plugin-ai-api 1.0.23 → 1.0.25

package/dist/server/usage.js

@@ -42,6 +42,24 @@ function getAiApiState(ctx) {
 function normalizeTokenCount(value) {
   return typeof value === "number" && Number.isSafeInteger(value) && value >= 0 ? value : null;
 }
+function extractPromptCacheTokens(source) {
+  const inputDetails = source.input_token_details;
+  if (inputDetails && typeof inputDetails === "object") {
+    const cacheRead = inputDetails.cache_read;
+    if (typeof cacheRead === "number") return cacheRead;
+    const cachedTokens = inputDetails.cached_tokens;
+    if (typeof cachedTokens === "number") return cachedTokens;
+  }
+  const promptDetails = source.prompt_tokens_details;
+  if (promptDetails && typeof promptDetails === "object") {
+    const cachedTokens = promptDetails.cached_tokens;
+    if (typeof cachedTokens === "number") return cachedTokens;
+  }
+  if (typeof source.cached_tokens === "number") return source.cached_tokens;
+  if (typeof source.cache_read_tokens === "number") return source.cache_read_tokens;
+  if (typeof source.cache_read === "number") return source.cache_read;
+  return null;
+}
 function normalizeUsage(value) {
   if (!value || typeof value !== "object") return void 0;
   const source = value;
@@ -55,11 +73,22 @@ function normalizeUsage(value) {
   return {
     prompt_tokens: prompt,
     completion_tokens: completion,
-    total_tokens: total
+    total_tokens: total,
+    prompt_cache_tokens: extractPromptCacheTokens(source)
   };
 }
-function setAiApiUsageResult(ctx, value, metadata = {}) {
-  const usage = normalizeUsage(value);
+function setAiApiUsageResult(ctx, value, metadata = {}, responseMetadata) {
+  let usage = normalizeUsage(value);
+  if ((usage == null ? void 0 : usage.prompt_cache_tokens) === null && responseMetadata && typeof responseMetadata === "object") {
+    const responseMetaRecord = responseMetadata;
+    let cacheTokens = extractPromptCacheTokens(responseMetaRecord);
+    if (cacheTokens === null && responseMetaRecord.usage && typeof responseMetaRecord.usage === "object") {
+      cacheTokens = extractPromptCacheTokens(responseMetaRecord.usage);
+    }
+    if (cacheTokens !== null) {
+      usage = { ...usage, prompt_cache_tokens: cacheTokens };
+    }
+  }
   getAiApiState(ctx).aiApiUsageResult = usage ? { source: "provider", usage, ...metadata } : { source: "unavailable", ...metadata };
   (0, import_app_observability.addAiApiUsage)(ctx, usage);
   return usage;
@@ -133,6 +162,7 @@ async function finishUsageRecord(ctx, id, startedAt, status) {
     inputTokens: (usage == null ? void 0 : usage.prompt_tokens) ?? null,
     outputTokens: (usage == null ? void 0 : usage.completion_tokens) ?? null,
     totalTokens: (usage == null ? void 0 : usage.total_tokens) ?? null,
+    promptCacheTokens: (usage == null ? void 0 : usage.prompt_cache_tokens) ?? null,
     resolvedService: ((_c = (_b = state.aiApiLlmBilling) == null ? void 0 : _b.resolution) == null ? void 0 : _c.service) ?? null,
     resolvedProvider: ((_e = (_d = state.aiApiLlmBilling) == null ? void 0 : _d.resolution) == null ? void 0 : _e.provider) ?? null,
     resolvedModel: ((_g = (_f = state.aiApiLlmBilling) == null ? void 0 : _f.resolution) == null ? void 0 : _g.model) ?? null,

