@omnicross/core 0.1.6 → 0.1.8

package/dist/{chunk-DLXZEZWO.cjs → chunk-2E42SAW3.cjs}

@@ -4,7 +4,7 @@
 
 var _pricingtypes = require('@omnicross/contracts/pricing-types');
 var PricingEngine = (_class = class {
-  constructor(store, logger, options = {}) {;_class.prototype.__init.call(this);_class.prototype.__init2.call(this);_class.prototype.__init3.call(this);_class.prototype.__init4.call(this);
+  constructor(store, logger, options = {}) {;_class.prototype.__init.call(this);_class.prototype.__init2.call(this);_class.prototype.__init3.call(this);_class.prototype.__init4.call(this);_class.prototype.__init5.call(this);
     this.store = store;
     this.logger = logger;
     this.options = options;
@@ -21,9 +21,18 @@ var PricingEngine = (_class = class {
    * non-wildcard entry per model wins.
    */
   __init2() {this.cacheByModel = /* @__PURE__ */ new Map()}
-  __init3() {this.cacheLoaded = false}
+  /**
+   * Last-resort fuzzy index: `[normalizedModelId, entry]` pairs over the same
+   * non-wildcard, non-openrouter rows as `cacheByModel`. Used when even the
+   * model-name fallback misses — e.g. usage recorded under a bare model id
+   * (`glm-5.2`) while the catalog only knows prefixed slugs
+   * (`cloudflare/@cf/zai-org/glm-5.2`). This is an ESTIMATE: the matched row
+   * may be an aggregate-route price rather than the direct-vendor rate.
+   */
+  __init3() {this.fuzzyEntries = []}
+  __init4() {this.cacheLoaded = false}
   /** Coalesce manual and background refreshes to prevent a catalog stampede. */
-  __init4() {this.refreshInFlight = null}
+  __init5() {this.refreshInFlight = null}
   // ===== Cache management =====
   /** Lazy-load and cache the full pricing table. */
   async ensureCache() {
@@ -31,9 +40,11 @@ var PricingEngine = (_class = class {
     const rows = await this.store.getAll();
     this.cache = new Map(rows.map((r) => [keyOf(r.providerId, r.modelId), r]));
     this.cacheByModel = /* @__PURE__ */ new Map();
+    this.fuzzyEntries = [];
     for (const r of rows) {
       if (r.providerId === "*" || r.providerId === "openrouter" || r.source === "openrouter") continue;
       if (!this.cacheByModel.has(r.modelId)) this.cacheByModel.set(r.modelId, r);
+      this.fuzzyEntries.push([normalizeModelId(r.modelId), r]);
     }
     this.cacheLoaded = true;
   }
@@ -53,10 +64,36 @@ var PricingEngine = (_class = class {
    *   2. wildcard provider ('*', modelId)
    *   3. model name alone (any provider) — tolerates a runtime-alias vs
    *      canonical provider-id mismatch for the same model.
+   *   4. normalized fuzzy match (last resort) — matches a bare id like
+   *      `glm-5.2` against prefixed catalog slugs like
+   *      `cloudflare/@cf/zai-org/glm-5.2`. ESTIMATE only: the matched row may
+   *      be an aggregate-route price, not the direct-vendor rate.
    */
   async getEntry(providerId, modelId) {
     await this.ensureCache();
-    return _nullishCoalesce(_nullishCoalesce(_nullishCoalesce(this.cache.get(keyOf(providerId, modelId)), () => ( this.cache.get(keyOf("*", modelId)))), () => ( this.cacheByModel.get(modelId))), () => ( null));
+    const exact = _nullishCoalesce(_nullishCoalesce(_nullishCoalesce(this.cache.get(keyOf(providerId, modelId)), () => ( this.cache.get(keyOf("*", modelId)))), () => ( this.cacheByModel.get(modelId))), () => ( null));
+    if (exact) return exact;
+    return this.fuzzyMatch(modelId);
+  }
+  /**
+   * Normalized substring match over the catalog's non-openrouter rows. Lowercases
+   * and strips all non-alphanumerics, then matches bidirectionally, preferring
+   * the shortest catalog key (tightest match) to reduce over-matching version
+   * variants. Returns null when nothing matches.
+   */
+  fuzzyMatch(modelId) {
+    const normQuery = normalizeModelId(modelId);
+    if (!normQuery) return null;
+    let best = null;
+    let bestLen = Infinity;
+    for (const [key, entry] of this.fuzzyEntries) {
+      if (!key || !key.includes(normQuery) && !normQuery.includes(key)) continue;
+      if (key.length < bestLen) {
+        best = entry;
+        bestLen = key.length;
+      }
+    }
+    return best;
   }
   // ===== Cost calculation =====
   /**
@@ -187,6 +224,7 @@ var PricingEngine = (_class = class {
   }
 }, _class);
 var keyOf = (providerId, modelId) => `${providerId}::${modelId}`;
+var normalizeModelId = (modelId) => (_nullishCoalesce(modelId, () => ( ""))).toLowerCase().replace(/[^a-z0-9]/g, "");
 var PROVIDER_ALIASES = {
   vertex_ai: "gemini",
   google: "gemini",

