fix: refresh model preset settings on resume (#1052)

Co-authored-by: Letta <noreply@letta.com>
Co-authored-by: letta-code <248085862+letta-code@users.noreply.github.com>
Co-authored-by: jnjpng <jnjpng@users.noreply.github.com>

src/agent/model.ts

@@ -2,6 +2,7 @@
  * Model resolution and handling utilities
  */
 import modelsData from "../models.json";
+import { OPENAI_CODEX_PROVIDER_NAME } from "../providers/openai-codex-provider";
 
 export const models = modelsData;
 
@@ -186,6 +187,68 @@ export function getModelUpdateArgs(
   return modelInfo?.updateArgs;
 }
 
+type AgentModelSnapshot = {
+  model?: string | null;
+  llm_config?: {
+    model?: string | null;
+    model_endpoint_type?: string | null;
+    reasoning_effort?: string | null;
+    enable_reasoner?: boolean | null;
+  } | null;
+};
+
+/**
+ * Resolve the current model preset + updateArgs for an existing agent.
+ *
+ * Used during startup/resume refresh to re-apply only preset-defined fields
+ * (without requiring an explicit --model flag).
+ */
+export function getModelPresetUpdateForAgent(
+  agent: AgentModelSnapshot,
+): { modelHandle: string; updateArgs: Record<string, unknown> } | null {
+  const directHandle =
+    typeof agent.model === "string" && agent.model.length > 0
+      ? agent.model
+      : null;
+
+  const endpointType = agent.llm_config?.model_endpoint_type;
+  const llmModel = agent.llm_config?.model;
+  const llmDerivedHandle =
+    typeof endpointType === "string" &&
+    endpointType.length > 0 &&
+    typeof llmModel === "string" &&
+    llmModel.length > 0
+      ? `${
+          endpointType === "chatgpt_oauth"
+            ? OPENAI_CODEX_PROVIDER_NAME
+            : endpointType
+        }/${llmModel}`
+      : typeof llmModel === "string" && llmModel.includes("/")
+        ? llmModel
+        : null;
+
+  const modelHandle = directHandle ?? llmDerivedHandle;
+  if (!modelHandle) return null;
+
+  const modelInfo = getModelInfoForLlmConfig(modelHandle, {
+    reasoning_effort: agent.llm_config?.reasoning_effort ?? null,
+    enable_reasoner: agent.llm_config?.enable_reasoner ?? null,
+  });
+
+  const updateArgs =
+    (modelInfo?.updateArgs as Record<string, unknown> | undefined) ??
+    getModelUpdateArgs(modelHandle);
+
+  if (!updateArgs || Object.keys(updateArgs).length === 0) {
+    return null;
+  }
+
+  return {
+    modelHandle: modelInfo?.handle ?? modelHandle,
+    updateArgs,
+  };
+}
+
 /**
  * Find a model entry by handle with fuzzy matching support
  * @param handle - The full model handle

src/headless.ts

@@ -878,23 +878,56 @@ export async function handleHeadlessCommand(
     (!forceNew && !fromAfFile)
   );
 
-  // If resuming and a model or system prompt was specified, apply those changes
+  // If resuming, always refresh model settings from presets to keep
-  if (isResumingAgent && (model || systemPromptPreset)) {
+  // preset-derived fields in sync, then apply optional command-line
+  // overrides (model/system prompt).
+  if (isResumingAgent) {
+    const { updateAgentLLMConfig } = await import("./agent/modify");
+
     if (model) {
       const { resolveModel } = await import("./agent/model");
       const modelHandle = resolveModel(model);
-      if (!modelHandle) {
+      if (typeof modelHandle !== "string") {
         console.error(`Error: Invalid model "${model}"`);
         process.exit(1);
       }
 