package/dist/server/utils/direct-llm-context.js

@@ -0,0 +1,319 @@
+/**
+ * This file is part of the NocoBase (R) project.
+ * Copyright (c) 2020-2024 NocoBase Co., Ltd.
+ * Authors: NocoBase Team.
+ *
+ * This project is dual-licensed under AGPL-3.0 and NocoBase Commercial License.
+ * For more information, please refer to: https://www.nocobase.com/agreement.
+ */
+
+var __defProp = Object.defineProperty;
+var __getOwnPropDesc = Object.getOwnPropertyDescriptor;
+var __getOwnPropNames = Object.getOwnPropertyNames;
+var __hasOwnProp = Object.prototype.hasOwnProperty;
+var __export = (target, all) => {
+  for (var name in all)
+    __defProp(target, name, { get: all[name], enumerable: true });
+};
+var __copyProps = (to, from, except, desc) => {
+  if (from && typeof from === "object" || typeof from === "function") {
+    for (let key of __getOwnPropNames(from))
+      if (!__hasOwnProp.call(to, key) && key !== except)
+        __defProp(to, key, { get: () => from[key], enumerable: !(desc = __getOwnPropDesc(from, key)) || desc.enumerable });
+  }
+  return to;
+};
+var __toCommonJS = (mod) => __copyProps(__defProp({}, "__esModule", { value: true }), mod);
+var direct_llm_context_exports = {};
+__export(direct_llm_context_exports, {
+  DirectLlmContextError: () => DirectLlmContextError,
+  parseImageDimensions: () => parseImageDimensions,
+  prepareDirectLlmContext: () => prepareDirectLlmContext
+});
+module.exports = __toCommonJS(direct_llm_context_exports);
+class DirectLlmContextError extends Error {
+  constructor(code, message) {
+    super(message);
+    this.code = code;
+    this.name = "DirectLlmContextError";
+  }
+}
+function getValue(record, key) {
+  var _a;
+  return ((_a = record == null ? void 0 : record.get) == null ? void 0 : _a.call(record, key)) ?? (record == null ? void 0 : record[key]);
+}
+function positiveInteger(value) {
+  const parsed = Number(value);
+  return Number.isSafeInteger(parsed) && parsed > 0 ? parsed : void 0;
+}
+function estimateValueTokens(value) {
+  if (value === void 0) return 0;
+  return Math.ceil(Buffer.byteLength(JSON.stringify(value), "utf8") / 3);
+}
+function isRecord(value) {
+  return typeof value === "object" && value !== null && !Array.isArray(value);
+}
+function readBigEndian(buf, offset) {
+  return buf.readUInt32BE(offset);
+}
+function readUInt16BE(buf, offset) {
+  return buf.readUInt16BE(offset);
+}
+function readLittleEndian(buf, offset) {
+  return buf.readUInt16LE(offset);
+}
+function parsePngDimensions(buffer) {
+  if (buffer.length < 24) return void 0;
+  return { width: readBigEndian(buffer, 16), height: readBigEndian(buffer, 20) };
+}
+function parseJpegDimensions(buffer) {
+  let offset = 2;
+  while (offset < buffer.length) {
+    if (buffer[offset] !== 255) {
+      offset++;
+      continue;
+    }
+    const marker = buffer[offset + 1];
+    if (marker >= 192 && marker <= 195 || marker >= 197 && marker <= 199 || marker >= 201 && marker <= 203 || marker >= 205 && marker <= 207) {
+      if (offset + 9 <= buffer.length) {
+        return { height: readUInt16BE(buffer, offset + 5), width: readUInt16BE(buffer, offset + 7) };
+      }
+      return void 0;
+    }
+    if (marker === 217 || offset + 4 >= buffer.length) break;
+    const segmentLength = buffer.readUInt16BE(offset + 2);
+    offset += 2 + segmentLength;
+  }
+  return void 0;
+}
+function parseGifDimensions(buffer) {
+  if (buffer.length < 10) return void 0;
+  return { width: readLittleEndian(buffer, 6), height: readLittleEndian(buffer, 8) };
+}
+function parseWebpDimensions(buffer) {
+  if (buffer.length < 30) return void 0;
+  const riff = buffer.toString("ascii", 0, 4);
+  const webp = buffer.toString("ascii", 8, 12);
+  if (riff !== "RIFF" || webp !== "WEBP") return void 0;
+  const chunkType = buffer.toString("ascii", 12, 16);
+  if (chunkType === "VP8 " && buffer.length >= 26) {
+    return { width: readLittleEndian(buffer, 26), height: readLittleEndian(buffer, 28) };
+  }
+  if (chunkType === "VP8L" && buffer.length >= 24) {
+    const bits = buffer.readUInt32LE(21);
+    return {
+      width: (bits & 16383) + 1,
+      height: (bits >> 14 & 16383) + 1
+    };
+  }
+  if (chunkType === "VP8X" && buffer.length >= 30) {
