@blockrun/llm 3.24.4 → 3.24.5

package/dist/index.cjs

@@ -5089,7 +5089,7 @@ function getCostSummary() {
 }
 
 // src/version.ts
-var SDK_VERSION = "3.24.4";
+var SDK_VERSION = "3.24.5";
 var USER_AGENT = `blockrun-ts/${SDK_VERSION}`;
 
 // src/client.ts
@@ -5157,8 +5157,15 @@ var LLMClient = class _LLMClient {
   modelPricingCache = null;
   modelPricingPromise = null;
   // Pre-auth cache: avoids the 402 round-trip on repeat requests to the same model.
-  // Key = "endpoint:model", value = cached payment header + timestamp.
+  // Key = "endpoint:model", value = cached payment header + timestamp + the
+  // length of the request body it was quoted for.
   // TTL: 1 hour — pre-auth quotes are stable server-side on that horizon.
+  //
+  // A quote prices ONE request (input size + max_tokens), and exact settles
+  // the whole signed amount, so reusing it for a smaller request overpaid and
+  // for a larger one underpaid. The gateway refuses both with a 402 (an
+  // over-signed payment since blockrun#874), so a reuse is only attempted for
+  // a body the same size as the one that was quoted.
   preAuthCache = /* @__PURE__ */ new Map();
   static PRE_AUTH_TTL_MS = 36e5;
   /**
@@ -5765,9 +5772,10 @@ var LLMClient = class _LLMClient {
     if (options?.responseFormat !== void 0) body.response_format = options.responseFormat;
     if (options?.stop !== void 0) body.stop = options.stop;
     const cacheKey2 = `/v1/chat/completions:${requestModel}`;
+    const requestBody = JSON.stringify(body);
     const cached = this.preAuthCache.get(cacheKey2);
     const now = Date.now();
-    if (cached && now - cached.cachedAt < _LLMClient.PRE_AUTH_TTL_MS) {
+    if (cached && now - cached.cachedAt < _LLMClient.PRE_AUTH_TTL_MS && cached.bodyLength === requestBody.length) {
       let signed;
       try {
         signed = await this.signPayment(cached.paymentHeader);
@@ -5836,7 +5844,7 @@ var LLMClient = class _LLMClient {
       }
     }
     if (!paymentHeader) throw new PaymentError("402 response but no payment requirements found");
-    this.preAuthCache.set(cacheKey2, { paymentHeader, cachedAt: now });
+    this.preAuthCache.set(cacheKey2, { paymentHeader, cachedAt: now, bodyLength: requestBody.length });
     const { paymentPayload, costUsd } = await this.signPayment(paymentHeader);
     const streamResp = await this.fetchWithTimeout(url, {
       method: "POST",

package/dist/index.js

@@ -4958,7 +4958,7 @@ function getCostSummary() {
 }
 
 // src/version.ts
-var SDK_VERSION = "3.24.4";
+var SDK_VERSION = "3.24.5";
 var USER_AGENT = `blockrun-ts/${SDK_VERSION}`;
 
 // src/client.ts
@@ -5026,8 +5026,15 @@ var LLMClient = class _LLMClient {
   modelPricingCache = null;
   modelPricingPromise = null;
   // Pre-auth cache: avoids the 402 round-trip on repeat requests to the same model.
-  // Key = "endpoint:model", value = cached payment header + timestamp.
+  // Key = "endpoint:model", value = cached payment header + timestamp + the
+  // length of the request body it was quoted for.
   // TTL: 1 hour — pre-auth quotes are stable server-side on that horizon.
+  //
+  // A quote prices ONE request (input size + max_tokens), and exact settles
+  // the whole signed amount, so reusing it for a smaller request overpaid and
+  // for a larger one underpaid. The gateway refuses both with a 402 (an
+  // over-signed payment since blockrun#874), so a reuse is only attempted for
+  // a body the same size as the one that was quoted.
   preAuthCache = /* @__PURE__ */ new Map();
   static PRE_AUTH_TTL_MS = 36e5;
   /**
@@ -5634,9 +5641,10 @@ var LLMClient = class _LLMClient {
     if (options?.responseFormat !== void 0) body.response_format = options.responseFormat;
     if (options?.stop !== void 0) body.stop = options.stop;
     const cacheKey2 = `/v1/chat/completions:${requestModel}`;
+    const requestBody = JSON.stringify(body);
     const cached = this.preAuthCache.get(cacheKey2);
     const now = Date.now();
-    if (cached && now - cached.cachedAt < _LLMClient.PRE_AUTH_TTL_MS) {
+    if (cached && now - cached.cachedAt < _LLMClient.PRE_AUTH_TTL_MS && cached.bodyLength === requestBody.length) {
       let signed;
       try {
         signed = await this.signPayment(cached.paymentHeader);
@@ -5705,7 +5713,7 @@ var LLMClient = class _LLMClient {
       }
     }
     if (!paymentHeader) throw new PaymentError("402 response but no payment requirements found");
-    this.preAuthCache.set(cacheKey2, { paymentHeader, cachedAt: now });
+    this.preAuthCache.set(cacheKey2, { paymentHeader, cachedAt: now, bodyLength: requestBody.length });
     const { paymentPayload, costUsd } = await this.signPayment(paymentHeader);
     const streamResp = await this.fetchWithTimeout(url, {
       method: "POST",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@blockrun/llm",
-  "version": "3.24.4",
+  "version": "3.24.5",
   "type": "module",
   "description": "TypeScript SDK for BlockRun - every frontier model behind one OpenAI-compatible client. Pay with a BlockRun API key or per-call in USDC. Smart routing picks the cheapest capable model. No rate limits.",
   "main": "dist/index.cjs",