       // Always apply model update - different model IDs can share the same
       // handle but have different settings (e.g., gpt-5.2-medium vs gpt-5.2-xhigh)
-      const { updateAgentLLMConfig } = await import("./agent/modify");
       const updateArgs = getModelUpdateArgs(model);
       await updateAgentLLMConfig(agent.id, modelHandle, updateArgs);
       // Refresh agent state after model update
       agent = await client.agents.retrieve(agent.id);
+    } else {
+      const { getModelPresetUpdateForAgent } = await import("./agent/model");
+      const presetRefresh = getModelPresetUpdateForAgent(agent);
+      if (presetRefresh) {
+        // Resume preset refresh is intentionally scoped for now.
+        // We only force-refresh max_output_tokens + parallel_tool_calls.
+        // Other preset fields available in models.json (for example:
+        // context_window, reasoning_effort, enable_reasoner,
+        // max_reasoning_tokens, verbosity, temperature,
+        // thinking_budget) are intentionally not auto-applied yet.
+        const resumeRefreshUpdateArgs: Record<string, unknown> = {};
+        if (typeof presetRefresh.updateArgs.max_output_tokens === "number") {
+          resumeRefreshUpdateArgs.max_output_tokens =
+            presetRefresh.updateArgs.max_output_tokens;
+        }
+        if (typeof presetRefresh.updateArgs.parallel_tool_calls === "boolean") {
+          resumeRefreshUpdateArgs.parallel_tool_calls =
+            presetRefresh.updateArgs.parallel_tool_calls;
+        }
+
+        if (Object.keys(resumeRefreshUpdateArgs).length > 0) {
+          await updateAgentLLMConfig(
+            agent.id,
+            presetRefresh.modelHandle,
+            resumeRefreshUpdateArgs,
+          );
+          // Refresh agent state after model update
+          agent = await client.agents.retrieve(agent.id);
+        }
+      }
     }
 
     if (systemPromptPreset) {

src/index.ts

@@ -1789,8 +1789,12 @@ async function main(): Promise<void> {
         );
         setIsResumingSession(resuming);
 
-        // If resuming and a model or system prompt was specified, apply those changes
+        // If resuming, always refresh model settings from presets to keep
-        if (resuming && (model || systemPromptPreset)) {
+        // preset-derived fields in sync, then apply optional command-line
+        // overrides (model/system prompt).
+        if (resuming) {
+          const { updateAgentLLMConfig } = await import("./agent/modify");
+
           if (model) {
             const { resolveModel, getModelUpdateArgs } = await import(
               "./agent/model"
@@ -1803,11 +1807,47 @@ async function main(): Promise<void> {
 
             // Always apply model update - different model IDs can share the same
             // handle but have different settings (e.g., gpt-5.2-medium vs gpt-5.2-xhigh)
-            const { updateAgentLLMConfig } = await import("./agent/modify");
             const updateArgs = getModelUpdateArgs(model);
             await updateAgentLLMConfig(agent.id, modelHandle, updateArgs);
             // Refresh agent state after model update
             agent = await client.agents.retrieve(agent.id);
+          } else {
+            const { getModelPresetUpdateForAgent } = await import(
+              "./agent/model"
+            );
+            const presetRefresh = getModelPresetUpdateForAgent(agent);
+            if (presetRefresh) {
+              // Resume preset refresh is intentionally scoped for now.
+              // We only force-refresh max_output_tokens + parallel_tool_calls.
+              // Other preset fields available in models.json (for example:
+              // context_window, reasoning_effort, enable_reasoner,
+              // max_reasoning_tokens, verbosity, temperature,
+              // thinking_budget) are intentionally not auto-applied yet.
+              const resumeRefreshUpdateArgs: Record<string, unknown> = {};
+              if (
+                typeof presetRefresh.updateArgs.max_output_tokens === "number"
+              ) {
+                resumeRefreshUpdateArgs.max_output_tokens =
+                  presetRefresh.updateArgs.max_output_tokens;
+              }
+              if (
+                typeof presetRefresh.updateArgs.parallel_tool_calls ===
+                "boolean"
+              ) {
+                resumeRefreshUpdateArgs.parallel_tool_calls =
+                  presetRefresh.updateArgs.parallel_tool_calls;
+              }
+
+              if (Object.keys(resumeRefreshUpdateArgs).length > 0) {
+                await updateAgentLLMConfig(
+                  agent.id,
+                  presetRefresh.modelHandle,
+                  resumeRefreshUpdateArgs,
+                );
+                // Refresh agent state after model update
+                agent = await client.agents.retrieve(agent.id);
+              }
+            }
           }
 