package/dist/{chunk-7AK7B36B.cjs → chunk-2Z4XM3H5.cjs}

@@ -1,15 +1,15 @@
 "use strict";Object.defineProperty(exports, "__esModule", {value: true});
 
-var _chunkV4TXMEJScjs = require('./chunk-V4TXMEJS.cjs');
+var _chunkOXHNSZ53cjs = require('./chunk-OXHNSZ53.cjs');
 
 
-var _chunk6VUQPOV3cjs = require('./chunk-6VUQPOV3.cjs');
+var _chunkV4TXMEJScjs = require('./chunk-V4TXMEJS.cjs');
 
 
-var _chunkOF3BECU2cjs = require('./chunk-OF3BECU2.cjs');
+var _chunkOVQBNVSVcjs = require('./chunk-OVQBNVSV.cjs');
 
 
-var _chunk4VLRHHFEcjs = require('./chunk-4VLRHHFE.cjs');
+var _chunkOF3BECU2cjs = require('./chunk-OF3BECU2.cjs');
 
 
 var _chunkOKMJLEXXcjs = require('./chunk-OKMJLEXX.cjs');
@@ -19,8 +19,8 @@ var BuiltinTransformers = {
   OpenAITransformer: _chunkOF3BECU2cjs.OpenAITransformer,
   GeminiTransformer: _chunkV4TXMEJScjs.GeminiTransformer,
   GeminiCodeAssistTransformer: _chunkOKMJLEXXcjs.GeminiCodeAssistTransformer,
-  AnthropicTransformer: _chunk4VLRHHFEcjs.AnthropicTransformer,
-  OpenAIResponseTransformer: _chunk6VUQPOV3cjs.OpenAIResponseTransformer
+  AnthropicTransformer: _chunkOXHNSZ53cjs.AnthropicTransformer,
+  OpenAIResponseTransformer: _chunkOVQBNVSVcjs.OpenAIResponseTransformer
 };
 function getBuiltinTransformers() {
   return BuiltinTransformers;

