purgedAt`, so a client may safely retry
+        after a timeout. No `Idempotency-Key` is needed.
+
+        Media objects are removed by a background sweep within about a minute,
+        so `mediaDeletionPending` can be true while `contentState` is already
+        `purged`. Deleting them inline would let one undeletable object block the
+        whole request.
+      parameters:
+        - $ref: '#/components/parameters/IdempotencyKey'
+      requestBody:
+        required: true
+        content:
+          application/json:
+            schema: { $ref: '#/components/schemas/TaskActionRequest' }
+      responses:
+        '200':
+          description: Content destroyed, or already destroyed by an earlier call.
+          headers:
+            Cache-Control:
+              schema: { type: string, const: no-store }
+          content:
+            application/json:
+              schema: { $ref: '#/components/schemas/TaskPurgeEnvelope' }
+        '400': { $ref: '#/components/responses/BadRequest' }
+        '401': { $ref: '#/components/responses/Unauthorized' }
+        '403': { $ref: '#/components/responses/Forbidden' }
+        '404': { $ref: '#/components/responses/NotFound' }
+        '429': { $ref: '#/components/responses/RateLimited' }
+        '500': { $ref: '#/components/responses/ServerError' }
   /api/v1/models:
     get:
       tags: [Models]
@@ -285,7 +355,7 @@ paths:
         - name: provider
           in: query
           schema: { type: string }
-          description: Exact model creator/author identifier. This never identifies an execution channel.
+          description: Exact model creator/author identifier, such as the research lab that built the model.
         - name: task
           in: query
           schema: { type: string }
@@ -362,7 +432,7 @@ paths:
         Returns only visible tasks created by this API key and its account,
         newest first by createdAt and taskId. The list contains metadata only;
         use recordInfo for a selected task's result. No input, output, signed
-        media URL, or execution-channel information is included.
+        media URL, or internal processing detail is included.
         Dates form a UTC half-open interval [from,to), up to 92 days.
         Default to is tomorrow in UTC; omitted from is seven days before to.
         Keep all filters, including explicit from/to dates, unchanged while
@@ -478,7 +548,7 @@ paths:
         The server generates the object key. Upload the exact declared number of
         bytes with PUT and copy every returned header. Supported media types are
         JPEG, PNG, WebP and GIF images (10 MiB), or MP4/WebM video and
-        MP3/WAV audio (100 MiB, at most 600 seconds). Audio/video commit verifies
+        MP3/WAV audio (90 MiB, at most 600 seconds). Audio/video commit verifies
         format and measurable duration; it does not inspect adult content. After PUT,
         call the file commit endpoint and use its spicy:// URI in task input.
       requestBody:
@@ -572,8 +642,8 @@ paths:
       description: |
         Returns every model that can be called right now. `id` is the SpicyAPI
         model identifier accepted by every `/v1` and `/api/v1` operation;
-        `owned_by` is the model publisher (for example a research lab), never
-        an execution supplier. Pricing is not included; use `/api/v1/models`.
+        `owned_by` is the model publisher (for example a research lab).
+        Pricing is not included; use `/api/v1/models`.
       responses:
         '200':
           description: OpenAI-style model list.
@@ -723,6 +793,102 @@ paths:
         '500': { $ref: '#/components/responses/AnthropicError' }
         '503': { $ref: '#/components/responses/AnthropicError' }
 
+  /v1beta/models/{model}:generateContent:
+    post:
+      tags: [OpenAI Compatible]
+      operationId: geminiGenerateContent
+      summary: Google Gemini generateContent on a SpicyAPI text model
+      description: |
+        Accepts a Google Gemini `generateContent` request (`contents` with
+        text / inlineData / functionCall / functionResponse parts,
+        `systemInstruction`, `tools.functionDeclarations`, `toolConfig`,
+        `generationConfig` such as `temperature`, `topP`, `maxOutputTokens`,
+        `stopSequences`, `responseMimeType`, `responseSchema`,
+        `thinkingConfig`). `{model}` is a SpicyAPI text model identifier from
+        `/v1/models` and may contain a slash (for example
+        `google/gemini-3-flash-preview`). Validation, pricing and availability are
+        identical to `/v1/chat/completions`. The response is one
+        `GenerateContentResponse`. Errors use the Google
+        `{error:{code,message,status}}` shape.