           if (systemPromptPreset) {

src/models.json

@@ -10,7 +10,8 @@
       "context_window": 200000,
       "max_output_tokens": 128000,
       "reasoning_effort": "high",
-      "enable_reasoner": true
+      "enable_reasoner": true,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -22,7 +23,8 @@
       "context_window": 1000000,
       "max_output_tokens": 128000,
       "reasoning_effort": "high",
-      "enable_reasoner": true
+      "enable_reasoner": true,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -34,7 +36,8 @@
       "context_window": 200000,
       "max_output_tokens": 128000,
       "reasoning_effort": "none",
-      "enable_reasoner": false
+      "enable_reasoner": false,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -47,7 +50,8 @@
       "max_output_tokens": 128000,
       "reasoning_effort": "low",
       "enable_reasoner": true,
-      "max_reasoning_tokens": 4000
+      "max_reasoning_tokens": 4000,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -60,7 +64,8 @@
       "max_output_tokens": 128000,
       "reasoning_effort": "medium",
       "enable_reasoner": true,
-      "max_reasoning_tokens": 12000
+      "max_reasoning_tokens": 12000,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -73,7 +78,8 @@
       "max_output_tokens": 128000,
       "reasoning_effort": "xhigh",
       "enable_reasoner": true,
-      "max_reasoning_tokens": 31999
+      "max_reasoning_tokens": 31999,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -84,7 +90,8 @@
     "updateArgs": {
       "context_window": 180000,
       "max_output_tokens": 64000,
-      "max_reasoning_tokens": 31999
+      "max_reasoning_tokens": 31999,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -95,7 +102,8 @@
     "updateArgs": {
       "enable_reasoner": false,
       "context_window": 180000,
-      "max_output_tokens": 64000
+      "max_output_tokens": 64000,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -108,7 +116,8 @@
       "context_window": 200000,
       "max_output_tokens": 128000,
       "reasoning_effort": "high",
-      "enable_reasoner": true
+      "enable_reasoner": true,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -120,7 +129,8 @@
       "context_window": 200000,
       "max_output_tokens": 128000,
       "reasoning_effort": "none",
-      "enable_reasoner": false
+      "enable_reasoner": false,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -133,7 +143,8 @@
       "max_output_tokens": 128000,
       "reasoning_effort": "low",
       "enable_reasoner": true,
-      "max_reasoning_tokens": 4000
+      "max_reasoning_tokens": 4000,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -146,7 +157,8 @@
       "max_output_tokens": 128000,
       "reasoning_effort": "medium",
       "enable_reasoner": true,
-      "max_reasoning_tokens": 12000
+      "max_reasoning_tokens": 12000,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -159,7 +171,8 @@
       "max_output_tokens": 128000,
       "reasoning_effort": "xhigh",
       "enable_reasoner": true,
-      "max_reasoning_tokens": 31999
+      "max_reasoning_tokens": 31999,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -170,7 +183,8 @@
     "updateArgs": {
       "context_window": 180000,
       "max_output_tokens": 64000,
-      "max_reasoning_tokens": 31999
+      "max_reasoning_tokens": 31999,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -183,7 +197,8 @@
     "updateArgs": {
       "context_window": 180000,
       "max_output_tokens": 64000,
-      "max_reasoning_tokens": 31999
+      "max_reasoning_tokens": 31999,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -194,7 +209,8 @@
     "isFeatured": true,
     "updateArgs": {
       "context_window": 180000,
-      "max_output_tokens": 64000
+      "max_output_tokens": 64000,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -206,7 +222,8 @@
       "reasoning_effort": "none",
       "verbosity": "low",
       "context_window": 272000,
-      "max_output_tokens": 128000
+      "max_output_tokens": 128000,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -218,7 +235,8 @@
       "reasoning_effort": "low",
       "verbosity": "low",
       "context_window": 272000,
-      "max_output_tokens": 128000
+      "max_output_tokens": 128000,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -230,7 +248,8 @@
       "reasoning_effort": "medium",
       "verbosity": "low",
       "context_window": 272000,
-      "max_output_tokens": 128000
+      "max_output_tokens": 128000,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -243,7 +262,8 @@
       "reasoning_effort": "high",
       "verbosity": "low",
       "context_window": 272000,
-      "max_output_tokens": 128000