+    return {
+      width: ((buffer[24] | buffer[25] << 8 | buffer[26] << 16) & 16777215) + 1,
+      height: ((buffer[27] | buffer[28] << 8 | buffer[29] << 16) & 16777215) + 1
+    };
+  }
+  return void 0;
+}
+function parseImageDimensions(buffer) {
+  if (buffer.length < 12) return void 0;
+  const header = buffer.toString("binary", 0, 4);
+  if (header === "\x89PNG") return parsePngDimensions(buffer);
+  if (header === "GIF8") return parseGifDimensions(buffer);
+  if (header === "RIFF") return parseWebpDimensions(buffer);
+  if (buffer[0] === 255 && buffer[1] === 216) return parseJpegDimensions(buffer);
+  return void 0;
+}
+const VISION_TILE_SIZE = 512;
+const VISION_LOW_DETAIL_TOKENS = 85;
+const VISION_TILE_TOKENS = 170;
+const VISION_HTTP_URL_ESTIMATE = 1024;
+const FILE_BASE64_FALLBACK_TOKENS = 1024;
+function estimateVisionTokensForDimensions(width, height) {
+  if (!Number.isFinite(width) || !Number.isFinite(height) || width <= 0 || height <= 0) {
+    return VISION_LOW_DETAIL_TOKENS;
+  }
+  const tilesX = Math.ceil(width / VISION_TILE_SIZE);
+  const tilesY = Math.ceil(height / VISION_TILE_SIZE);
+  return VISION_LOW_DETAIL_TOKENS + tilesX * tilesY * VISION_TILE_TOKENS;
+}
+function decodeBase64DataUrl(url) {
+  const match = /^data:([^;]+);base64,([A-Za-z0-9+/]+=?=?)$/.exec(url);
+  if (!match) return void 0;
+  try {
+    const buffer = Buffer.from(match[2], "base64");
+    return { mimeType: match[1].toLowerCase(), buffer };
+  } catch {
+    return void 0;
+  }
+}
+function estimateImageUrlTokens(imageUrl) {
+  const url = typeof imageUrl === "string" ? imageUrl : isRecord(imageUrl) ? String(imageUrl.url ?? "") : "";
+  if (!url) return 0;
+  if (url.startsWith("data:")) {
+    const decoded = decodeBase64DataUrl(url);
+    if (!decoded) return FILE_BASE64_FALLBACK_TOKENS;
+    if (!decoded.mimeType.startsWith("image/")) return FILE_BASE64_FALLBACK_TOKENS;
+    const dimensions = parseImageDimensions(decoded.buffer);
+    return dimensions ? estimateVisionTokensForDimensions(dimensions.width, dimensions.height) : VISION_LOW_DETAIL_TOKENS;
+  }
+  if (url.startsWith("http://") || url.startsWith("https://")) {
+    return VISION_HTTP_URL_ESTIMATE;
+  }
+  return 0;
+}
+function estimateFileBlockTokens(block) {
+  const file = isRecord(block.file) ? block.file : void 0;
+  if (!file) return 0;
+  const fileData = String(file.file_data ?? "");
+  if (fileData.startsWith("data:")) {
+    const decoded = decodeBase64DataUrl(fileData);
+    if (decoded) {
+      return Math.max(1, Math.ceil(decoded.buffer.length / 3));
+    }
+    return FILE_BASE64_FALLBACK_TOKENS;
+  }
+  return FILE_BASE64_FALLBACK_TOKENS;
+}
+function estimateContentBlockTokens(block) {
+  if (!isRecord(block)) return estimateValueTokens(block);
+  const type = typeof block.type === "string" ? block.type : void 0;
+  if (type === "text") {
+    return typeof block.text === "string" ? estimateValueTokens(block.text) + 4 : 4;
+  }
+  if (type === "image_url") {
+    return estimateImageUrlTokens(block.image_url) + 4;
+  }
+  if (type === "file") {
+    return estimateFileBlockTokens(block) + 4;
+  }
+  if (type === "file_url") {
+    return VISION_HTTP_URL_ESTIMATE + 4;
+  }
+  return estimateValueTokens(block) + 4;
+}
+function estimateMessageTokens(message) {
+  if (Array.isArray(message.content)) {
+    return message.content.reduce((total, block) => total + estimateContentBlockTokens(block), 4);
+  }
+  return estimateValueTokens(message) + 4;
+}
+function estimateMessagesTokens(messages) {
+  return messages.reduce((total, message) => total + estimateMessageTokens(message), 0);
+}
+function isInstruction(message) {
+  return message.role === "system" || message.role === "developer";
+}
+function splitTurns(messages) {
+  const fixed = [];
+  const turns = [];
+  let currentTurn;
+  for (const message of messages) {
+    if (isInstruction(message)) {
+      fixed.push(message);
+      continue;
+    }
+    if (message.role === "user" || !currentTurn) {
+      currentTurn = [message];
+      turns.push(currentTurn);
+      continue;
+    }
+    currentTurn.push(message);
+  }
+  return { fixed, turns };
+}
+function messagesWithTurns(messages, turns) {