package/dist/{chunk-OEWZUX4G.js → chunk-32O35CIF.js}

@@ -1,3 +1,8 @@
+import {
+  chatUsageToResponsesUsage,
+  responsesUsageToChatUsage
+} from "./chunk-FU7EAR73.js";
+
 // src/transformer/transformers/OpenAIResponseTransformer.ts
 var OpenAIResponseTransformer = class {
   static TransformerName = "openai-response";
@@ -433,11 +438,7 @@ function convertResponseApiJsonToOpenAI(data) {
         finish_reason: toolCalls.length > 0 ? "tool_calls" : "stop"
       }
     ],
-    usage: usage ? {
-      prompt_tokens: usage.input_tokens || 0,
-      completion_tokens: usage.output_tokens || 0,
-      total_tokens: (usage.input_tokens || 0) + (usage.output_tokens || 0)
-    } : void 0
+    usage: responsesUsageToChatUsage(usage)
   };
 }
 function convertOpenAIJsonToResponseApi(data, codexTools) {
@@ -475,11 +476,7 @@ function convertOpenAIJsonToResponseApi(data, codexTools) {
     status: "completed",
     model: data.model || "unknown",
     output,
-    usage: usage ? {
-      input_tokens: usage.prompt_tokens || 0,
-      output_tokens: usage.completion_tokens || 0,
-      total_tokens: usage.total_tokens || (usage.prompt_tokens || 0) + (usage.completion_tokens || 0)
-    } : void 0
+    usage: chatUsageToResponsesUsage(usage)
   };
 }
 function convertResponseApiStreamToOpenAI(responseApiStream) {
@@ -589,11 +586,7 @@ function convertResponseApiStreamToOpenAI(responseApiStream) {
                 case "response.completed": {
                   const resp = event.response;
                   const respUsage = resp?.usage;
-                  const usage = respUsage ? {
-                    prompt_tokens: respUsage.input_tokens || 0,
-                    completion_tokens: respUsage.output_tokens || 0,
-                    total_tokens: (respUsage.input_tokens || 0) + (respUsage.output_tokens || 0)
-                  } : void 0;
+                  const usage = responsesUsageToChatUsage(respUsage);
                   emitChunk([{ index: 0, delta: {}, finish_reason: hasToolCalls ? "tool_calls" : "stop" }], usage);
                   safeEnqueue("data: [DONE]\n\n");
                   break;
@@ -801,11 +794,7 @@ function convertOpenAIStreamToResponseApi(openaiStream, codexTools) {
                     status: "completed",
                     model,
                     output,
-                    usage: chunk.usage ? {
-                      input_tokens: chunk.usage.prompt_tokens || 0,
-                      output_tokens: chunk.usage.completion_tokens || 0,
-                      total_tokens: chunk.usage.total_tokens || 0
-                    } : void 0
+                    usage: chatUsageToResponsesUsage(chunk.usage)
                   }
                 });
               }

package/dist/chunk-3SCOLIPT.cjs

@@ -0,0 +1,63 @@
+"use strict";Object.defineProperty(exports, "__esModule", {value: true});// src/provider-proxy/identity/headerMerge.ts
+function fillMissingHeaders(headers, source) {
+  const present = new Set(Object.keys(headers).map((k) => k.toLowerCase()));
+  for (const [key, value] of Object.entries(source)) {
+    const lower = key.toLowerCase();
+    if (present.has(lower)) continue;
+    headers[lower] = value;
+    present.add(lower);
+  }
+}
+function flattenHeaderValue(raw) {
+  if (raw === void 0) return void 0;
+  const value = Array.isArray(raw) ? raw.filter((v) => typeof v === "string").join(", ") : raw;
+  return typeof value === "string" && value.length > 0 ? value : void 0;
+}
+var NEVER_FORWARD_HEADERS = /* @__PURE__ */ new Set([
+  "authorization",
+  "x-api-key",
+  "cookie",
+  "proxy-authorization",
+  "host",
+  "connection",
+  "content-length",
+  "content-type",
+  "accept-encoding"
+]);
+
+// src/provider-proxy/identity/codexCliHeaders.ts
+var DEFAULT_CODEX_CLI_HEADERS = Object.freeze({
+  originator: "codex_cli_rs",
+  "user-agent": "codex_cli_rs/0.144.5",
+  version: "0.144.5"
+});
+var CODEX_FORWARD_ALLOWLIST = /* @__PURE__ */ new Set([
+  "version",
+  "openai-beta",
+  "session_id",
+  "originator",
+  "user-agent"
+]);
+function extractCodexClientHeaders(reqHeaders) {
+  const out = {};
+  for (const [rawKey, rawVal] of Object.entries(reqHeaders)) {
+    const key = rawKey.toLowerCase();
+    if (NEVER_FORWARD_HEADERS.has(key)) continue;
+    if (!CODEX_FORWARD_ALLOWLIST.has(key)) continue;
+    const value = flattenHeaderValue(rawVal);
+    if (value !== void 0) out[key] = value;
+  }
+  return out;
+}
+function codexAcceptHeader(isStream) {
+  return isStream ? "text/event-stream" : "application/json";
+}
+
+
+
+
+
+
+
+
+exports.fillMissingHeaders = fillMissingHeaders; exports.flattenHeaderValue = flattenHeaderValue; exports.NEVER_FORWARD_HEADERS = NEVER_FORWARD_HEADERS; exports.DEFAULT_CODEX_CLI_HEADERS = DEFAULT_CODEX_CLI_HEADERS; exports.extractCodexClientHeaders = extractCodexClientHeaders; exports.codexAcceptHeader = codexAcceptHeader;