+      "max_output_tokens": 128000,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -255,7 +275,8 @@
       "reasoning_effort": "xhigh",
       "verbosity": "low",
       "context_window": 272000,
-      "max_output_tokens": 128000
+      "max_output_tokens": 128000,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -267,7 +288,8 @@
       "reasoning_effort": "medium",
       "verbosity": "medium",
       "context_window": 272000,
-      "max_output_tokens": 128000
+      "max_output_tokens": 128000,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -279,7 +301,8 @@
       "reasoning_effort": "high",
       "verbosity": "medium",
       "context_window": 272000,
-      "max_output_tokens": 128000
+      "max_output_tokens": 128000,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -291,7 +314,8 @@
       "reasoning_effort": "medium",
       "verbosity": "medium",
       "context_window": 272000,
-      "max_output_tokens": 128000
+      "max_output_tokens": 128000,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -303,7 +327,8 @@
       "reasoning_effort": "high",
       "verbosity": "medium",
       "context_window": 272000,
-      "max_output_tokens": 128000
+      "max_output_tokens": 128000,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -315,7 +340,8 @@
       "reasoning_effort": "medium",
       "verbosity": "medium",
       "context_window": 272000,
-      "max_output_tokens": 128000
+      "max_output_tokens": 128000,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -327,7 +353,8 @@
       "reasoning_effort": "high",
       "verbosity": "medium",
       "context_window": 272000,
-      "max_output_tokens": 128000
+      "max_output_tokens": 128000,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -339,7 +366,8 @@
       "reasoning_effort": "medium",
       "verbosity": "medium",
       "context_window": 272000,
-      "max_output_tokens": 128000
+      "max_output_tokens": 128000,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -351,7 +379,8 @@
       "reasoning_effort": "high",
       "verbosity": "medium",
       "context_window": 272000,
-      "max_output_tokens": 128000
+      "max_output_tokens": 128000,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -363,7 +392,8 @@
       "reasoning_effort": "xhigh",
       "verbosity": "medium",
       "context_window": 272000,
-      "max_output_tokens": 128000
+      "max_output_tokens": 128000,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -375,7 +405,8 @@
       "reasoning_effort": "medium",
       "verbosity": "medium",
       "context_window": 272000,
-      "max_output_tokens": 128000
+      "max_output_tokens": 128000,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -387,7 +418,8 @@
       "reasoning_effort": "none",
       "verbosity": "medium",
       "context_window": 272000,
-      "max_output_tokens": 128000
+      "max_output_tokens": 128000,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -399,7 +431,8 @@
       "reasoning_effort": "low",
       "verbosity": "medium",
       "context_window": 272000,
-      "max_output_tokens": 128000
+      "max_output_tokens": 128000,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -412,7 +445,8 @@
       "reasoning_effort": "medium",
       "verbosity": "medium",
       "context_window": 272000,
-      "max_output_tokens": 128000
+      "max_output_tokens": 128000,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -424,7 +458,8 @@
       "reasoning_effort": "high",
       "verbosity": "medium",
       "context_window": 272000,
-      "max_output_tokens": 128000
+      "max_output_tokens": 128000,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -436,7 +471,8 @@
       "reasoning_effort": "xhigh",
       "verbosity": "medium",
       "context_window": 272000,
-      "max_output_tokens": 128000
+      "max_output_tokens": 128000,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -448,7 +484,8 @@
       "reasoning_effort": "none",
       "verbosity": "medium",
       "context_window": 272000,
-      "max_output_tokens": 128000
+      "max_output_tokens": 128000,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -460,7 +497,8 @@
       "reasoning_effort": "low",
       "verbosity": "medium",
       "context_window": 272000,
-      "max_output_tokens": 128000
+      "max_output_tokens": 128000,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -473,7 +511,8 @@
       "reasoning_effort": "medium",
       "verbosity": "medium",
       "context_window": 272000,
-      "max_output_tokens": 128000
+      "max_output_tokens": 128000,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -485,7 +524,8 @@
       "reasoning_effort": "high",
       "verbosity": "medium",
