diff --git a/apps/x/apps/main/src/ipc.ts b/apps/x/apps/main/src/ipc.ts
index 2532755a..cd8c3551 100644
--- a/apps/x/apps/main/src/ipc.ts
+++ b/apps/x/apps/main/src/ipc.ts
@@ -69,6 +69,7 @@ import { rankSlackHomeMessages } from '@x/core/dist/knowledge/sources/rank_slack
 import { syncSlackKnowledgeSources, triggerSync as triggerSlackKnowledgeSync, getSlackKnowledgeSyncStatus } from '@x/core/dist/knowledge/sources/sync_slack.js';
 import { isOnboardingComplete, markOnboardingComplete } from '@x/core/dist/config/note_creation_config.js';
 import { loadNotificationSettings, saveNotificationSettings } from '@x/core/dist/config/notification_config.js';
+import { loadTurnLimitsSettings, saveTurnLimitsSettings } from '@x/core/dist/config/turn_limits.js';
 import { saveAppSettings } from '@x/core/dist/config/app_settings.js';
 import { setSelfCaptureActive } from '@x/core/dist/meetings/detector.js';
 import { notifyIfEnabled } from '@x/core/dist/application/notification/notifier.js';
@@ -2559,6 +2560,13 @@ export function setupIpcHandlers() {
       saveNotificationSettings(args);
       return { success: true };
     },
+    'turnLimits:getSettings': async () => {
+      return await loadTurnLimitsSettings();
+    },
+    'turnLimits:setSettings': async (_event, args) => {
+      await saveTurnLimitsSettings(args);
+      return { success: true };
+    },
     // Embedded browser handlers (WebContentsView + navigation)
     ...browserIpcHandlers,
   });
diff --git a/apps/x/apps/renderer/src/App.tsx b/apps/x/apps/renderer/src/App.tsx
index 692a055a..58b78d14 100644
--- a/apps/x/apps/renderer/src/App.tsx
+++ b/apps/x/apps/renderer/src/App.tsx
@@ -3021,6 +3021,13 @@ function App() {
       // via the agent resolver; keep them session-sticky where possible so the
       // provider prefix cache survives across turns.
       const reasoningEffort = reasoningEffortByTabRef.current.get(submitTabId)
+      // The runtime defaults omitted maxModelCalls to the global limit; the
+      // chat-specific override is the UI's job to pass explicitly. A failed
+      // settings read just falls back to the global limit.
+      const chatMaxModelCalls = await window.ipc
+        .invoke('turnLimits:getSettings', null)
+        .then((settings) => settings.chatMaxModelCalls)
+        .catch(() => undefined)
       const sendConfig = {
         agent: {
           agentId,
@@ -3039,6 +3046,7 @@ function App() {
         },
         autoPermission: (permissionMode ?? 'manual') === 'auto',
         ...(reasoningEffort ? { reasoningEffort } : {}),
+        ...(chatMaxModelCalls !== undefined ? { maxModelCalls: chatMaxModelCalls } : {}),
       }
       const userMessageContextFor = (middlePane: Awaited<ReturnType<typeof buildMiddlePaneContext>>) => ({
         currentDateTime: new Date().toISOString(),
diff --git a/apps/x/apps/renderer/src/components/settings-dialog.tsx b/apps/x/apps/renderer/src/components/settings-dialog.tsx
index cc881247..864b92ef 100644
--- a/apps/x/apps/renderer/src/components/settings-dialog.tsx
+++ b/apps/x/apps/renderer/src/components/settings-dialog.tsx
@@ -2,7 +2,7 @@
 
 import * as React from "react"
 import { useState, useEffect, useCallback, useMemo } from "react"
-import { Server, Key, Shield, Palette, Monitor, Sun, Moon, Loader2, CheckCircle2, Plus, X, Wrench, Search, ChevronRight, Link2, Tags, Mail, BookOpen, User, Plug, HelpCircle, MessageCircle, Terminal, AlertTriangle, RefreshCw, PanelRight, Bell, Smartphone } from "lucide-react"
+import { Server, Key, Shield, Palette, Monitor, Sun, Moon, Loader2, CheckCircle2, Plus, Minus, X, Wrench, Search, ChevronRight, Link2, Tags, Mail, BookOpen, User, Plug, HelpCircle, MessageCircle, Terminal, AlertTriangle, RefreshCw, PanelRight, Bell, Smartphone } from "lucide-react"
 