package/dist/{chunk-SYQF7OIR.js → chunk-3WD4HTR2.js}

@@ -2,6 +2,7 @@
 import { appendFile, mkdir } from "fs/promises";
 import { dirname } from "path";
 var tracePath = null;
+var REDACTED_BODY = "***REDACTED (credential exchange)***";
 var SECRET_HEADER_NAMES = /* @__PURE__ */ new Set([
   "authorization",
   "api-key",
@@ -44,6 +45,7 @@ function __resetUpstreamTraceForTests() {
 }
 
 export {
+  REDACTED_BODY,
   setUpstreamTracePath,
   getUpstreamTracePath,
   redactHeaders,

package/dist/{chunk-AT7B7SOZ.js → chunk-5NJLQ6T6.js}

@@ -81,6 +81,20 @@ function flattenToolResultContent(content) {
 }
 
 // src/api-converter/anthropic-to-openai.ts
+function anthropicUsageToChatUsage(usage) {
+  const input = usage.input_tokens || 0;
+  const output = usage.output_tokens || 0;
+  const cacheRead = usage.cache_read_input_tokens || 0;
+  const cacheCreation = usage.cache_creation_input_tokens || 0;
+  const promptTokens = input + cacheRead + cacheCreation;
+  const out = {
+    prompt_tokens: promptTokens,
+    completion_tokens: output,
+    total_tokens: promptTokens + output
+  };
+  if (cacheRead > 0) out.prompt_tokens_details = { cached_tokens: cacheRead };
+  return out;
+}
 function convertAnthropicToOpenAI(response) {
   const content = [];
   const toolCalls = [];
@@ -118,11 +132,7 @@ function convertAnthropicToOpenAI(response) {
         finish_reason: finishReason
       }
     ],
-    usage: {
-      prompt_tokens: response.usage.input_tokens,
-      completion_tokens: response.usage.output_tokens,
-      total_tokens: response.usage.input_tokens + response.usage.output_tokens
-    }
+    usage: anthropicUsageToChatUsage(response.usage)
   };
   if (reasoningContent) {
     result.reasoning_content = reasoningContent;
@@ -352,11 +362,12 @@ async function* convertAnthropicStreamToOpenAI(anthropicStream) {
                 delta: {},
                 finish_reason: finishReason
               }],
-              usage: event.usage ? {
-                prompt_tokens: event.usage.input_tokens || 0,
-                completion_tokens: event.usage.output_tokens || 0,
-                total_tokens: (event.usage.input_tokens || 0) + (event.usage.output_tokens || 0)
-              } : void 0
+              usage: event.usage ? anthropicUsageToChatUsage({
+                input_tokens: event.usage.input_tokens || 0,
+                output_tokens: event.usage.output_tokens || 0,
+                cache_read_input_tokens: event.usage.cache_read_input_tokens,
+                cache_creation_input_tokens: event.usage.cache_creation_input_tokens
+              }) : void 0
             })}
 