       "context_window": 272000,
-      "max_output_tokens": 128000
+      "max_output_tokens": 128000,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -497,7 +537,8 @@
       "reasoning_effort": "xhigh",
       "verbosity": "medium",
       "context_window": 272000,
-      "max_output_tokens": 128000
+      "max_output_tokens": 128000,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -509,7 +550,8 @@
       "reasoning_effort": "none",
       "verbosity": "medium",
       "context_window": 272000,
-      "max_output_tokens": 128000
+      "max_output_tokens": 128000,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -521,7 +563,8 @@
       "reasoning_effort": "low",
       "verbosity": "medium",
       "context_window": 272000,
-      "max_output_tokens": 128000
+      "max_output_tokens": 128000,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -533,7 +576,8 @@
       "reasoning_effort": "medium",
       "verbosity": "medium",
       "context_window": 272000,
-      "max_output_tokens": 128000
+      "max_output_tokens": 128000,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -545,7 +589,8 @@
       "reasoning_effort": "high",
       "verbosity": "medium",
       "context_window": 272000,
-      "max_output_tokens": 128000
+      "max_output_tokens": 128000,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -557,7 +602,8 @@
       "reasoning_effort": "none",
       "verbosity": "medium",
       "context_window": 272000,
-      "max_output_tokens": 128000
+      "max_output_tokens": 128000,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -570,7 +616,8 @@
       "reasoning_effort": "medium",
       "verbosity": "medium",
       "context_window": 272000,
-      "max_output_tokens": 128000
+      "max_output_tokens": 128000,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -582,7 +629,8 @@
       "reasoning_effort": "high",
       "verbosity": "medium",
       "context_window": 272000,
-      "max_output_tokens": 128000
+      "max_output_tokens": 128000,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -594,7 +642,8 @@
       "reasoning_effort": "medium",
       "verbosity": "medium",
       "context_window": 272000,
-      "max_output_tokens": 128000
+      "max_output_tokens": 128000,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -606,7 +655,8 @@
       "reasoning_effort": "high",
       "verbosity": "medium",
       "context_window": 272000,
-      "max_output_tokens": 128000
+      "max_output_tokens": 128000,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -618,7 +668,8 @@
       "reasoning_effort": "xhigh",
       "verbosity": "medium",
       "context_window": 272000,
-      "max_output_tokens": 128000
+      "max_output_tokens": 128000,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -630,7 +681,8 @@
       "reasoning_effort": "minimal",
       "verbosity": "medium",
       "context_window": 272000,
-      "max_output_tokens": 128000
+      "max_output_tokens": 128000,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -642,7 +694,8 @@
       "reasoning_effort": "low",
       "verbosity": "medium",
       "context_window": 272000,
-      "max_output_tokens": 128000
+      "max_output_tokens": 128000,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -654,7 +707,8 @@
       "reasoning_effort": "medium",
       "verbosity": "medium",
       "context_window": 272000,
-      "max_output_tokens": 128000
+      "max_output_tokens": 128000,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -666,7 +720,8 @@
       "reasoning_effort": "high",
       "verbosity": "medium",
       "context_window": 272000,
-      "max_output_tokens": 128000
+      "max_output_tokens": 128000,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -678,7 +733,8 @@
       "reasoning_effort": "medium",
       "verbosity": "medium",
       "context_window": 272000,
-      "max_output_tokens": 128000
+      "max_output_tokens": 128000,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -690,7 +746,8 @@
       "reasoning_effort": "medium",
       "verbosity": "medium",
       "context_window": 272000,
-      "max_output_tokens": 128000
+      "max_output_tokens": 128000,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -701,7 +758,8 @@
     "isFeatured": true,
     "free": true,
     "updateArgs": {
-      "context_window": 200000
+      "context_window": 200000,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -712,7 +770,8 @@
     "isFeatured": true,
     "free": true,
     "updateArgs": {
-      "context_window": 200000
+      "context_window": 200000,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -724,7 +783,8 @@
     "free": true,
     "updateArgs": {
       "context_window": 160000,