+      parameters:
+        - name: model
+          in: path
+          required: true
+          schema: { type: string }
+        - $ref: '#/components/parameters/IdempotencyKey'
+      requestBody:
+        required: true
+        content:
+          application/json:
+            schema:
+              $ref: '#/components/schemas/GeminiGenerateContentRequest'
+      responses:
+        '200':
+          description: One `GenerateContentResponse`.
+          content:
+            application/json:
+              schema:
+                $ref: '#/components/schemas/GeminiGenerateContentResponse'
+        '400': { $ref: '#/components/responses/GeminiError' }
+        '401': { $ref: '#/components/responses/GeminiError' }
+        '402': { $ref: '#/components/responses/GeminiError' }
+        '403': { $ref: '#/components/responses/GeminiError' }
+        '404': { $ref: '#/components/responses/GeminiError' }
+        '409': { $ref: '#/components/responses/GeminiError' }
+        '413': { $ref: '#/components/responses/GeminiError' }
+        '429': { $ref: '#/components/responses/GeminiError' }
+        '500': { $ref: '#/components/responses/GeminiError' }
+        '503': { $ref: '#/components/responses/GeminiError' }
+
+  /v1beta/models/{model}:streamGenerateContent:
+    post:
+      tags: [OpenAI Compatible]
+      operationId: geminiStreamGenerateContent
+      summary: Google Gemini streamGenerateContent on a SpicyAPI text model
+      description: |
+        Same request as `generateContent`. With `?alt=sse` (what the Google
+        GenAI SDKs send) the response is a server-sent event stream where each
+        `data:` line is one `GenerateContentResponse`; the final one carries
+        `finishReason` and `usageMetadata`. There is no `[DONE]` sentinel.
+        Without `alt=sse` the response is a JSON array of
+        `GenerateContentResponse` objects written progressively.
+      parameters:
+        - name: model
+          in: path
+          required: true
+          schema: { type: string }
+        - name: alt
+          in: query
+          required: false
+          schema: { type: string, enum: [sse] }
+        - $ref: '#/components/parameters/IdempotencyKey'
+      requestBody:
+        required: true
+        content:
+          application/json:
+            schema:
+              $ref: '#/components/schemas/GeminiGenerateContentRequest'
+      responses:
+        '200':
+          description: An SSE stream (`alt=sse`) or a JSON array of `GenerateContentResponse`.
+          content:
+            text/event-stream:
+              schema: { type: string }
+            application/json:
+              schema:
+                type: array
+                items: { $ref: '#/components/schemas/GeminiGenerateContentResponse' }
+        '400': { $ref: '#/components/responses/GeminiError' }
+        '401': { $ref: '#/components/responses/GeminiError' }
+        '402': { $ref: '#/components/responses/GeminiError' }
+        '403': { $ref: '#/components/responses/GeminiError' }
+        '404': { $ref: '#/components/responses/GeminiError' }
+        '409': { $ref: '#/components/responses/GeminiError' }
+        '413': { $ref: '#/components/responses/GeminiError' }
+        '429': { $ref: '#/components/responses/GeminiError' }
+        '500': { $ref: '#/components/responses/GeminiError' }
+        '503': { $ref: '#/components/responses/GeminiError' }
+
   /v1/videos:
     post:
       tags: [OpenAI Compatible]
@@ -839,6 +1005,28 @@ components:
       required: false
       description: Stable key for one logical task submission; retained for 24 hours per account and bound to both the submitting API key and normalized request fingerprint. A sibling key receives 409 without the original task ID.
       schema: { type: string, minLength: 1, maxLength: 128 }
+    SpicyRetention:
+      name: X-Spicy-Retention
+      in: header
+      required: false
+      description: >-
+        Retention for this one task, in whole seconds. It applies to both the
+        generated outputs and the stored request text. The effective value is
+        always `min(header, account setting, platform maximum)`: a request can
+        only shorten retention, never extend it.
+
+        `0` is valid and means the outputs are removed as soon as the task
+        reaches a terminal state. Values that cannot be applied never fail the
+        request: a negative number is clamped to `0`, a value above the ceiling
+        is clamped to the ceiling, and a value that is not a decimal integer is
+        ignored so the account setting applies. The effective result is always
+        reported back in `retention` on the task record.