 `;

package/dist/chunk-5RSZYQJH.cjs

@@ -0,0 +1,62 @@
+"use strict";Object.defineProperty(exports, "__esModule", {value: true}); function _optionalChain(ops) { let lastAccessLHS = undefined; let value = ops[0]; let i = 1; while (i < ops.length) { const op = ops[i]; const fn = ops[i + 1]; i += 2; if ((op === 'optionalAccess' || op === 'optionalCall') && value == null) { return undefined; } if (op === 'access' || op === 'optionalAccess') { lastAccessLHS = value; value = fn(value); } else if (op === 'call' || op === 'optionalCall') { value = fn((...args) => value.call(lastAccessLHS, ...args)); lastAccessLHS = undefined; } } return value; }// src/transformer/transformers/utils/usage-mapping.ts
+function num(obj, key) {
+  const v = _optionalChain([obj, 'optionalAccess', _ => _[key]]);
+  return typeof v === "number" && Number.isFinite(v) ? v : 0;
+}
+function detailCachedTokens(obj, detailKey) {
+  const detail = _optionalChain([obj, 'optionalAccess', _2 => _2[detailKey]]);
+  return num(detail, "cached_tokens");
+}
+function anthropicUsageToChatUsage(usage) {
+  if (!usage) return void 0;
+  const input = num(usage, "input_tokens");
+  const output = num(usage, "output_tokens");
+  const cacheRead = num(usage, "cache_read_input_tokens");
+  const cacheCreation = num(usage, "cache_creation_input_tokens");
+  const promptTokens = input + cacheRead + cacheCreation;
+  const out = {
+    prompt_tokens: promptTokens,
+    completion_tokens: output,
+    total_tokens: promptTokens + output
+  };
+  if (cacheRead > 0) {
+    out.prompt_tokens_details = { cached_tokens: cacheRead };
+  }
+  return out;
+}
+function chatUsageToResponsesUsage(usage) {
+  if (!usage) return void 0;
+  const prompt = num(usage, "prompt_tokens");
+  const completion = num(usage, "completion_tokens");
+  const out = {
+    input_tokens: prompt,
+    output_tokens: completion,
+    total_tokens: num(usage, "total_tokens") || prompt + completion
+  };
+  const cacheRead = detailCachedTokens(usage, "prompt_tokens_details");
+  if (cacheRead > 0) {
+    out.input_tokens_details = { cached_tokens: cacheRead };
+  }
+  return out;
+}
+function responsesUsageToChatUsage(usage) {
+  if (!usage) return void 0;
+  const input = num(usage, "input_tokens");
+  const output = num(usage, "output_tokens");
+  const out = {
+    prompt_tokens: input,
+    completion_tokens: output,
+    total_tokens: num(usage, "total_tokens") || input + output
+  };
+  const cacheRead = detailCachedTokens(usage, "input_tokens_details");
+  if (cacheRead > 0) {
+    out.prompt_tokens_details = { cached_tokens: cacheRead };
+  }
+  return out;
+}
+
+
+
+
+
+exports.anthropicUsageToChatUsage = anthropicUsageToChatUsage; exports.chatUsageToResponsesUsage = chatUsageToResponsesUsage; exports.responsesUsageToChatUsage = responsesUsageToChatUsage;