-      "max_output_tokens": 64000
+      "max_output_tokens": 64000,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -736,7 +796,8 @@
     "free": true,
     "updateArgs": {
       "context_window": 160000,
-      "max_output_tokens": 64000
+      "max_output_tokens": 64000,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -746,7 +807,8 @@
     "description": "Minimax's latest model",
     "updateArgs": {
       "context_window": 160000,
-      "max_output_tokens": 64000
+      "max_output_tokens": 64000,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -755,7 +817,8 @@
     "label": "Kimi K2",
     "description": "Kimi's K2 model",
     "updateArgs": {
-      "context_window": 262144
+      "context_window": 262144,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -766,7 +829,8 @@
     "updateArgs": {
       "context_window": 256000,
       "max_output_tokens": 16000,
-      "temperature": 1.0
+      "temperature": 1.0,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -776,7 +840,8 @@
     "description": "Kimi's latest coding model",
     "isFeatured": true,
     "updateArgs": {
-      "context_window": 262144
+      "context_window": 262144,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -785,7 +850,8 @@
     "label": "DeepSeek Chat V3.1",
     "description": "DeepSeek V3.1 model",
     "updateArgs": {
-      "context_window": 128000
+      "context_window": 128000,
+      "parallel_tool_calls": true
     }
   },
   {
@@ -794,7 +860,11 @@
     "label": "Gemini 3.1 Pro",
     "description": "Google's latest and smartest model",
     "isFeatured": true,
-    "updateArgs": { "context_window": 180000, "temperature": 1.0 }
+    "updateArgs": {
+      "context_window": 180000,
+      "temperature": 1.0,
+      "parallel_tool_calls": true
+    }
   },
   {
     "id": "gemini-3",
@@ -802,7 +872,11 @@
     "label": "Gemini 3 Pro",
     "description": "Google's smartest model",
     "isFeatured": true,
-    "updateArgs": { "context_window": 180000, "temperature": 1.0 }
+    "updateArgs": {
+      "context_window": 180000,
+      "temperature": 1.0,
+      "parallel_tool_calls": true
+    }
   },
   {
     "id": "gemini-3-flash",
@@ -810,55 +884,63 @@
     "label": "Gemini 3 Flash",
     "description": "Google's fastest Gemini 3 model",
     "isFeatured": true,
-    "updateArgs": { "context_window": 180000, "temperature": 1.0 }
+    "updateArgs": {
+      "context_window": 180000,
+      "temperature": 1.0,
+      "parallel_tool_calls": true
+    }
   },
   {
     "id": "gemini-flash",
     "handle": "google_ai/gemini-2.5-flash",
     "label": "Gemini 2.5 Flash",
     "description": "Google's fastest model",
-    "updateArgs": { "context_window": 180000 }
+    "updateArgs": { "context_window": 180000, "parallel_tool_calls": true }
   },
   {
     "id": "gemini-pro",
     "handle": "google_ai/gemini-2.5-pro",
     "label": "Gemini 2.5 Pro",
     "description": "Google's last generation flagship model",
-    "updateArgs": { "context_window": 180000 }
+    "updateArgs": { "context_window": 180000, "parallel_tool_calls": true }
   },
   {
     "id": "gpt-4.1",
     "handle": "openai/gpt-4.1",
     "label": "GPT-4.1",
     "description": "OpenAI's most recent non-reasoner model",
-    "updateArgs": { "context_window": 1047576 }
+    "updateArgs": { "context_window": 1047576, "parallel_tool_calls": true }
   },
   {
     "id": "gpt-4.1-mini",
     "handle": "openai/gpt-4.1-mini-2025-04-14",
     "label": "GPT-4.1-Mini",
     "description": "OpenAI's most recent non-reasoner model (mini version)",
-    "updateArgs": { "context_window": 1047576 }
+    "updateArgs": { "context_window": 1047576, "parallel_tool_calls": true }
   },
   {
     "id": "gpt-4.1-nano",
     "handle": "openai/gpt-4.1-nano-2025-04-14",
     "label": "GPT-4.1-Nano",
     "description": "OpenAI's most recent non-reasoner model (nano version)",
-    "updateArgs": { "context_window": 1047576 }
+    "updateArgs": { "context_window": 1047576, "parallel_tool_calls": true }
   },
   {
     "id": "o4-mini",
     "handle": "openai/o4-mini",
     "label": "o4-mini",
     "description": "OpenAI's latest o-series reasoning model",
-    "updateArgs": { "context_window": 180000 }
+    "updateArgs": { "context_window": 180000, "parallel_tool_calls": true }
   },
   {
     "id": "gemini-3-vertex",
     "handle": "google_vertex/gemini-3-pro-preview",
     "label": "Gemini 3 Pro",
     "description": "Google's smartest Gemini 3 Pro model (via Vertex AI)",
-    "updateArgs": { "context_window": 180000, "temperature": 1.0 }
+    "updateArgs": {
+      "context_window": 180000,
+      "temperature": 1.0,
+      "parallel_tool_calls": true
+    }
   }
 ]