+  const retainedMessages = new Set(turns.flat());
+  return messages.filter((message) => isInstruction(message) || retainedMessages.has(message));
+}
+async function loadModelMetadata(ctx, serviceName, modelId) {
+  const row = await ctx.db.getRepository("aiApiModelMetadata").findOne({
+    filter: { llmService: serviceName, model: modelId, enabled: true }
+  });
+  const contextWindow = positiveInteger(getValue(row, "contextWindow"));
+  const maxCompletionTokens = positiveInteger(getValue(row, "maxCompletionTokens"));
+  if (!contextWindow || !maxCompletionTokens) {
+    throw new DirectLlmContextError(
+      "model_context_metadata_not_configured",
+      `Context metadata is not configured for '${serviceName}/${modelId}'. Configure context window and max completion tokens.`
+    );
+  }
+  return { contextWindow, maxCompletionTokens };
+}
+async function resolveOverflowBehavior(ctx) {
+  var _a;
+  const userId = (_a = ctx.state.currentUser) == null ? void 0 : _a.id;
+  if (userId === null || userId === void 0) return "reject";
+  const policy = await ctx.db.getRepository("aiApiUserQuotaPolicies").findOne({
+    filter: { userId, enabled: true }
+  });
+  return getValue(policy, "contextOverflowBehavior") === "truncate" ? "truncate" : "reject";
+}
+function resolveReservedOutputTokens(options, metadata) {
+  const requested = positiveInteger(options.maxCompletionTokens ?? options.maxTokens);
+  if (requested && requested > metadata.maxCompletionTokens) {
+    throw new DirectLlmContextError(
+      "max_completion_tokens_exceeds_model_limit",
+      `Requested max completion tokens (${requested}) exceeds the model limit (${metadata.maxCompletionTokens}).`
+    );
+  }
+  return requested ?? metadata.maxCompletionTokens;
+}
+async function prepareDirectLlmContext(ctx, options) {
+  const [metadata, behavior] = await Promise.all([
+    loadModelMetadata(ctx, options.serviceName, options.modelId),
+    resolveOverflowBehavior(ctx)
+  ]);
+  const reservedOutputTokens = resolveReservedOutputTokens(options, metadata);
+  const inputTokenBudget = metadata.contextWindow - reservedOutputTokens;
+  if (inputTokenBudget <= 0) {
+    throw new DirectLlmContextError(
+      "context_length_exceeded",
+      `The model context window (${metadata.contextWindow}) leaves no input capacity after reserving ${reservedOutputTokens} output tokens.`
+    );
+  }
+  const fixedOverheadTokens = estimateValueTokens(options.tools) + (options.tools === void 0 ? 0 : 4);
+  const originalEstimate = estimateMessagesTokens(options.messages) + fixedOverheadTokens;
+  if (originalEstimate <= inputTokenBudget) {
+    return {
+      messages: options.messages,
+      estimatedInputTokens: originalEstimate,
+      inputTokenBudget,
+      reservedOutputTokens,
+      truncated: false
+    };
+  }
+  if (behavior === "reject") {
+    throw new DirectLlmContextError(
+      "context_length_exceeded",
+      `Estimated input tokens (${originalEstimate}) exceed the allowed input budget (${inputTokenBudget}).`
+    );
+  }
+  const { turns } = splitTurns(options.messages);
+  let remainingTurns = turns;
+  let messages = messagesWithTurns(options.messages, remainingTurns);
+  let estimatedInputTokens = estimateMessagesTokens(messages) + fixedOverheadTokens;
+  while (remainingTurns.length > 1 && estimatedInputTokens > inputTokenBudget) {
+    remainingTurns = remainingTurns.slice(1);
+    messages = messagesWithTurns(options.messages, remainingTurns);
+    estimatedInputTokens = estimateMessagesTokens(messages) + fixedOverheadTokens;
+  }
+  if (estimatedInputTokens > inputTokenBudget) {
+    throw new DirectLlmContextError(
+      "context_length_exceeded",
+      `The fixed instructions, tools, and newest conversation turn require ${estimatedInputTokens} input tokens, exceeding the allowed budget (${inputTokenBudget}).`
+    );
+  }
+  return {
+    messages,
+    estimatedInputTokens,
+    inputTokenBudget,
+    reservedOutputTokens,
+    truncated: messages.length !== options.messages.length
+  };
+}
+// Annotate the CommonJS export names for ESM import in node:
+0 && (module.exports = {
+  DirectLlmContextError,
+  parseImageDimensions,
+  prepareDirectLlmContext
+});