+
+        Accepted by `createTask` and `stream`. `retry` inherits the source
+        task's effective value and ignores this header, so one task family never
+        ends up with two different expiry times. `quote` does not accept it
+        because it does not create a task.
+      schema: { type: string, pattern: '^-?[0-9]+$' }
   headers:
     RateLimitLimit:
       description: Capacity of the active account-level rate-limit window.
@@ -893,6 +1081,9 @@ components:
     AnthropicError:
       description: Error in the Anthropic `{type:"error",error:{type,message}}` shape.
       content: { application/json: { schema: { $ref: '#/components/schemas/AnthropicErrorBody' } } }
+    GeminiError:
+      description: Error in the Google `{error:{code,message,status}}` shape. `status` is the google.rpc code name (for example `INVALID_ARGUMENT`, `UNAUTHENTICATED`, `RESOURCE_EXHAUSTED`).
+      content: { application/json: { schema: { $ref: '#/components/schemas/GeminiErrorBody' } } }
   schemas:
     FundingGrantItem:
       type: object
@@ -952,7 +1143,9 @@ components:
             code:
               type: integer
               enum: [400, 401, 40201, 40202, 403, 40301, 40302, 40303, 404, 409, 40901, 413, 429, 500, 50301]
-            msg: { type: string }
+            msg:
+              type: string
+              description: Human-readable explanation, English unless `Accept-Language` or the account's API error language selects another supported language. Never parse it; branch on `code`.
     CreateTaskRequest:
       type: object
       required: [model, input]
@@ -1047,6 +1240,46 @@ components:
             code: { type: integer, const: 200 }
             msg: { type: string, const: success }
             data: { $ref: '#/components/schemas/RetryTaskResponse' }
+    TaskPurgeResponse:
+      type: object
+      required: [taskId, contentState, billingRetained, mediaDeletionPending]
+      properties:
+        taskId: { type: string }
+        contentState:
+          type: string
+          enum: [present, expired, purged]
+          description: >-
+            `purged` after a successful destroy, and on any idempotent repeat.
+            Still `present` means this call removed nothing and may be retried
+            later; the reason is deliberately not disclosed.
+        purgedAt:
+          type: string
+          format: date-time
+          description: Time of the **first** destroy, not of this call.
+        contentRemovedBy:
+          type: string
+          enum: [user, system]
+          description: Omitted while `contentState` is `present`.
+        billingRetained:
+          type: boolean
+          const: true
+          description: >-
+            Always true. Destroying content never changes the ledger, the
+            charged amount, or any other billing fact. It is stated explicitly
+            because it is the one thing a caller is likely to get wrong about
+            an endpoint named "purge".
+        mediaDeletionPending:
+          type: boolean
+          description: True while the background sweep has not finished removing the stored media objects.
+    TaskPurgeEnvelope:
+      allOf:
+        - $ref: '#/components/schemas/EnvelopeBase'
+        - type: object
+          required: [data]
+          properties:
+            code: { type: integer, const: 200 }
+            msg: { type: string, const: success }
+            data: { $ref: '#/components/schemas/TaskPurgeResponse' }
     CreateTaskEnvelope:
       allOf:
         - $ref: '#/components/schemas/EnvelopeBase'
@@ -1096,8 +1329,23 @@ components:
           type: object
           additionalProperties: true
         output: { $ref: '#/components/schemas/TaskOutput' }
-        errorCode: { type: string, description: Stable SpicyAPI failure identifier; never a raw execution-channel code. }
-        errorMessage: { type: string, description: Safe normalized explanation without internal service names, hosts, task IDs, or raw errors. }
+        errorCode:
+          type: string
+          description: >-
+            Stable SpicyAPI failure identifier from a closed set; never an internal diagnostic code.
+            The set is closed: any other value a client sees should be handled as
+            `upstream_failed`.
