llmist 15.8.1 → 15.10.0

package/dist/index.cjs

@@ -7972,22 +7972,24 @@ var init_openrouter_models = __esm({
         provider: "openrouter",
         modelId: "google/gemini-2.5-flash",
         displayName: "Gemini 2.5 Flash (OpenRouter)",
-        contextWindow: 1e6,
-        maxOutputTokens: 65536,
+        contextWindow: 1048576,
+        maxOutputTokens: 65535,
         pricing: {
-          input: 0.15,
-          output: 0.6
+          input: 0.3,
+          output: 2.5
         },
-        knowledgeCutoff: "2025-01",
+        knowledgeCutoff: "2025-06",
         features: {
           streaming: true,
           functionCalling: true,
           vision: true,
-          reasoning: true
+          reasoning: true,
+          structuredOutputs: true
         },
         metadata: {
           family: "Gemini 2.5",
-          notes: "Google Gemini 2.5 Flash via OpenRouter. Fast and cost-effective."
+          releaseDate: "2025-06-17",
+          notes: "Google Gemini 2.5 Flash via OpenRouter. Workhorse model for reasoning, coding, and math."
         }
       },
       {
@@ -8012,6 +8014,54 @@ var init_openrouter_models = __esm({
           notes: "Google Gemini 2.5 Pro via OpenRouter."
         }
       },
+      {
+        provider: "openrouter",
+        modelId: "google/gemini-3-pro-preview",
+        displayName: "Gemini 3 Pro Preview (OpenRouter)",
+        contextWindow: 1048576,
+        maxOutputTokens: 65536,
+        pricing: {
+          input: 2,
+          output: 12
+        },
+        knowledgeCutoff: "2025-11",
+        features: {
+          streaming: true,
+          functionCalling: true,
+          vision: true,
+          reasoning: true,
+          structuredOutputs: true
+        },
+        metadata: {
+          family: "Gemini 3",
+          releaseDate: "2025-11-18",
+          notes: "Gemini 3 Pro Preview via OpenRouter. Flagship frontier model for multimodal reasoning."
+        }
+      },
+      {
+        provider: "openrouter",
+        modelId: "google/gemini-3-flash-preview",
+        displayName: "Gemini 3 Flash Preview (OpenRouter)",
+        contextWindow: 1048576,
+        maxOutputTokens: 65535,
+        pricing: {
+          input: 0.5,
+          output: 3
+        },
+        knowledgeCutoff: "2025-12",
+        features: {
+          streaming: true,
+          functionCalling: true,
+          vision: true,
+          reasoning: true,
+          structuredOutputs: true
+        },
+        metadata: {
+          family: "Gemini 3",
+          releaseDate: "2025-12-17",
+          notes: "Gemini 3 Flash Preview via OpenRouter. High-speed thinking model for agentic workflows."
+        }
+      },
       // ============================================================
       // Meta Llama Models (via OpenRouter)
       // ============================================================
@@ -8103,6 +8153,29 @@ var init_openrouter_models = __esm({
           notes: "DeepSeek Chat via OpenRouter. Very cost-effective."
         }
       },
+      {
+        provider: "openrouter",
+        modelId: "deepseek/deepseek-v3.2",
+        displayName: "DeepSeek V3.2 (OpenRouter)",
+        contextWindow: 163840,
+        maxOutputTokens: 65536,
+        pricing: {
+          input: 0.25,
+          output: 0.38
+        },
+        knowledgeCutoff: "2025-12",
+        features: {
+          streaming: true,
+          functionCalling: true,
+          vision: false,
+          reasoning: true
+        },
+        metadata: {
+          family: "DeepSeek V3.2",
+          releaseDate: "2025-12-01",
+          notes: "DeepSeek V3.2 via OpenRouter. GPT-5 class performance with DeepSeek Sparse Attention."
+        }
+      },
       // ============================================================
       // Mistral Models (via OpenRouter)
       // ============================================================
@@ -9158,6 +9231,7 @@ var init_builder = __esm({
       textOnlyHandler;
       textWithGadgetsHandler;
       defaultGadgetTimeoutMs;
+      gadgetExecutionMode;
       gadgetOutputLimit;
       gadgetOutputLimitPercent;
       compactionConfig;
@@ -9533,6 +9607,29 @@ var init_builder = __esm({
         this.defaultGadgetTimeoutMs = timeoutMs;
         return this;
       }
+      /**
+       * Set the gadget execution mode.
+       *
+       * Controls how multiple gadgets are executed when the LLM calls them:
+       * - `'parallel'` (default): Gadgets without dependencies execute concurrently
+       * - `'sequential'`: Gadgets execute one at a time, each awaiting completion
+       *
+       * @param mode - Execution mode ('parallel' or 'sequential')
+       * @returns This builder for chaining
+       *
+       * @example
+       * ```typescript
+       * // Sequential execution for ordered file operations
+       * .withGadgetExecutionMode('sequential')
+       *
+       * // Parallel execution (default) for independent operations
+       * .withGadgetExecutionMode('parallel')
+       * ```
+       */
+      withGadgetExecutionMode(mode) {
+        this.gadgetExecutionMode = mode;
+        return this;
+      }
       /**
        * Enable or disable gadget output limiting.