package/dist/server/utils/openai-format.js

@@ -99,7 +99,19 @@ function toOpenAIResponse(options) {
         finish_reason: finishReason
       }
     ],
-    usage: usage || { prompt_tokens: null, completion_tokens: null, total_tokens: null }
+    usage: usage ? {
+      prompt_tokens: usage.prompt_tokens ?? null,
+      completion_tokens: usage.completion_tokens ?? null,
+      total_tokens: usage.total_tokens ?? null,
+      prompt_tokens_details: {
+        cached_tokens: usage.prompt_cache_tokens ?? null
+      }
+    } : {
+      prompt_tokens: null,
+      completion_tokens: null,
+      total_tokens: null,
+      prompt_tokens_details: { cached_tokens: null }
+    }
   };
 }
 function toOpenAIStreamChunk(options) {
@@ -129,7 +141,14 @@ function toOpenAIUsageChunk(options) {
     created: Math.floor(Date.now() / 1e3),
     model,
     choices: [],
-    usage
+    usage: {
+      prompt_tokens: usage.prompt_tokens,
+      completion_tokens: usage.completion_tokens,
+      total_tokens: usage.total_tokens,
+      prompt_tokens_details: {
+        cached_tokens: usage.prompt_cache_tokens ?? null
+      }
+    }
   };
 }
 function toOpenAIEmbeddingsResponse(options) {

package/dist/server/validation.js

@@ -101,6 +101,9 @@ function validateQuotaPolicy(model) {
   if (!["allow", "use_reserved"].includes(String(model.get("missingUsageBehavior")))) {
     throw new Error("missingUsageBehavior must be allow or use_reserved.");
   }
+  if (!["reject", "truncate"].includes(String(model.get("contextOverflowBehavior") ?? "reject"))) {
+    throw new Error("contextOverflowBehavior must be reject or truncate.");
+  }
   try {
     (0, import_dayjs.default)().tz(String(model.get("timezone") || "UTC"));
   } catch {

