@blockrun/clawrouter 0.9.3 → 0.9.5

package/dist/index.d.ts

@@ -509,8 +509,8 @@ type ProxyOptions = {
      */
     sessionConfig?: Partial<SessionConfig>;
     /**
-     * Auto-compress large requests to fit within API limits.
-     * When enabled, requests approaching 200KB are automatically compressed using
+     * Auto-compress large requests to reduce network usage.
+     * When enabled, requests are automatically compressed using
      * LLM-safe context compression (15-40% reduction).
      * Default: true
      */
@@ -521,11 +521,6 @@ type ProxyOptions = {
      * Set to 0 to compress all requests.
      */
     compressionThresholdKB?: number;
-    /**
-     * Maximum request size in KB after compression (default: 200).
-     * Hard limit enforced by BlockRun API.
-     */
-    maxRequestSizeKB?: number;
     onReady?: (port: number) => void;
     onError?: (error: Error) => void;
     onPayment?: (info: {
@@ -586,7 +581,12 @@ declare const blockrunProvider: ProviderPlugin;
 declare const MODEL_ALIASES: Record<string, string>;
 /**
  * Resolve a model alias to its full model ID.
- * Returns the original model if not an alias.
+ * Also strips "blockrun/" prefix for direct model paths.
+ * Examples:
+ *   - "claude" -> "anthropic/claude-sonnet-4" (alias)
+ *   - "blockrun/claude" -> "anthropic/claude-sonnet-4" (alias with prefix)
+ *   - "blockrun/anthropic/claude-sonnet-4" -> "anthropic/claude-sonnet-4" (prefix stripped)
+ *   - "openai/gpt-4o" -> "openai/gpt-4o" (unchanged)
  */
 declare function resolveModelAlias(model: string): string;
 type BlockRunModel = {
@@ -678,7 +678,8 @@ declare class RequestDeduplicator {
     markInflight(key: string): void;
     /** Complete an in-flight request — cache result and notify waiters. */
     complete(key: string, result: CachedResponse): void;
-    /** Remove an in-flight entry on error (don't cache failures). */
+    /** Remove an in-flight entry on error (don't cache failures).
+     *  Also rejects any waiters so they can retry independently. */
     removeInflight(key: string): void;
     /** Prune expired completed entries. */
     private prune;

package/dist/index.js

@@ -37,6 +37,7 @@ function resolveModelAlias(model) {
     const withoutPrefix = normalized.slice("blockrun/".length);
     const resolvedWithoutPrefix = MODEL_ALIASES[withoutPrefix];
     if (resolvedWithoutPrefix) return resolvedWithoutPrefix;
+    return withoutPrefix;
   }
   return model;
 }
@@ -2082,26 +2083,14 @@ var RequestDeduplicator = class {
   getInflight(key) {
     const entry = this.inflight.get(key);
     if (!entry) return void 0;
-    const promise = new Promise((resolve) => {
-      entry.waiters.push(
-        new Promise((r) => {
-          const orig = entry.resolve;
-          entry.resolve = (result) => {
-            orig(result);
-            resolve(result);
-            r(result);
-          };
-        })
-      );
+    return new Promise((resolve) => {
+      entry.resolvers.push(resolve);
     });
-    return promise;
   }
   /** Mark a request as in-flight. */
   markInflight(key) {
     this.inflight.set(key, {
-      resolve: () => {
-      },
-      waiters: []
+      resolvers: []
     });
   }
   /** Complete an in-flight request — cache result and notify waiters. */
@@ -2111,14 +2100,33 @@ var RequestDeduplicator = class {
     }
     const entry = this.inflight.get(key);
     if (entry) {
-      entry.resolve(result);
+      for (const resolve of entry.resolvers) {
+        resolve(result);
+      }
       this.inflight.delete(key);
     }
     this.prune();
   }
-  /** Remove an in-flight entry on error (don't cache failures). */
+  /** Remove an in-flight entry on error (don't cache failures).
+   *  Also rejects any waiters so they can retry independently. */
   removeInflight(key) {
-    this.inflight.delete(key);
+    const entry = this.inflight.get(key);
+    if (entry) {
+      const errorBody = Buffer.from(
+        JSON.stringify({
+          error: { message: "Original request failed, please retry", type: "dedup_origin_failed" }
+        })
+      );
+      for (const resolve of entry.resolvers) {
+        resolve({
+          status: 503,
+          headers: { "content-type": "application/json" },
+          body: errorBody,
+          completedAt: Date.now()
+        });
+      }
+      this.inflight.delete(key);
+    }
   }
   /** Prune expired completed entries. */
   prune() {
@@ -4026,7 +4034,6 @@ async function proxyRequest(req, res, apiBase, payFetch, options, routerOpts, de
   }
   const autoCompress = options.autoCompressRequests ?? true;
   const compressionThreshold = options.compressionThresholdKB ?? 180;
-  const sizeLimit = options.maxRequestSizeKB ?? 200;
   const requestSizeKB = Math.ceil(body.length / 1024);
   if (autoCompress && requestSizeKB > compressionThreshold) {
     try {
@@ -4068,21 +4075,6 @@ async function proxyRequest(req, res, apiBase, payFetch, options, routerOpts, de
         );
         parsed.messages = compressionResult.messages;
         body = Buffer.from(JSON.stringify(parsed));
-        if (compressedSizeKB > sizeLimit) {
-          const errorMsg = {
-            error: {
-              message: `Request size ${compressedSizeKB}KB still exceeds limit after compression (original: ${requestSizeKB}KB). Please reduce context size.`,
-              type: "request_too_large",
-              original_size_kb: requestSizeKB,
-              compressed_size_kb: compressedSizeKB,
-              limit_kb: sizeLimit,
-              help: "Try: 1) Remove old messages from history, 2) Summarize large tool results, 3) Use direct API for very large contexts"
-            }
-          };
-          res.writeHead(413, { "Content-Type": "application/json" });
-          res.end(JSON.stringify(errorMsg));
-          return;
-        }
       }
     } catch (err) {
       console.warn(
@@ -4090,21 +4082,6 @@ async function proxyRequest(req, res, apiBase, payFetch, options, routerOpts, de
       );
     }
   }
-  const finalSizeKB = Math.ceil(body.length / 1024);
-  if (finalSizeKB > sizeLimit) {
-    const errorMsg = {
-      error: {
-        message: `Request size ${finalSizeKB}KB exceeds limit ${sizeLimit}KB. Please reduce context size.`,
-        type: "request_too_large",
-        size_kb: finalSizeKB,
-        limit_kb: sizeLimit,
-        help: "Try: 1) Remove old messages from history, 2) Summarize large tool results, 3) Enable compression (autoCompressRequests: true)"
-      }
-    };
-    res.writeHead(413, { "Content-Type": "application/json" });
-    res.end(JSON.stringify(errorMsg));
-    return;
-  }
   const dedupKey = RequestDeduplicator.hash(body);
   const cached = deduplicator.getCached(dedupKey);
   if (cached) {