 import {
   Dialog,
@@ -29,12 +29,13 @@ import { AccountSettings } from "@/components/settings/account-settings"
 import { ConnectedAccountsSettings } from "@/components/settings/connected-accounts-settings"
 import { MobileChannelsSettings } from "@/components/settings/mobile-channels-settings"
 import type { ApprovalPolicy } from "@x/shared/src/code-mode.js"
+import { DEFAULT_TURN_LIMITS_SETTINGS } from "@x/shared/src/turn-limits.js"
 import type { ipc as ipcShared } from "@x/shared"
 import { startProvisioning, useProvisioning, enabledOptimistic, type AgentStatus, type CodeModeAgentStatus } from "@/lib/code-mode-provisioning"
 import { useProviderModels } from "@/hooks/use-provider-models"
 import { useChatGPT } from "@/hooks/useChatGPT"
 
-type ConfigTab = "account" | "connections" | "mobile" | "models" | "mcp" | "security" | "code-mode" | "appearance" | "notifications" | "note-tagging" | "help"
+type ConfigTab = "account" | "connections" | "mobile" | "models" | "mcp" | "security" | "code-mode" | "appearance" | "notifications" | "note-tagging" | "advanced" | "help"
 
 interface TabConfig {
   id: ConfigTab
@@ -109,6 +110,12 @@ const tabs: TabConfig[] = [
     path: "config/tags.json",
     description: "Configure tags for notes and emails",
   },
+  {
+    id: "advanced",
+    label: "Advanced",
+    icon: Wrench,
+    description: "Advanced runtime and cost controls",
+  },
   {
     id: "help",
     label: "Help",
@@ -120,7 +127,7 @@ const tabs: TabConfig[] = [
 /** Sidebar nav grouping: identity first, capabilities, then app-level. */
 const NAV_SECTIONS: { label: string | null; ids: ConfigTab[] }[] = [
   { label: null, ids: ["account", "connections", "mobile"] },
-  { label: "Configure", ids: ["models", "mcp", "security", "code-mode", "note-tagging"] },
+  { label: "Configure", ids: ["models", "mcp", "security", "code-mode", "note-tagging", "advanced"] },
   { label: "App", ids: ["appearance", "notifications", "help"] },
 ]
 
@@ -2632,6 +2639,226 @@ function NotificationSettings({ dialogOpen }: { dialogOpen: boolean }) {
   )
 }
 