+          enum:
+            - invalid_request
+            - unsupported_combination
+            - content_rejected
+            - rate_limited
+            - upstream_unavailable
+            - generation_failed
+            - timeout
+            - invalid_asset
+            - upstream_failed
+        errorMessage: { type: string, description: 'Safe normalized explanation without internal service names, hosts, task IDs, or raw errors. Follows the same language selection as `msg`; `errorCode` never changes.' }
         cost: { $ref: '#/components/schemas/USDString' }
         settled:
           type: boolean
@@ -1105,6 +1353,52 @@ components:
         createdAt: { type: string, format: date-time }
         deadlineAt: { type: string, format: date-time, description: Server execution deadline. Present in current responses; historical stored webhook events may omit it. Not the result retention or URL expiry time. }
         completedAt: { type: string, format: date-time }
+        contentState:
+          type: string
+          enum: [present, expired, purged]
+          description: >-
+            Whether the task content is still stored, and why it is not.
+            `expired` means the retention policy removed it; `purged` means the
+            account owner destroyed it with `purgeTask`. The two are deliberately
+            distinct: telling a customer their own deletion was an expiry reads
+            as the platform losing their data.
+        contentRemovedBy:
+          type: string
+          enum: [user, system]
+          description: Present only once content has been removed. Omitted while `contentState` is `present`.
+        purgedAt:
+          type: string
+          format: date-time
+          description: Present only for a task destroyed with `purgeTask`. Expiry does not set it.
+        retention: { $ref: '#/components/schemas/TaskRetention' }
+    TaskRetention:
+      type: object
+      description: >-
+        When this task's content is scheduled to be removed, and which layer
+        decided it. Omitted for tasks created before per-task retention existed;
+        those follow the platform maximum.
+      required: [outputsExpireAt, promptsExpireAt, source]
+      properties:
+        outputsExpireAt:
+          type: string
+          format: date-time
+          description: >-
+            When the generated media objects and the result payload are removed.
+            Before the task finishes this is measured from acceptance; when the
+            task reaches a terminal state it is re-based on the completion time
+            using the same window, so it only ever moves later, never earlier.
+        promptsExpireAt:
+          type: string
+          format: date-time
+          description: When the stored request text is erased. The task row, its state, amounts and request_id remain.
+        source:
+          type: string
+          enum: [header, account, platform]
+          description: >-
+            Which layer actually constrained the value: the per-request
+            `X-Spicy-Retention` header, the account setting, or the platform
+            maximum. A request that sends the header and still sees `account`
+            has an account setting that is already shorter than it asked for.
     TaskRecordEnvelope:
       allOf:
         - $ref: '#/components/schemas/EnvelopeBase'
@@ -1251,17 +1545,17 @@ components:
         displayName: { type: string }
         provider:
           type: string
-          description: Model creator/author identifier. It never identifies an execution channel.
+          description: Model creator/author identifier, such as the research lab that built the model.
         modality: { type: string, enum: [image, video, audio, text] }
         tasks: { type: array, items: { type: string } }
         async: { type: boolean }
         mature:
           type: boolean
-          description: Informational model-creator capability metadata only. It does not participate in routing, authorization, or request rejection.
+          description: Informational model-creator capability metadata only. It does not participate in authorization, availability decisions, or request rejection.
         policyTier:
           type: string
           enum: [unrestricted, borderline, unspecified]
-          description: Informational model-creator policy metadata only. It does not participate in routing, authorization, or request rejection.
+          description: Informational model-creator policy metadata only. It does not participate in authorization, availability decisions, or request rejection.
         taskTimeoutSeconds:
           type: integer
           minimum: 1
@@ -1351,7 +1645,7 @@ components:
       additionalProperties: false
       properties:
         contentType: { type: string, enum: [image/jpeg, image/png, image/webp, image/gif, video/mp4, video/webm, audio/mpeg, audio/wav] }
-        bytes: { type: integer, minimum: 1, maximum: 104857600, description: Images are limited to 10 MiB; supported audio/video to 100 MiB. }
+        bytes: { type: integer, minimum: 1, maximum: 94371840, description: Images are limited to 10 MiB; supported audio/video to 90 MiB. }
     UploadURLResponse:
       type: object
       required: [fileId, key, uploadUrl, method, headers, expiresAt, maxBytes]
@@ -1605,6 +1899,71 @@ components:
             input_tokens: { type: integer }
             output_tokens: { type: integer }
             total_tokens: { type: integer }
+    GeminiGenerateContentRequest:
+      type: object
+      required: [contents]
+      description: Google Gemini `generateContent` request. Which generation parameters a model accepts comes from that model's `inputSchema` in `/v1/models` and the catalog; unsupported parameters return 400.
