@compilr-dev/agents 0.3.7 → 0.3.8

package/dist/providers/fireworks.d.ts

@@ -57,6 +57,14 @@ export declare class FireworksProvider extends OpenAICompatibleProvider {
      * Fireworks AI uses standard OpenAI body format
      */
     protected buildProviderSpecificBody(_options?: ChatOptions): Record<string, unknown>;
+    /**
+     * Extract cache statistics from Fireworks response headers.
+     * Fireworks returns cache stats in headers rather than the JSON body.
+     * @see https://docs.fireworks.ai/guides/prompt-caching
+     */
+    protected extractCacheStatsFromHeaders(headers: Headers): {
+        cacheReadTokens?: number;
+    };
     /**
      * Map HTTP errors with Fireworks AI-specific messages
      */

package/dist/providers/fireworks.js

@@ -65,6 +65,21 @@ export class FireworksProvider extends OpenAICompatibleProvider {
     buildProviderSpecificBody(_options) {
         return {};
     }
+    /**
+     * Extract cache statistics from Fireworks response headers.
+     * Fireworks returns cache stats in headers rather than the JSON body.
+     * @see https://docs.fireworks.ai/guides/prompt-caching
+     */
+    extractCacheStatsFromHeaders(headers) {
+        const cachedTokens = headers.get('fireworks-cached-prompt-tokens');
+        if (cachedTokens) {
+            const parsed = parseInt(cachedTokens, 10);
+            if (!isNaN(parsed) && parsed > 0) {
+                return { cacheReadTokens: parsed };
+            }
+        }
+        return {};
+    }
     /**
      * Map HTTP errors with Fireworks AI-specific messages
      */

package/dist/providers/openai-compatible.d.ts

@@ -149,6 +149,15 @@ export declare abstract class OpenAICompatibleProvider implements LLMProvider {
      * @returns ProviderError with appropriate message
      */
     protected abstract mapConnectionError(error: Error): ProviderError;
+    /**
+     * Extract cache statistics from response headers.
+     * Override in subclasses for providers that return cache stats in headers (e.g., Fireworks).
+     * @param headers Response headers
+     * @returns Partial LLMUsage with cache stats
+     */
+    protected extractCacheStatsFromHeaders(_headers: Headers): {
+        cacheReadTokens?: number;
+    };
     /**
      * Stream chat completion from the provider
      *

package/dist/providers/openai-compatible.js

@@ -42,6 +42,15 @@ export class OpenAICompatibleProvider {
         this.defaultMaxTokens = config.maxTokens ?? DEFAULT_MAX_TOKENS;
         this.timeout = config.timeout ?? DEFAULT_TIMEOUT;
     }
+    /**
+     * Extract cache statistics from response headers.
+     * Override in subclasses for providers that return cache stats in headers (e.g., Fireworks).
+     * @param headers Response headers
+     * @returns Partial LLMUsage with cache stats
+     */
+    extractCacheStatsFromHeaders(_headers) {
+        return {};
+    }
     // ==================== SHARED IMPLEMENTATION ====================
     /**
      * Stream chat completion from the provider
@@ -106,6 +115,8 @@ export class OpenAICompatibleProvider {
                 const errorBody = await response.text();
                 throw this.mapHttpError(response.status, errorBody, model);
             }
+            // Extract cache stats from headers (for providers like Fireworks)
+            const headerCacheStats = this.extractCacheStatsFromHeaders(response.headers);
             const reader = response.body?.getReader();
             if (!reader) {
                 throw new ProviderError('No response body', this.name);
@@ -153,12 +164,12 @@ export class OpenAICompatibleProvider {
                     }
                 }
             }
-            // Yield done chunk with usage
+            // Yield done chunk with usage (merge header-based cache stats)
             yield {
                 type: 'done',
                 usage: usage
-                    ? { ...usage, debugPayload }
-                    : { inputTokens: 0, outputTokens: 0, debugPayload },
+                    ? { ...usage, ...headerCacheStats, debugPayload }
+                    : { inputTokens: 0, outputTokens: 0, ...headerCacheStats, debugPayload },
             };
         }
         catch (error) {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@compilr-dev/agents",
-  "version": "0.3.7",
+  "version": "0.3.8",
   "description": "Lightweight multi-LLM agent library for building CLI AI assistants",
   "type": "module",
   "main": "dist/index.js",