+// --- Advanced (runtime/cost controls) tab ---
+
+const MODEL_CALL_LIMIT_MIN = 1
+const MODEL_CALL_LIMIT_MAX = 500
+
+function parseLimit(value: string): number | null {
+  const n = Number(value.trim())
+  if (!Number.isInteger(n) || n < MODEL_CALL_LIMIT_MIN || n > MODEL_CALL_LIMIT_MAX) return null
+  return n
+}
+
+/**
+ * Compact segmented − / value / + stepper. The native number-input spinners
+ * are replaced entirely: typing is free-form digits, stepping clamps to the
+ * range and commits immediately. An empty value steps from `fallback` (the
+ * chat field starts from the global limit).
+ */
+function LimitStepper({
+  value,
+  fallback,
+  placeholder,
+  onInput,
+  onCommit,
+}: {
+  value: string
+  fallback: number
+  placeholder?: string
+  /** Every keystroke (no save). */
+  onInput: (next: string) => void
+  /** A settled change: step click or blur. */
+  onCommit: (next: string) => void
+}) {
+  const current = parseLimit(value)
+
+  const step = (delta: number) => {
+    // From an empty/invalid field, the first step lands on the fallback so
+    // the override starts where the effective value already is.
+    const next = current === null
+      ? Math.min(MODEL_CALL_LIMIT_MAX, Math.max(MODEL_CALL_LIMIT_MIN, fallback))
+      : Math.min(MODEL_CALL_LIMIT_MAX, Math.max(MODEL_CALL_LIMIT_MIN, current + delta))
+    onCommit(String(next))
+  }
+
+  return (
+    <div className="flex h-8 items-center overflow-hidden rounded-md border border-input bg-background shadow-xs shrink-0">
+      <button
+        type="button"
+        aria-label="Decrease limit"
+        onClick={() => step(-1)}
+        disabled={current !== null && current <= MODEL_CALL_LIMIT_MIN}
+        className="flex h-full w-7 items-center justify-center text-muted-foreground transition-colors hover:bg-muted hover:text-foreground disabled:pointer-events-none disabled:opacity-30"
+      >
+        <Minus className="size-3" />
+      </button>
+      <input
+        type="text"
+        inputMode="numeric"
+        value={value}
+        placeholder={placeholder}
+        onChange={(e) => onInput(e.target.value.replace(/[^0-9]/g, ""))}
+        onBlur={() => onCommit(value)}
+        className={cn(
+          "h-full border-x border-input bg-transparent text-center text-sm tabular-nums outline-none",
+          // The 11px placeholder sits on the 14px text baseline, so it reads
+          // slightly low; nudge it up for optical centering. Only applies
+          // while the placeholder is visible, so typed text is unaffected.
+          "placeholder:text-[11px] placeholder:text-muted-foreground/70 placeholder-shown:pb-1",
+          placeholder ? "w-24" : "w-16",
+        )}
+      />
+      <button
+        type="button"
+        aria-label="Increase limit"
+        onClick={() => step(1)}
+        disabled={current !== null && current >= MODEL_CALL_LIMIT_MAX}
+        className="flex h-full w-7 items-center justify-center text-muted-foreground transition-colors hover:bg-muted hover:text-foreground disabled:pointer-events-none disabled:opacity-30"
+      >
+        <Plus className="size-3" />
+      </button>
+    </div>
+  )
+}
+
+function AdvancedSettings({ dialogOpen }: { dialogOpen: boolean }) {
+  // Inputs are kept as strings so the user can clear a field while typing;
+  // validation happens on commit (step click or blur).
+  const [globalLimit, setGlobalLimit] = useState("")
+  const [chatLimit, setChatLimit] = useState("")
+  const [loaded, setLoaded] = useState(false)
+
+  useEffect(() => {
+    if (!dialogOpen) return
+    let cancelled = false
+    window.ipc.invoke("turnLimits:getSettings", null)
+      .then((settings) => {
+        if (cancelled) return
+        setGlobalLimit(String(settings.maxModelCalls))
+        // A chat override equal to the global limit is no override — show
+        // "Same as above" (legacy files; saves already collapse this).
+        setChatLimit(
+          settings.chatMaxModelCalls !== undefined && settings.chatMaxModelCalls !== settings.maxModelCalls
+            ? String(settings.chatMaxModelCalls)
+            : ""
+        )
+        setLoaded(true)
+      })
+      .catch(() => {
+        if (!cancelled) toast.error("Failed to load advanced settings")
+      })
+    return () => { cancelled = true }
+  }, [dialogOpen])
+
+  // Saves silently on success (a toast per stepper click would be noisy,
+  // matching the notification toggles); errors still surface.
+  const saveLimits = useCallback(async (globalStr: string, chatStr: string) => {
+    const global = parseLimit(globalStr)
+    if (global === null) {
+      toast.error(`Model-call limit must be a whole number between ${MODEL_CALL_LIMIT_MIN} and ${MODEL_CALL_LIMIT_MAX}`)
+      return
+    }
+    let chat: number | undefined
+    if (chatStr.trim() !== "") {
+      const parsed = parseLimit(chatStr)
+      if (parsed === null) {
+        toast.error(`Chat limit must be empty or a whole number between ${MODEL_CALL_LIMIT_MIN} and ${MODEL_CALL_LIMIT_MAX}`)
+        return
+      }
+      chat = parsed
+    }
+    // An override equal to the global limit is meaningless — persist it as
+    // "use the global limit" so the field reopens as "Same as above".
+    if (chat === global) chat = undefined
+    try {
+      await window.ipc.invoke("turnLimits:setSettings", {
+        maxModelCalls: global,
+        ...(chat === undefined ? {} : { chatMaxModelCalls: chat }),
+      })
+    } catch {
+      toast.error("Failed to save model-call limits")
+    }
+  }, [])
+
+  if (!loaded) {
+    return (
+      <div className="h-full flex items-center justify-center text-muted-foreground text-sm">
+        <Loader2 className="size-4 animate-spin mr-2" />
+        Loading...
+      </div>
+    )
+  }
+
+  return (
+    <div className="space-y-5">
+      <div className="text-sm text-muted-foreground leading-relaxed">
+        Runtime cost and safety controls. A turn is stopped once it reaches its model-call limit;
+        changes apply to newly started turns only.
+      </div>
+
+      <div className="space-y-2">
+        <div className="rounded-md border px-3 py-3 flex items-center gap-4">
+          <div className="flex-1 min-w-0">
+            <div className="text-sm font-medium">Model-call limit</div>
+            <div className="text-xs text-muted-foreground mt-0.5">
+              Maximum model calls per turn. Applies to everything by default — background and
+              knowledge work, scheduled agents, and sub-agents (it also caps sub-agent budgets).
+            </div>
+          </div>
+          <LimitStepper
+            value={globalLimit}
+            fallback={DEFAULT_TURN_LIMITS_SETTINGS.maxModelCalls}
+            onInput={setGlobalLimit}
+            onCommit={(next) => {
+              setGlobalLimit(next)
+              void saveLimits(next, chatLimit)
+            }}
+          />
+        </div>
+
+        <div className="rounded-md border px-3 py-3 flex items-center gap-4">
+          <div className="flex-1 min-w-0">
+            <div className="text-sm font-medium">Chat model-call limit</div>
+            <div className="text-xs text-muted-foreground mt-0.5">
+              Optional separate limit for interactive chat turns. Leave empty to use the
+              model-call limit above.
+            </div>
+          </div>
+          <div className="flex items-center gap-1.5 shrink-0">
+            {chatLimit.trim() !== "" && (
+              <button
+                type="button"
+                aria-label="Use the global limit"
+                title="Use the global limit"
+                onClick={() => {
+                  setChatLimit("")
+                  void saveLimits(globalLimit, "")
+                }}
+                className="flex size-5 items-center justify-center rounded-sm text-muted-foreground transition-colors hover:bg-muted hover:text-foreground"
+              >
+                <X className="size-3.5" />
+              </button>
+            )}
+            <LimitStepper
+              value={chatLimit}
+              fallback={parseLimit(globalLimit) ?? DEFAULT_TURN_LIMITS_SETTINGS.maxModelCalls}
+              placeholder="Same as above"
+              onInput={setChatLimit}
+              onCommit={(next) => {
+                setChatLimit(next)
+                // An emptied chat field on blur means "use the global
+                // limit" — persist the override removal.
+                void saveLimits(globalLimit, next)
+              }}
+            />
+          </div>
+        </div>
+      </div>
+    </div>
+  )
+}
+
 // --- Main Settings Dialog ---
 