        *
@@ -10013,6 +10110,7 @@ ${endPrefix}`
           textOnlyHandler: this.textOnlyHandler,
           textWithGadgetsHandler: this.textWithGadgetsHandler,
           defaultGadgetTimeoutMs: this.defaultGadgetTimeoutMs,
+          gadgetExecutionMode: this.gadgetExecutionMode,
           gadgetOutputLimit: this.gadgetOutputLimit,
           gadgetOutputLimitPercent: this.gadgetOutputLimitPercent,
           compactionConfig: this.compactionConfig,
@@ -10199,6 +10297,7 @@ ${endPrefix}`
           textOnlyHandler: this.textOnlyHandler,
           textWithGadgetsHandler: this.textWithGadgetsHandler,
           defaultGadgetTimeoutMs: this.defaultGadgetTimeoutMs,
+          gadgetExecutionMode: this.gadgetExecutionMode,
           gadgetOutputLimit: this.gadgetOutputLimit,
           gadgetOutputLimitPercent: this.gadgetOutputLimitPercent,
           compactionConfig: this.compactionConfig,
@@ -11795,6 +11894,8 @@ var init_stream_processor = __esm({
       tree;
       parentNodeId;
       baseDepth;
+      // Gadget execution mode
+      gadgetExecutionMode;
       responseText = "";
       observerFailureCount = 0;
       // Dependency tracking for gadget execution DAG
@@ -11830,6 +11931,7 @@ var init_stream_processor = __esm({
         this.tree = options.tree;
         this.parentNodeId = options.parentNodeId ?? null;
         this.baseDepth = options.baseDepth ?? 0;
+        this.gadgetExecutionMode = options.gadgetExecutionMode ?? "parallel";
         this.priorCompletedInvocations = options.priorCompletedInvocations ?? /* @__PURE__ */ new Set();
         this.priorFailedInvocations = options.priorFailedInvocations ?? /* @__PURE__ */ new Set();
         this.subagentConfig = options.subagentConfig;
@@ -12141,7 +12243,13 @@ var init_stream_processor = __esm({
           this.concurrencyQueue.set(call.gadgetName, queue);
           return;
         }
-        this.startGadgetWithConcurrencyTracking(call);
+        if (this.gadgetExecutionMode === "sequential") {
+          for await (const evt of this.executeGadgetGenerator(call)) {
+            yield evt;
+          }
+        } else {
+          this.startGadgetWithConcurrencyTracking(call);
+        }
       }
       /**
        * Get the effective concurrency limit for a gadget.
@@ -12593,25 +12701,37 @@ var init_stream_processor = __esm({
             progress = true;
           }
           if (readyToExecute.length > 0) {
-            this.logger.debug("Executing ready gadgets in parallel", {
-              count: readyToExecute.length,
-              invocationIds: readyToExecute.map((c) => c.invocationId)
-            });
             for (const call of readyToExecute) {
               this.gadgetsAwaitingDependencies.delete(call.invocationId);
             }
-            const eventSets = await Promise.all(
-              readyToExecute.map(async (call) => {
-                const events = [];
+            if (this.gadgetExecutionMode === "sequential") {
+              this.logger.debug("Executing ready gadgets sequentially", {
+                count: readyToExecute.length,
+                invocationIds: readyToExecute.map((c) => c.invocationId)
+              });
+              for (const call of readyToExecute) {
                 for await (const evt of this.executeGadgetGenerator(call)) {
-                  events.push(evt);
+                  yield evt;
+                }
+              }
+            } else {
+              this.logger.debug("Executing ready gadgets in parallel", {
+                count: readyToExecute.length,
+                invocationIds: readyToExecute.map((c) => c.invocationId)
+              });
+              const eventSets = await Promise.all(
+                readyToExecute.map(async (call) => {
+                  const events = [];
+                  for await (const evt of this.executeGadgetGenerator(call)) {
+                    events.push(evt);
+                  }
+                  return events;
+                })
+              );
+              for (const events of eventSets) {
+                for (const evt of events) {
+                  yield evt;
                 }
-                return events;
-              })
-            );
-            for (const events of eventSets) {
-              for (const evt of events) {
-                yield evt;
               }
             }
             progress = true;
@@ -12765,6 +12885,7 @@ var init_agent = __esm({
       textOnlyHandler;
       textWithGadgetsHandler;
       defaultGadgetTimeoutMs;
+      gadgetExecutionMode;
       defaultMaxTokens;
       hasUserPrompt;
       // Gadget output limiting
@@ -12820,6 +12941,7 @@ var init_agent = __esm({
         this.textOnlyHandler = options.textOnlyHandler ?? "terminate";
         this.textWithGadgetsHandler = options.textWithGadgetsHandler;
         this.defaultGadgetTimeoutMs = options.defaultGadgetTimeoutMs;
+        this.gadgetExecutionMode = options.gadgetExecutionMode ?? "parallel";
         this.defaultMaxTokens = this.resolveMaxTokensFromCatalog(options.model);
         this.outputLimitEnabled = options.gadgetOutputLimit ?? DEFAULT_GADGET_OUTPUT_LIMIT;
         this.outputStore = new GadgetOutputStore();
@@ -13151,6 +13273,7 @@ var init_agent = __esm({
                     logger: this.logger.getSubLogger({ name: "stream-processor" }),
                     requestHumanInput: this.requestHumanInput,
                     defaultGadgetTimeoutMs: this.defaultGadgetTimeoutMs,
+                    gadgetExecutionMode: this.gadgetExecutionMode,
                     client: this.client,
                     mediaStore: this.mediaStore,
                     agentConfig: this.agentContextConfig,