package/dist/{chunk-IFXJ2BG7.js → chunk-7C7DET6S.js}

@@ -11,6 +11,29 @@ import {
 
 // src/outbound-api/auditCapture.ts
 import { randomUUID } from "crypto";
+function completeUtf8PrefixLength(buf) {
+  if (buf.length === 0) return 0;
+  let leadIndex = buf.length - 1;
+  while (leadIndex >= 0 && (buf[leadIndex] & 192) === 128) leadIndex -= 1;
+  if (leadIndex < 0) return 0;
+  const lead = buf[leadIndex];
+  let expectedLength = 1;
+  if ((lead & 224) === 192) expectedLength = 2;
+  else if ((lead & 240) === 224) expectedLength = 3;
+  else if ((lead & 248) === 240) expectedLength = 4;
+  return buf.length - leadIndex < expectedLength ? leadIndex : buf.length;
+}
+function truncateToBytes(text, maxBytes) {
+  if (maxBytes < 0) return text;
+  const buf = Buffer.from(text, "utf8");
+  if (buf.length <= maxBytes) return text;
+  const head = buf.subarray(0, maxBytes);
+  return head.subarray(0, completeUtf8PrefixLength(head)).toString("utf8");
+}
+function decodeCapturedBody(body, truncated) {
+  if (!truncated) return body.toString("utf8");
+  return body.subarray(0, completeUtf8PrefixLength(body)).toString("utf8");
+}
 function resolveClientIp(req, trustForwardedFor) {
   if (trustForwardedFor) {
     const xff = req.headers["x-forwarded-for"];
@@ -25,18 +48,39 @@ function resolveClientIp(req, trustForwardedFor) {
 function beginAuditCapture(req, res, now) {
   const config = getAuditCaptureConfig();
   if (!config) return null;
-  let rawRequestBody;
+  let requestBody;
   const responseChunks = [];
+  let responseBytes = 0;
+  let responseTruncated = false;
   let finished = false;
   const ctx = {
     setRequestBody(raw) {
-      if (config.captureBodies) rawRequestBody = raw;
+      if (config.captureBodies) {
+        requestBody = truncateToBytes(redactAuditText(raw), config.maxBodyBytes);
+      }
     }
   };
   if (config.captureBodies) {
     installResponseCapture(res, {
       push(chunk) {
+        if (config.maxBodyBytes < 0) {
+          responseChunks.push(chunk);
+          responseBytes += chunk.length;
+          return;
+        }
+        const remaining = config.maxBodyBytes - responseBytes;
+        if (remaining <= 0) {
+          responseTruncated = true;
+          return;
+        }
+        if (chunk.length > remaining) {
+          responseChunks.push(chunk.subarray(0, remaining));
+          responseBytes += remaining;
+          responseTruncated = true;
+          return;
+        }
         responseChunks.push(chunk);
+        responseBytes += chunk.length;
       }
     });
   }
@@ -69,12 +113,13 @@ function beginAuditCapture(req, res, now) {
       }
       if (ctx.error) record.error = redactAuditText(ctx.error);
       if (config.captureBodies) {
-        if (rawRequestBody != null && rawRequestBody.length > 0) {
-          record.requestBody = redactAuditText(rawRequestBody);
+        if (requestBody != null && requestBody.length > 0) {
+          record.requestBody = requestBody;
         }
         if (responseChunks.length > 0) {
-          const body = Buffer.concat(responseChunks).toString("utf8");
-          record.responseBody = redactAuditText(body);
+          const captured = Buffer.concat(responseChunks, responseBytes);
+          const body = decodeCapturedBody(captured, responseTruncated);
+          record.responseBody = truncateToBytes(redactAuditText(body), config.maxBodyBytes);
         }
       }
       recordAudit(record);

package/dist/{chunk-SNYBEXLB.cjs → chunk-EPN67EGP.cjs}

@@ -78,7 +78,12 @@ var UsageRecorder = class {
       costSavedByCacheUsd,
       rawUsage: input.rawUsage != null ? safeStringify(input.rawUsage) : null,
       runId: _nullishCoalesce(input.runId, () => ( null)),
-      eventId: _nullishCoalesce(input.eventId, () => ( null))
+      routeLeaseId: _nullishCoalesce(input.routeLeaseId, () => ( null)),
+      routeLeaseConsumer: _nullishCoalesce(input.routeLeaseConsumer, () => ( null)),
+      routeLeaseStageId: _nullishCoalesce(input.routeLeaseStageId, () => ( null)),
+      eventId: _nullishCoalesce(input.eventId, () => ( null)),
+      ...input.cacheKeySource !== void 0 ? { cacheKeySource: input.cacheKeySource } : {},
+      ...input.cacheKeyInjected !== void 0 ? { cacheKeyInjected: input.cacheKeyInjected } : {}
     };
     try {
       return await this.store.insert(row);