 export function SettingsDialog({ children, defaultTab = "account", open: controlledOpen, onOpenChange }: SettingsDialogProps) {
@@ -2684,7 +2911,7 @@ export function SettingsDialog({ children, defaultTab = "account", open: control
   }
 
   const loadConfig = useCallback(async (tab: ConfigTab) => {
-    if (tab === "appearance" || tab === "models" || tab === "note-tagging" || tab === "account" || tab === "connections" || tab === "help" || tab === "code-mode" || tab === "notifications") return
+    if (tab === "appearance" || tab === "models" || tab === "note-tagging" || tab === "account" || tab === "connections" || tab === "help" || tab === "code-mode" || tab === "notifications" || tab === "advanced") return
     const tabConfig = tabs.find((t) => t.id === tab)!
     if (!tabConfig.path) return
     setLoading(true)
@@ -2806,7 +3033,7 @@ export function SettingsDialog({ children, defaultTab = "account", open: control
             </div>
 
             {/* Content */}
-            <div className={cn("flex-1 px-6 pb-5 min-h-0", (activeTab === "models" || activeTab === "connections" || activeTab === "mobile" || activeTab === "account" || activeTab === "code-mode" || activeTab === "notifications") ? "overflow-y-auto" : activeTab === "note-tagging" ? "overflow-hidden flex flex-col" : "overflow-hidden")}>
+            <div className={cn("flex-1 px-6 pb-5 min-h-0", (activeTab === "models" || activeTab === "connections" || activeTab === "mobile" || activeTab === "account" || activeTab === "code-mode" || activeTab === "notifications" || activeTab === "advanced") ? "overflow-y-auto" : activeTab === "note-tagging" ? "overflow-hidden flex flex-col" : "overflow-hidden")}>
               {activeTab === "account" ? (
                 <AccountSettings dialogOpen={open} />
               ) : activeTab === "connections" ? (
@@ -2845,6 +3072,8 @@ export function SettingsDialog({ children, defaultTab = "account", open: control
                 <AppearanceSettings />
               ) : activeTab === "notifications" ? (
                 <NotificationSettings dialogOpen={open} />
+              ) : activeTab === "advanced" ? (
+                <AdvancedSettings dialogOpen={open} />
               ) : activeTab === "help" ? (
                 <HelpSettings />
               ) : activeTab === "code-mode" ? (
diff --git a/apps/x/apps/renderer/src/lib/session-chat/turn-view.ts b/apps/x/apps/renderer/src/lib/session-chat/turn-view.ts
index bb697e60..e12e47c3 100644
--- a/apps/x/apps/renderer/src/lib/session-chat/turn-view.ts
+++ b/apps/x/apps/renderer/src/lib/session-chat/turn-view.ts
@@ -6,6 +6,7 @@ import type {
   ToolPermissionRequestEvent,
 } from '@x/shared/src/runs.js'
 import {
+  MODEL_CALL_LIMIT_ERROR_CODE,
   deriveTurnStatus,
   outstandingAsyncTools,
   outstandingPermissions,
@@ -331,10 +332,17 @@ export function buildTurnConversation(state: TurnState): ConversationItem[] {
   }
 
