llmist 15.8.0 → 15.8.1

package/dist/index.cjs

@@ -1487,6 +1487,8 @@ var init_rate_limit = __esm({
       dailyTokens = 0;
       /** Date string (YYYY-MM-DD UTC) for daily reset tracking */
       dailyResetDate;
+      /** Count of pending reservations (for backward compatibility) */
+      pendingReservations = 0;
       constructor(config) {
         this.config = resolveRateLimitConfig(config);
         this.dailyResetDate = this.getCurrentDateUTC();
@@ -1494,13 +1496,21 @@ var init_rate_limit = __esm({
       /**
        * Record a completed request with its token usage.
        *
+       * If reserveRequest() was called before the LLM call (recommended for concurrent
+       * scenarios), the request timestamp was already recorded. Otherwise, this method
+       * will add it for backward compatibility.
+       *
        * @param inputTokens - Number of input tokens used
        * @param outputTokens - Number of output tokens generated
        */
       recordUsage(inputTokens, outputTokens) {
         const now = Date.now();
         const totalTokens = inputTokens + outputTokens;
-        this.requestTimestamps.push(now);
+        if (this.pendingReservations > 0) {
+          this.pendingReservations--;
+        } else {
+          this.requestTimestamps.push(now);
+        }
         this.tokenUsage.push({ timestamp: now, tokens: totalTokens });
         this.checkDailyReset();
         this.dailyTokens += totalTokens;
@@ -1632,6 +1642,7 @@ var init_rate_limit = __esm({
         this.tokenUsage = [];
         this.dailyTokens = 0;
         this.dailyResetDate = this.getCurrentDateUTC();
+        this.pendingReservations = 0;
       }
       /**
        * Update configuration dynamically.
@@ -1642,6 +1653,37 @@ var init_rate_limit = __esm({
       updateConfig(config) {
         this.config = resolveRateLimitConfig(config);
       }
+      /**
+       * Reserve a request slot before making an LLM call.
+       *
+       * This is critical for concurrent subagents sharing a rate limiter.
+       * Without reservation, multiple subagents checking getRequiredDelayMs()
+       * simultaneously would all see zero usage and proceed, causing rate limit errors.
+       *
+       * Call this AFTER waiting for getRequiredDelayMs() but BEFORE making the LLM call.
+       * The reservation ensures subsequent concurrent checks see the pending request.
+       *
+       * @example
+       * ```typescript
+       * // Proactive rate limiting with reservation
+       * const delay = tracker.getRequiredDelayMs();
+       * if (delay > 0) await sleep(delay);
+       *
+       * tracker.reserveRequest(); // Claim slot BEFORE making call
+       * try {
+       *   const result = await llm.call();
+       *   tracker.recordUsage(result.inputTokens, result.outputTokens);
+       * } catch (error) {
+       *   // Request already reserved; recordUsage updates token count
+       *   throw error;
+       * }
+       * ```
+       */
+      reserveRequest() {
+        const now = Date.now();
+        this.requestTimestamps.push(now);
+        this.pendingReservations++;
+      }
       // ─────────────────────────────────────────────────────────────────────────
       // Private methods
       // ─────────────────────────────────────────────────────────────────────────
@@ -13345,6 +13387,7 @@ var init_agent = __esm({
             });
             await this.sleep(throttleDelay);
           }
+          this.rateLimitTracker.reserveRequest();
         }
         return this.client.stream(llmOptions);
       }