package/dist/swagger.js

@@ -285,6 +285,11 @@ var swagger_default = {
             maximum: 100,
             default: 10,
             description: "Max request body size in megabytes. Requests above this return 413. The gateway buffers each body in memory, so values above 100 are rejected."
+          },
+          pdfRenderPagesAsImages: {
+            type: "boolean",
+            default: false,
+            description: "When true, PDF file/file_url blocks are rendered to per-page PNG images and sent as image_url blocks. Requires a registered PdfToImageRenderer. When false or no renderer is available, PDFs are forwarded as file blocks."
           }
         }
       },
@@ -299,9 +304,9 @@ var swagger_default = {
       },
       ContentBlock: {
         type: "object",
-        description: "A multimodal content block. Only text and image_url blocks are forwarded to the provider; any other type is rejected with 400 unsupported_content_block.",
+        description: "A multimodal content block. text, image_url, file and file_url blocks are forwarded; file and file_url blocks are first run through the configurable file processor service.",
         properties: {
-          type: { type: "string", enum: ["text", "image_url"] },
+          type: { type: "string", enum: ["text", "image_url", "file", "file_url"] },
           text: { type: "string" },
           image_url: {
             type: "object",
@@ -314,6 +319,34 @@ var swagger_default = {
               detail: { type: "string", enum: ["auto", "low", "high"] }
             },
             required: ["url"]
+          },
+          file: {
+            type: "object",
+            properties: {
+              file_data: {
+                type: "string",
+                description: "A base64 data URL, e.g. data:application/pdf;base64,JVBERi0...",
+                example: "data:application/pdf;base64,JVBERi0..."
+              },
+              filename: { type: "string" },
+              mime_type: {
+                type: "string",
+                description: "MIME type of the file, e.g. application/pdf",
+                example: "application/pdf"
+              }
+            },
+            required: ["file_data"]
+          },
+          file_url: {
+            type: "object",
+            properties: {
+              url: {
+                type: "string",
+                description: "An http(s) URL pointing to a file. The gateway downloads the file and converts it to a file block.",
+                example: "https://example.com/document.pdf"
+              }
+            },
+            required: ["url"]
           }
         },
         required: ["type"]
@@ -374,7 +407,13 @@ var swagger_default = {
             properties: {
               prompt_tokens: { type: "integer" },
               completion_tokens: { type: "integer" },
-              total_tokens: { type: "integer" }
+              total_tokens: { type: "integer" },
+              prompt_tokens_details: {
+                type: "object",
+                properties: {
+                  cached_tokens: { type: "integer" }
+                }
+              }
             }
           }
         }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "plugin-ai-api",
-  "version": "1.0.23",
+  "version": "1.0.25",
   "main": "dist/server/index.js",
   "dependencies": {},
   "peerDependencies": {

package/src/client-v2/pages/UsagePage.tsx

@@ -32,6 +32,7 @@ interface UsageRecord {
   inputTokens?: number;
   outputTokens?: number;
   totalTokens?: number;
+  promptCacheTokens?: number;
   estimatedCost?: string;
   currency?: string;
   costStatus?: string;
@@ -51,6 +52,7 @@ interface UsageSummary {
   inputTokens: number;
   outputTokens: number;
   totalTokens: number;
+  promptCacheTokens: number;
   costsByCurrency: { currency: string; totalCost: string }[];
 }
 
@@ -65,6 +67,7 @@ const emptySummary: UsageSummary = {
   inputTokens: 0,
   outputTokens: 0,
   totalTokens: 0,
+  promptCacheTokens: 0,
   costsByCurrency: [],
 };
 
@@ -153,6 +156,7 @@ export default function UsagePage() {
     { title: t('Input tokens'), dataIndex: 'inputTokens', key: 'inputTokens', width: 110 },
     { title: t('Output tokens'), dataIndex: 'outputTokens', key: 'outputTokens', width: 110 },
     { title: t('Total tokens'), dataIndex: 'totalTokens', key: 'totalTokens', width: 110 },
+    { title: t('Prompt cache tokens'), dataIndex: 'promptCacheTokens', key: 'promptCacheTokens', width: 140 },
     {
       title: t('Cost'),
       key: 'cost',
@@ -234,6 +238,11 @@ export default function UsagePage() {
             <Statistic title={t('Total tokens')} value={summary.totalTokens} />
           </Card>
         </Col>
+        <Col xs={24} sm={12} lg={6}>
+          <Card size="small">
+            <Statistic title={t('Prompt cache tokens')} value={summary.promptCacheTokens} />
+          </Card>
+        </Col>
         <Col xs={24}>
           <Card size="small">
             <Statistic title={t('Total cost')} value={totalCost} />

package/src/client-v2/pages/UserQuotasPage.tsx

@@ -39,6 +39,7 @@ interface QuotaPolicy {
   currency: string;
   rejectUnpricedModel: boolean;
   missingUsageBehavior: 'allow' | 'use_reserved';
+  contextOverflowBehavior: 'reject' | 'truncate';
 }
 