   if (state.terminal?.type === 'turn_failed') {
+    // Interactive turns normally wrap up gracefully before hitting the
+    // limit; if a hard limit failure still lands here, explain it and point
+    // at the setting instead of showing the raw runtime error.
+    const message = state.terminal.code === MODEL_CALL_LIMIT_ERROR_CODE
+      ? `This turn hit its model-call limit of ${state.definition.config.maxModelCalls} before it could finish. ` +
+        'You can raise the limit in Settings → Advanced.'
+      : state.terminal.error
     items.push({
       id: `${turnId}:error`,
       kind: 'error',
-      message: state.terminal.error,
+      message,
       timestamp: ts(),
     } satisfies ErrorMessage)
   }
diff --git a/apps/x/packages/core/docs/turn-runtime-design.md b/apps/x/packages/core/docs/turn-runtime-design.md
index bda228ad..4f8d4e36 100644
--- a/apps/x/packages/core/docs/turn-runtime-design.md
+++ b/apps/x/packages/core/docs/turn-runtime-design.md
@@ -512,7 +512,13 @@ Rules:
 - `input` is the user message that defines this turn boundary.
 - `autoPermission` defaults to `false` before persistence.
 - `humanAvailable` is required explicitly.
-- `maxModelCalls` defaults to `20` before persistence.
+- `maxModelCalls`, when omitted, defaults at creation to the user's global
+  model-call limit (`config/turn_limits.json`, built-in default `50`). The
+  runtime makes no chat/headless distinction: the optional chat override is
+  the chat UI's job — its send path passes an explicit `maxModelCalls`.
+  Spawned sub-agents default to the global limit, which also caps the
+  budget a parent may grant them. Settings changes affect only newly
+  created turns.
 - Persisted values are fully resolved and immutable.
 