package/dist/chunk-FU7EAR73.js

@@ -0,0 +1,62 @@
+// src/transformer/transformers/utils/usage-mapping.ts
+function num(obj, key) {
+  const v = obj?.[key];
+  return typeof v === "number" && Number.isFinite(v) ? v : 0;
+}
+function detailCachedTokens(obj, detailKey) {
+  const detail = obj?.[detailKey];
+  return num(detail, "cached_tokens");
+}
+function anthropicUsageToChatUsage(usage) {
+  if (!usage) return void 0;
+  const input = num(usage, "input_tokens");
+  const output = num(usage, "output_tokens");
+  const cacheRead = num(usage, "cache_read_input_tokens");
+  const cacheCreation = num(usage, "cache_creation_input_tokens");
+  const promptTokens = input + cacheRead + cacheCreation;
+  const out = {
+    prompt_tokens: promptTokens,
+    completion_tokens: output,
+    total_tokens: promptTokens + output
+  };
+  if (cacheRead > 0) {
+    out.prompt_tokens_details = { cached_tokens: cacheRead };
+  }
+  return out;
+}
+function chatUsageToResponsesUsage(usage) {
+  if (!usage) return void 0;
+  const prompt = num(usage, "prompt_tokens");
+  const completion = num(usage, "completion_tokens");
+  const out = {
+    input_tokens: prompt,
+    output_tokens: completion,
+    total_tokens: num(usage, "total_tokens") || prompt + completion
+  };
+  const cacheRead = detailCachedTokens(usage, "prompt_tokens_details");
+  if (cacheRead > 0) {
+    out.input_tokens_details = { cached_tokens: cacheRead };
+  }
+  return out;
+}
+function responsesUsageToChatUsage(usage) {
+  if (!usage) return void 0;
+  const input = num(usage, "input_tokens");
+  const output = num(usage, "output_tokens");
+  const out = {
+    prompt_tokens: input,
+    completion_tokens: output,
+    total_tokens: num(usage, "total_tokens") || input + output
+  };
+  const cacheRead = detailCachedTokens(usage, "input_tokens_details");
+  if (cacheRead > 0) {
+    out.prompt_tokens_details = { cached_tokens: cacheRead };
+  }
+  return out;
+}
+
+export {
+  anthropicUsageToChatUsage,
+  chatUsageToResponsesUsage,
+  responsesUsageToChatUsage
+};

package/dist/{chunk-UX7VWA46.js → chunk-FXR6N3SR.js}

@@ -21,6 +21,15 @@ var PricingEngine = class {
    * non-wildcard entry per model wins.
    */
   cacheByModel = /* @__PURE__ */ new Map();
+  /**
+   * Last-resort fuzzy index: `[normalizedModelId, entry]` pairs over the same
+   * non-wildcard, non-openrouter rows as `cacheByModel`. Used when even the
+   * model-name fallback misses — e.g. usage recorded under a bare model id
+   * (`glm-5.2`) while the catalog only knows prefixed slugs
+   * (`cloudflare/@cf/zai-org/glm-5.2`). This is an ESTIMATE: the matched row
+   * may be an aggregate-route price rather than the direct-vendor rate.
+   */
+  fuzzyEntries = [];
   cacheLoaded = false;
   /** Coalesce manual and background refreshes to prevent a catalog stampede. */
   refreshInFlight = null;
@@ -31,9 +40,11 @@ var PricingEngine = class {
     const rows = await this.store.getAll();
     this.cache = new Map(rows.map((r) => [keyOf(r.providerId, r.modelId), r]));
     this.cacheByModel = /* @__PURE__ */ new Map();
+    this.fuzzyEntries = [];
     for (const r of rows) {
       if (r.providerId === "*" || r.providerId === "openrouter" || r.source === "openrouter") continue;
       if (!this.cacheByModel.has(r.modelId)) this.cacheByModel.set(r.modelId, r);
+      this.fuzzyEntries.push([normalizeModelId(r.modelId), r]);
     }
     this.cacheLoaded = true;
   }
@@ -53,10 +64,36 @@ var PricingEngine = class {
    *   2. wildcard provider ('*', modelId)
    *   3. model name alone (any provider) — tolerates a runtime-alias vs
    *      canonical provider-id mismatch for the same model.
+   *   4. normalized fuzzy match (last resort) — matches a bare id like
+   *      `glm-5.2` against prefixed catalog slugs like
+   *      `cloudflare/@cf/zai-org/glm-5.2`. ESTIMATE only: the matched row may
+   *      be an aggregate-route price, not the direct-vendor rate.
    */
   async getEntry(providerId, modelId) {
     await this.ensureCache();
-    return this.cache.get(keyOf(providerId, modelId)) ?? this.cache.get(keyOf("*", modelId)) ?? this.cacheByModel.get(modelId) ?? null;
+    const exact = this.cache.get(keyOf(providerId, modelId)) ?? this.cache.get(keyOf("*", modelId)) ?? this.cacheByModel.get(modelId) ?? null;
+    if (exact) return exact;
+    return this.fuzzyMatch(modelId);
+  }
+  /**
+   * Normalized substring match over the catalog's non-openrouter rows. Lowercases
+   * and strips all non-alphanumerics, then matches bidirectionally, preferring
+   * the shortest catalog key (tightest match) to reduce over-matching version
+   * variants. Returns null when nothing matches.
+   */
+  fuzzyMatch(modelId) {
+    const normQuery = normalizeModelId(modelId);
+    if (!normQuery) return null;
+    let best = null;
+    let bestLen = Infinity;
+    for (const [key, entry] of this.fuzzyEntries) {
+      if (!key || !key.includes(normQuery) && !normQuery.includes(key)) continue;
+      if (key.length < bestLen) {
+        best = entry;
+        bestLen = key.length;
+      }
+    }
+    return best;
   }
   // ===== Cost calculation =====
   /**
@@ -187,6 +224,7 @@ var PricingEngine = class {
   }
 };
 var keyOf = (providerId, modelId) => `${providerId}::${modelId}`;
+var normalizeModelId = (modelId) => (modelId ?? "").toLowerCase().replace(/[^a-z0-9]/g, "");
 var PROVIDER_ALIASES = {
   vertex_ai: "gemini",
   google: "gemini",