+      properties:
+        contents:
+          type: array
+          items:
+            type: object
+            properties:
+              role: { type: string, enum: [user, model] }
+              parts:
+                type: array
+                items:
+                  type: object
+                  description: One of `text`, `inlineData {mimeType,data}`, `fileData {mimeType,fileUri}`, `functionCall {name,args}`, `functionResponse {name,response}`.
+        systemInstruction: { type: object }
+        tools: { type: array, items: { type: object } }
+        toolConfig: { type: object }
+        generationConfig:
+          type: object
+          properties:
+            temperature: { type: number }
+            topP: { type: number }
+            maxOutputTokens: { type: integer }
+            stopSequences: { type: array, items: { type: string } }
+            responseMimeType: { type: string }
+            responseSchema: { type: object }
+            thinkingConfig: { type: object }
+    GeminiGenerateContentResponse:
+      type: object
+      properties:
+        candidates:
+          type: array
+          items:
+            type: object
+            properties:
+              index: { type: integer }
+              content:
+                type: object
+                properties:
+                  role: { type: string }
+                  parts: { type: array, items: { type: object } }
+              finishReason: { type: string }
+        usageMetadata:
+          type: object
+          properties:
+            promptTokenCount: { type: integer }
+            candidatesTokenCount: { type: integer }
+            totalTokenCount: { type: integer }
+            cachedContentTokenCount: { type: integer }
+            thoughtsTokenCount: { type: integer }
+        modelVersion: { type: string, description: The SpicyAPI model identifier. }
+        responseId: { type: string }
+    GeminiErrorBody:
+      type: object
+      required: [error]
+      properties:
+        error:
+          type: object
+          required: [code, message, status]
+          properties:
+            code: { type: integer }
+            message: { type: string }
+            status: { type: string }
     AnthropicMessageRequest:
       type: object
       required: [model, messages, max_tokens]

package/dist/src/docs/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/docs/index.ts"],"names":[],"mappings":"AAAA,MAAM,WAAW,kBAAkB;IACjC,IAAI,EAAE,MAAM,CAAC;IACb,KAAK,EAAE,MAAM,CAAC;IACd,OAAO,EAAE,MAAM,CAAC;IAChB,GAAG,EAAE,MAAM,CAAC;IACZ,QAAQ,EAAE,MAAM,EAAE,CAAC;CACpB;AAiID,eAAO,MAAM,aAAa,EAAE,kBAAkB,EAG3C,CAAC;AA6CJ,wBAAgB,mBAAmB,CAAC,KAAK,SAAK,EAAE,KAAK,SAAK,GAAG,kBAAkB,EAAE,CAoChF"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/docs/index.ts"],"names":[],"mappings":"AAAA,MAAM,WAAW,kBAAkB;IACjC,IAAI,EAAE,MAAM,CAAC;IACb,KAAK,EAAE,MAAM,CAAC;IACd,OAAO,EAAE,MAAM,CAAC;IAChB,GAAG,EAAE,MAAM,CAAC;IACZ,QAAQ,EAAE,MAAM,EAAE,CAAC;CACpB;AA4HD,eAAO,MAAM,aAAa,EAAE,kBAAkB,EAG3C,CAAC;AA6CJ,wBAAgB,mBAAmB,CAAC,KAAK,SAAK,EAAE,KAAK,SAAK,GAAG,kBAAkB,EAAE,CAoChF"}

package/dist/src/docs/index.js

@@ -4,55 +4,55 @@ const definitions = [
         slug: "",
         title: "Quickstart",
         summary: "Create a task, follow its state, and retrieve generated media.",
-        keywords: ["start", "guide", "quickstart", "入门", "开始"],
+        keywords: ["start", "guide", "quickstart", "introduction", "tutorial"],
     },
     {
         slug: "authentication",
         title: "Authentication",
         summary: "API keys, per-key controls, rotation, and incident response.",
-        keywords: ["api key", "bearer", "security", "认证", "密钥"],
+        keywords: ["api key", "bearer", "security"],
     },
     {
         slug: "models",
         title: "Models and parameters",
         summary: "Live model IDs, account-specific pricing, input schemas, and examples.",
-        keywords: ["catalog", "schema", "price", "model", "模型", "价格", "参数"],
+        keywords: ["catalog", "schema", "price", "model"],