 The capability is named `humanAvailable`, not `headless`. `headless` describes
diff --git a/apps/x/packages/core/src/config/turn_limits.test.ts b/apps/x/packages/core/src/config/turn_limits.test.ts
new file mode 100644
index 00000000..83f19a5a
--- /dev/null
+++ b/apps/x/packages/core/src/config/turn_limits.test.ts
@@ -0,0 +1,100 @@
+import fs from "node:fs/promises";
+import os from "node:os";
+import path from "node:path";
+import { afterEach, beforeEach, describe, expect, it, vi } from "vitest";
+
+// WorkDir is resolved at module load, so each test gets a fresh temp workdir
+// via ROWBOAT_WORKDIR + resetModules + dynamic import (same pattern as
+// app_version.test.ts).
+let tmpDir: string;
+
+beforeEach(async () => {
+  tmpDir = await fs.mkdtemp(path.join(os.tmpdir(), "rowboat-turn-limits-test-"));
+  process.env.ROWBOAT_WORKDIR = tmpDir;
+  vi.resetModules();
+  // config.js fire-and-forgets a git init + Today.md migration on import;
+  // mock them out so no repo appears (and races teardown) in the temp workdir.
+  vi.doMock("../knowledge/version_history.js", () => ({
+    commitAll: vi.fn(async () => undefined),
+    initRepo: vi.fn(async () => undefined),
+  }));
+  vi.doMock("../knowledge/deprecate_today_note.js", () => ({
+    deprecateTodayNote: vi.fn(async () => undefined),
+  }));
+});
+
+afterEach(async () => {
+  delete process.env.ROWBOAT_WORKDIR;
+  vi.doUnmock("../knowledge/version_history.js");
+  vi.doUnmock("../knowledge/deprecate_today_note.js");
+  vi.resetModules();
+  await fs.rm(tmpDir, { recursive: true, force: true });
+});
+
+async function loadTurnLimits() {
+  return import("./turn_limits.js");
+}
+
+const settingsPath = () => path.join(tmpDir, "config", "turn_limits.json");
+
+async function writeSettings(content: string): Promise<void> {
+  await fs.mkdir(path.dirname(settingsPath()), { recursive: true });
+  await fs.writeFile(settingsPath(), content);
+}
+
+describe("loadTurnLimitsSettings", () => {
+  it("defaults to the built-in limit (50) when no file exists", async () => {
+    const { loadTurnLimitsSettings } = await loadTurnLimits();
+    expect(await loadTurnLimitsSettings()).toEqual({ maxModelCalls: 50 });
+  });
+
+  it("reads persisted settings", async () => {
+    await writeSettings(JSON.stringify({ maxModelCalls: 60, chatMaxModelCalls: 10 }));
+    const { loadTurnLimitsSettings } = await loadTurnLimits();
+    expect(await loadTurnLimitsSettings()).toEqual({
+      maxModelCalls: 60,
+      chatMaxModelCalls: 10,
+    });
+  });
+
+  it("fills a missing global limit from the default", async () => {
+    await writeSettings(JSON.stringify({ chatMaxModelCalls: 5 }));
+    const { loadTurnLimitsSettings } = await loadTurnLimits();
+    expect(await loadTurnLimitsSettings()).toEqual({
+      maxModelCalls: 50,
+      chatMaxModelCalls: 5,
+    });
+  });
+
+  it("falls back to defaults on a corrupt file", async () => {
+    await writeSettings("{not json");
+    const { loadTurnLimitsSettings } = await loadTurnLimits();
+    expect(await loadTurnLimitsSettings()).toEqual({ maxModelCalls: 50 });
+  });
+
+  it("falls back to defaults on out-of-range values", async () => {
+    await writeSettings(JSON.stringify({ maxModelCalls: 5000 }));
+    const { loadTurnLimitsSettings } = await loadTurnLimits();
+    expect(await loadTurnLimitsSettings()).toEqual({ maxModelCalls: 50 });
+  });
+});
+
+describe("saveTurnLimitsSettings", () => {
+  it("persists valid settings for the next load", async () => {
+    const { saveTurnLimitsSettings, loadTurnLimitsSettings } = await loadTurnLimits();
+    await saveTurnLimitsSettings({ maxModelCalls: 80, chatMaxModelCalls: 15 });
+    expect(await loadTurnLimitsSettings()).toEqual({
+      maxModelCalls: 80,
+      chatMaxModelCalls: 15,
+    });
+  });
+
+  it("rejects out-of-range values", async () => {
+    const { saveTurnLimitsSettings } = await loadTurnLimits();
+    await expect(saveTurnLimitsSettings({ maxModelCalls: 0 })).rejects.toThrow();
+    await expect(saveTurnLimitsSettings({ maxModelCalls: 501 })).rejects.toThrow();
+    await expect(
+      saveTurnLimitsSettings({ maxModelCalls: 20, chatMaxModelCalls: 501 }),
+    ).rejects.toThrow();
+  });
+});
diff --git a/apps/x/packages/core/src/config/turn_limits.ts b/apps/x/packages/core/src/config/turn_limits.ts
new file mode 100644
index 00000000..75af9e81
--- /dev/null
+++ b/apps/x/packages/core/src/config/turn_limits.ts
@@ -0,0 +1,42 @@
+import fs from 'fs/promises';
+import path from 'path';
+import {
+    TurnLimitsSettingsSchema,
+    DEFAULT_TURN_LIMITS_SETTINGS,
+    type TurnLimitsSettings,
+} from '@x/shared/dist/turn-limits.js';
+import { WorkDir } from './config.js';
+
+const TURN_LIMITS_CONFIG_PATH = path.join(WorkDir, 'config', 'turn_limits.json');
+
+/**
+ * Load the model-call limit settings, falling back to the defaults (global
+ * limit DEFAULT_MAX_MODEL_CALLS, no chat override) when the file is absent
+ * or malformed.
+ */
+export async function loadTurnLimitsSettings(): Promise<TurnLimitsSettings> {
+    try {
+        const content = await fs.readFile(TURN_LIMITS_CONFIG_PATH, 'utf-8');
+        const parsed = JSON.parse(content);
+        return TurnLimitsSettingsSchema.parse({
+            ...DEFAULT_TURN_LIMITS_SETTINGS,
+            ...(parsed && typeof parsed === 'object' ? parsed : {}),
+        });
+    } catch (error) {
+        if ((error as NodeJS.ErrnoException).code !== 'ENOENT') {
+            console.error('[TurnLimits] Error loading turn limit settings:', error);
+        }
+        return DEFAULT_TURN_LIMITS_SETTINGS;
+    }
+}
+
+export async function saveTurnLimitsSettings(
+    settings: TurnLimitsSettings,
+): Promise<void> {
+    const validated = TurnLimitsSettingsSchema.parse(settings);
+    await fs.mkdir(path.dirname(TURN_LIMITS_CONFIG_PATH), { recursive: true });
+    await fs.writeFile(
+        TURN_LIMITS_CONFIG_PATH,
+        JSON.stringify(validated, null, 2),
+    );
+}
diff --git a/apps/x/packages/core/src/runtime/assembly/spawn-agent.test.ts b/apps/x/packages/core/src/runtime/assembly/spawn-agent.test.ts
index f6440301..7b61abd4 100644
--- a/apps/x/packages/core/src/runtime/assembly/spawn-agent.test.ts
+++ b/apps/x/packages/core/src/runtime/assembly/spawn-agent.test.ts
@@ -48,6 +48,7 @@ function fakeServices(opts: {
     parentEvents?: Array<z.infer<typeof TurnEvent>>;
     childResult?: Partial<HeadlessAgentResult>;
     startError?: string;
+    globalMaxModelCalls?: number;
 }) {
     const started: HeadlessAgentOptions[] = [];
     const turnRuntime = {
@@ -79,7 +80,16 @@ function fakeServices(opts: {
             throw new Error("unused");
         },
     };
-    return { services: { turnRuntime, headlessRunner }, started };
+    return {
+        services: {
+            turnRuntime,
+            headlessRunner,
+            ...(opts.globalMaxModelCalls === undefined
+                ? {}
+                : { globalMaxModelCalls: opts.globalMaxModelCalls }),
+        },
+        started,
+    };
 }
 