 export default function UserQuotasPage() {
@@ -85,6 +86,7 @@ export default function UserQuotasPage() {
       currency: 'USD',
       rejectUnpricedModel: true,
       missingUsageBehavior: 'use_reserved',
+      contextOverflowBehavior: 'reject',
     } as QuotaPolicy);
     setOpen(true);
   };
@@ -157,6 +159,14 @@ export default function UserQuotasPage() {
       width: 120,
       render: (value, record) => (value == null ? t('Unlimited') : `${value} ${record.currency}`),
     },
+    {
+      title: t('Context overflow behavior'),
+      dataIndex: 'contextOverflowBehavior',
+      key: 'contextOverflowBehavior',
+      width: 150,
+      render: (value: QuotaPolicy['contextOverflowBehavior']) =>
+        value === 'truncate' ? t('Truncate oldest conversation turns') : t('Reject request'),
+    },
     { title: t('Timezone'), dataIndex: 'timezone', key: 'timezone', width: 140 },
     {
       title: t('Status'),
@@ -248,6 +258,14 @@ export default function UserQuotasPage() {
               ]}
             />
           </Form.Item>
+          <Form.Item name="contextOverflowBehavior" label={t('Context overflow behavior')} rules={[{ required: true }]}>
+            <Select
+              options={[
+                { label: t('Reject request'), value: 'reject' },
+                { label: t('Truncate oldest conversation turns'), value: 'truncate' },
+              ]}
+            />
+          </Form.Item>
           <Form.Item name="enabled" label={t('Enabled')} valuePropName="checked">
             <Switch />
           </Form.Item>

package/src/locale/en-US.json

@@ -53,6 +53,9 @@
   "Missing usage behavior": "Missing usage behavior",
   "Use reserved estimate": "Use reserved estimate",
   "Allow without token charge": "Allow without token charge",
+  "Context overflow behavior": "Context overflow behavior",
+  "Reject request": "Reject request",
+  "Truncate oldest conversation turns": "Truncate oldest conversation turns",
   "Started at": "Started at",
   "Requested model": "Requested model",
   "Resolved service": "Resolved service",
@@ -60,6 +63,7 @@
   "Input tokens": "Input tokens",
   "Output tokens": "Output tokens",
   "Total tokens": "Total tokens",
+  "Prompt cache tokens": "Prompt cache tokens",
   "Cost": "Cost",
   "Cost status": "Cost status",
   "Request ID": "Request ID",

package/src/locale/vi-VN.json

@@ -53,6 +53,9 @@
   "Missing usage behavior": "Xử lý khi thiếu token usage",
   "Use reserved estimate": "Dùng số liệu giữ chỗ để ước tính",
   "Allow without token charge": "Cho phép và không tính token",
+  "Context overflow behavior": "Xử lý khi vượt context",
+  "Reject request": "Từ chối request",
+  "Truncate oldest conversation turns": "Cắt các lượt hội thoại cũ nhất",
   "Started at": "Bắt đầu lúc",
   "Requested model": "Model được yêu cầu",
   "Resolved service": "Service đã resolve",
@@ -60,6 +63,7 @@
   "Input tokens": "Input token",
   "Output tokens": "Output token",
   "Total tokens": "Tổng token",
+  "Prompt cache tokens": "Prompt cache token",
   "Cost": "Chi phí",
   "Cost status": "Trạng thái chi phí",
   "Request ID": "Request ID",

package/src/locale/zh-CN.json

@@ -53,6 +53,9 @@
   "Missing usage behavior": "缺少用量时的行为",
   "Use reserved estimate": "使用预留估算",
   "Allow without token charge": "允许且不计令牌",
+  "Context overflow behavior": "上下文超限处理",
+  "Reject request": "拒绝请求",
+  "Truncate oldest conversation turns": "截断最早的对话轮次",
   "Started at": "开始时间",
   "Requested model": "请求模型",
   "Resolved service": "解析后的服务",
@@ -60,6 +63,7 @@
   "Input tokens": "输入令牌",
   "Output tokens": "输出令牌",
   "Total tokens": "总令牌",
+  "Prompt cache tokens": "提示缓存令牌",
   "Cost": "费用",
   "Cost status": "费用状态",
   "Request ID": "请求 ID",