     },
     {
         slug: "tasks",
         title: "Asynchronous tasks",
         summary: "Create tasks, inspect task records, states, costs, and outputs.",
-        keywords: ["createTask", "recordInfo", "poll", "state", "任务", "轮询"],
+        keywords: ["createTask", "recordInfo", "poll", "polling", "state"],
     },
     {
         slug: "task-actions",
         title: "Task actions",
         summary: "Retry failed or expired tasks safely and understand unsupported actions.",
-        keywords: ["retry", "cancel", "重试", "取消"],
+        keywords: ["retry", "cancel"],
     },
     {
         slug: "media",
         title: "Media transfer",
         summary: "Public HTTPS media inputs, optional local uploads, and direct signed result URLs.",
-        keywords: ["upload", "download", "file", "image", "上传", "下载", "文件"],
+        keywords: ["upload", "download", "file", "image"],
     },
     {
         slug: "billing",
         title: "Billing",
         summary: "Prices, holds, settlement, failures, refunds, and balance fields.",
-        keywords: ["cost", "credit", "balance", "payment", "billing", "余额", "付费", "计费"],
+        keywords: ["cost", "credit", "balance", "payment", "billing"],
     },
     {
         slug: "idempotency",
         title: "Idempotency",
         summary: "Retry one logical generation without duplicate work or duplicate holds.",
-        keywords: ["idempotency-key", "retry", "duplicate", "幂等", "重复"],
+        keywords: ["idempotency-key", "retry", "duplicate"],
     },
     {
         slug: "webhooks",
         title: "Webhooks",
         summary: "Payload versions, raw-body signatures, delivery IDs, and retries.",
-        keywords: ["callback", "hmac", "signature", "webhook", "回调", "签名"],
+        keywords: ["callback", "hmac", "signature", "webhook"],
     },
     {
         slug: "errors",
@@ -67,34 +67,31 @@ const definitions = [
             "service unavailable",
             "no route",
             "customer price",
-            "错误",
-            "不可用",
-            "重试",
         ],
     },
     {
         slug: "limits",
         title: "Rate limits",
         summary: "Shared account rate limits and recommended backoff behavior.",
-        keywords: ["rate", "quota", "retry-after", "限流", "配额"],
+        keywords: ["rate", "quota", "retry-after", "throttling"],
     },
     {
         slug: "retention",
         title: "Data retention",
-        summary: "Task input, output, media, and operational-record retention boundaries.",
-        keywords: ["privacy", "delete", "storage", "retention", "隐私", "删除", "保留"],
+        summary: "Retention boundaries, per-request retention, and destroying one task's stored content.",
+        keywords: ["privacy", "delete", "storage", "retention", "purge", "destroy"],
     },
     {
         slug: "policy",
         title: "Content policy",
         summary: "Model capabilities, no per-request platform content review, and caller responsibility.",
-        keywords: ["model capabilities", "content review", "content", "policy", "透明传递", "内容"],
+        keywords: ["model capabilities", "content review", "content", "policy"],
     },
     {
         slug: "agents",
         title: "Agent integrations",
         summary: "Safe patterns for AI agents using the public API contract.",
-        keywords: ["agent", "mcp", "tool", "智能体", "代理"],
+        keywords: ["agent", "mcp", "tool"],
     },
     {
         slug: "quotes-and-compatibility",
@@ -109,16 +106,13 @@ const definitions = [
             "openai",
             "responses",
             "chat completions",
-            "报价",
-            "价格变化",
-            "文本",
         ],
     },
     {
         slug: "api-reference",
         title: "API reference",
         summary: "The complete public endpoint and envelope reference.",
-        keywords: ["openapi", "endpoint", "reference", "接口", "参考"],
+        keywords: ["openapi", "endpoint", "reference"],
     },
 ];
 export const DOCUMENTATION = definitions.map((entry) => ({