 const signal = new AbortController().signal;
@@ -181,14 +191,35 @@ describe("runSpawnedAgent", () => {
         expect(agent.inline.instructions).toMatch(/headlessly/);
     });
 
-    it("rejects a model-call budget above the cap at the schema boundary", async () => {
-        const { services } = fakeServices({});
+    it("clamps a model-call budget above the global limit to it", async () => {
+        const { services, started } = fakeServices({});
         const result = await runSpawnedAgent(
-            { task: "t", instructions: "x", max_model_calls: 51 },
+            { task: "t", instructions: "x", max_model_calls: 80 },
             { parentTurnId: "parent-1", signal, services },
         );
-        expect(result.isError).toBe(true);
-        expect(result.output).toMatch(/invalid input/);
+        expect(result.isError).toBe(false);
+        expect(started[0].maxModelCalls).toBe(50);
+    });
+
+    it("uses the configured global limit as the sub-agent default and cap", async () => {
+        const { services, started } = fakeServices({ globalMaxModelCalls: 60 });
+        await runSpawnedAgent(
+            { task: "t", instructions: "x" },
+            { parentTurnId: "parent-1", signal, services },
+        );
+        expect(started[0].maxModelCalls).toBe(60);
+
+        await runSpawnedAgent(
+            { task: "t", instructions: "x", max_model_calls: 80 },
+            { parentTurnId: "parent-1", signal, services },
+        );
+        expect(started[1].maxModelCalls).toBe(60);
+
+        await runSpawnedAgent(
+            { task: "t", instructions: "x", max_model_calls: 5 },
+            { parentTurnId: "parent-1", signal, services },
+        );
+        expect(started[2].maxModelCalls).toBe(5);
     });
 
     it("refuses to spawn from a child parent (depth cap)", async () => {
diff --git a/apps/x/packages/core/src/runtime/assembly/spawn-agent.ts b/apps/x/packages/core/src/runtime/assembly/spawn-agent.ts
index 3acfd24d..2b7ad77d 100644
--- a/apps/x/packages/core/src/runtime/assembly/spawn-agent.ts
+++ b/apps/x/packages/core/src/runtime/assembly/spawn-agent.ts
@@ -54,10 +54,9 @@ export const SpawnAgentInput = z.object({
         .number()
         .int()
         .min(1)
-        .max(DEFAULT_MAX_MODEL_CALLS)
         .optional()
         .describe(
-            `Model-call budget for the sub-agent (default and cap: ${DEFAULT_MAX_MODEL_CALLS}).`,
+            "Model-call budget for the sub-agent. Defaults to the configured global limit, which is also the cap — higher values are clamped to it.",
         ),
     reasoning_effort: z
         .enum(["low", "medium", "high"])
@@ -89,6 +88,9 @@ export interface SpawnedAgentCallbacks {
     services?: {
         turnRuntime: import("../turns/api.js").ITurnRuntime;
         headlessRunner: import("./headless.js").IHeadlessAgentRunner;
+        // Default and cap for the child's model-call budget; production
+        // reads the user's global limit setting.
+        globalMaxModelCalls?: number;
     };
 }
 