package/dist/{chunk-WYTEQ4Q5.cjs → chunk-GDIXXKF2.cjs}

@@ -81,6 +81,20 @@ function flattenToolResultContent(content) {
 }
 
 // src/api-converter/anthropic-to-openai.ts
+function anthropicUsageToChatUsage(usage) {
+  const input = usage.input_tokens || 0;
+  const output = usage.output_tokens || 0;
+  const cacheRead = usage.cache_read_input_tokens || 0;
+  const cacheCreation = usage.cache_creation_input_tokens || 0;
+  const promptTokens = input + cacheRead + cacheCreation;
+  const out = {
+    prompt_tokens: promptTokens,
+    completion_tokens: output,
+    total_tokens: promptTokens + output
+  };
+  if (cacheRead > 0) out.prompt_tokens_details = { cached_tokens: cacheRead };
+  return out;
+}
 function convertAnthropicToOpenAI(response) {
   const content = [];
   const toolCalls = [];
@@ -118,11 +132,7 @@ function convertAnthropicToOpenAI(response) {
         finish_reason: finishReason
       }
     ],
-    usage: {
-      prompt_tokens: response.usage.input_tokens,
-      completion_tokens: response.usage.output_tokens,
-      total_tokens: response.usage.input_tokens + response.usage.output_tokens
-    }
+    usage: anthropicUsageToChatUsage(response.usage)
   };
   if (reasoningContent) {
     result.reasoning_content = reasoningContent;
@@ -352,11 +362,12 @@ async function* convertAnthropicStreamToOpenAI(anthropicStream) {
                 delta: {},
                 finish_reason: finishReason
               }],
-              usage: event.usage ? {
-                prompt_tokens: event.usage.input_tokens || 0,
-                completion_tokens: event.usage.output_tokens || 0,
-                total_tokens: (event.usage.input_tokens || 0) + (event.usage.output_tokens || 0)
-              } : void 0
+              usage: event.usage ? anthropicUsageToChatUsage({
+                input_tokens: event.usage.input_tokens || 0,
+                output_tokens: event.usage.output_tokens || 0,
+                cache_read_input_tokens: event.usage.cache_read_input_tokens,
+                cache_creation_input_tokens: event.usage.cache_creation_input_tokens
+              }) : void 0
             })}
 
 `;