src/tests/agent/model-preset-refresh.wiring.test.ts

@@ -0,0 +1,81 @@
+import { describe, expect, test } from "bun:test";
+import { readFileSync } from "node:fs";
+import { fileURLToPath } from "node:url";
+
+describe("model preset refresh wiring", () => {
+  test("model.ts exports preset refresh helper", () => {
+    const path = fileURLToPath(
+      new URL("../../agent/model.ts", import.meta.url),
+    );
+    const source = readFileSync(path, "utf-8");
+
+    expect(source).toContain("export function getModelPresetUpdateForAgent(");
+    expect(source).toContain("OPENAI_CODEX_PROVIDER_NAME");
+    expect(source).toContain("getModelInfoForLlmConfig(modelHandle");
+  });
+
+  test("modify.ts keeps direct updateArgs-driven model update flow", () => {
+    const path = fileURLToPath(
+      new URL("../../agent/modify.ts", import.meta.url),
+    );
+    const source = readFileSync(path, "utf-8");
+
+    const start = source.indexOf("export async function updateAgentLLMConfig(");
+    const end = source.indexOf(
+      "export interface SystemPromptUpdateResult",
+      start,
+    );
+    expect(start).toBeGreaterThanOrEqual(0);
+    expect(end).toBeGreaterThan(start);
+    const updateSegment = source.slice(start, end);
+
+    expect(updateSegment).toContain(
+      "buildModelSettings(modelHandle, updateArgs)",
+    );
+    expect(updateSegment).toContain("getModelContextWindow(modelHandle)");
+    expect(updateSegment).not.toContain(
+      "const currentAgent = await client.agents.retrieve(",
+    );
+    expect(source).not.toContain(
+      'hasUpdateArg(updateArgs, "parallel_tool_calls")',
+    );
+  });
+
+  test("interactive resume flow refreshes model preset without explicit --model", () => {
+    const path = fileURLToPath(new URL("../../index.ts", import.meta.url));
+    const source = readFileSync(path, "utf-8");
+
+    expect(source).toContain("if (resuming)");
+    expect(source).toContain("getModelPresetUpdateForAgent");
+    expect(source).toContain(
+      "const presetRefresh = getModelPresetUpdateForAgent(agent)",
+    );
+    expect(source).toContain("resumeRefreshUpdateArgs");
+    expect(source).toContain("presetRefresh.updateArgs.max_output_tokens");
+    expect(source).toContain("presetRefresh.updateArgs.parallel_tool_calls");
+    expect(source).toContain("await updateAgentLLMConfig(");
+    expect(source).toContain("presetRefresh.modelHandle");
+    expect(source).not.toContain(
+      "await updateAgentLLMConfig(\n                agent.id,\n                presetRefresh.modelHandle,\n                presetRefresh.updateArgs,",
+    );
+  });
+
+  test("headless resume flow refreshes model preset without explicit --model", () => {
+    const path = fileURLToPath(new URL("../../headless.ts", import.meta.url));
+    const source = readFileSync(path, "utf-8");
+
+    expect(source).toContain("if (isResumingAgent)");
+    expect(source).toContain("getModelPresetUpdateForAgent");
+    expect(source).toContain(
+      "const presetRefresh = getModelPresetUpdateForAgent(agent)",
+    );
+    expect(source).toContain("resumeRefreshUpdateArgs");
+    expect(source).toContain("presetRefresh.updateArgs.max_output_tokens");
+    expect(source).toContain("presetRefresh.updateArgs.parallel_tool_calls");
+    expect(source).toContain("await updateAgentLLMConfig(");
+    expect(source).toContain("presetRefresh.modelHandle");
+    expect(source).not.toContain(
+      "await updateAgentLLMConfig(\n          agent.id,\n          presetRefresh.modelHandle,\n          presetRefresh.updateArgs,",
+    );
+  });
+});