@@ -112,7 +114,7 @@ export async function runSpawnedAgent(
 
     // Lazy: this module is imported by the BuiltinTools catalog, which the
     // DI container's bridges import at startup.
-    const { turnRuntime, headlessRunner } =
+    const { turnRuntime, headlessRunner, globalMaxModelCalls } =
         opts.services ?? (await resolveServices());
 
     let parentModel: z.infer<typeof ModelDescriptor> | undefined;
@@ -167,9 +169,12 @@ export async function runSpawnedAgent(
               },
           };
 
+    // The user's global limit is both the default and the cap: a parent can
+    // grant a child less budget than the setting allows, never more.
+    const spawnCap = globalMaxModelCalls ?? DEFAULT_MAX_MODEL_CALLS;
     const maxModelCalls = Math.min(
-        input.max_model_calls ?? DEFAULT_MAX_MODEL_CALLS,
-        DEFAULT_MAX_MODEL_CALLS,
+        input.max_model_calls ?? spawnCap,
+        spawnCap,
     );
 
     let handle: Awaited<ReturnType<typeof headlessRunner.start>>;
@@ -231,6 +236,9 @@ async function resolveServices(): Promise<
     NonNullable<SpawnedAgentCallbacks["services"]>
 > {
     const { lazyResolve } = await import("../../di/lazy-resolve.js");
+    const { loadTurnLimitsSettings } = await import(
+        "../../config/turn_limits.js"
+    );
     return {
         turnRuntime:
             await lazyResolve<import("../turns/api.js").ITurnRuntime>(
@@ -239,6 +247,7 @@ async function resolveServices(): Promise<
         headlessRunner: await lazyResolve<
             import("./headless.js").IHeadlessAgentRunner
         >("headlessAgentRunner"),
+        globalMaxModelCalls: (await loadTurnLimitsSettings()).maxModelCalls,
     };
 }
 
diff --git a/apps/x/packages/core/src/runtime/turns/runtime.test.ts b/apps/x/packages/core/src/runtime/turns/runtime.test.ts
index 8478867e..60afb0e6 100644
--- a/apps/x/packages/core/src/runtime/turns/runtime.test.ts
+++ b/apps/x/packages/core/src/runtime/turns/runtime.test.ts
@@ -1,4 +1,4 @@
-import { describe, expect, it } from "vitest";
+import { describe, expect, it, vi } from "vitest";
 import type { z } from "zod";
 import {
     MODEL_CALL_LIMIT_ERROR_CODE,
@@ -41,6 +41,16 @@ import type {
     ToolExecutionContext,
 } from "./tool-registry.js";
 
+// createTurn defaults an omitted maxModelCalls from the fs-backed settings
+// module; mock it so tests stay hermetic (no real WorkDir reads) and can
+// steer the configured global limit.
+const turnLimitsMock = vi.hoisted(() => ({ maxModelCalls: 50 }));
+vi.mock("../../config/turn_limits.js", () => ({
+    loadTurnLimitsSettings: async () => ({
+        maxModelCalls: turnLimitsMock.maxModelCalls,
+    }),
+}));
+
 // ---------------------------------------------------------------------------
 // Fixtures
 // ---------------------------------------------------------------------------
@@ -3060,3 +3070,45 @@ describe("turn event bus", () => {
         expect(durable.every((e) => e.sessionId === null)).toBe(true);
     });
 });
+
+// ---------------------------------------------------------------------------
+// Model-call limit resolution at createTurn (issue #768)
+// ---------------------------------------------------------------------------
+
+describe("model-call limit resolution", () => {
+    async function createdMaxModelCalls(
+        runtime: TurnRuntime,
+        repo: InMemoryTurnRepo,
+        config: { humanAvailable: boolean; maxModelCalls?: number },
+    ): Promise<number> {
+        const turnId = await newTurn(runtime, { config });
+        const [created] = await persisted(repo, turnId);
+        if (created.type !== "turn_created") throw new Error("no turn_created");
+        return created.config.maxModelCalls;
+    }
+
+    it("an omitted limit defaults to the configured global limit", async () => {
+        turnLimitsMock.maxModelCalls = 42;
+        try {
+            const { runtime, repo } = makeRuntime();
+            expect(
+                await createdMaxModelCalls(runtime, repo, { humanAvailable: true }),
+            ).toBe(42);
+            expect(
+                await createdMaxModelCalls(runtime, repo, { humanAvailable: false }),
+            ).toBe(42);
+        } finally {
+            turnLimitsMock.maxModelCalls = 50;
+        }
+    });
+
+    it("an explicit per-call limit wins over the setting", async () => {
+        const { runtime, repo } = makeRuntime();
+        expect(
+            await createdMaxModelCalls(runtime, repo, {
+                humanAvailable: true,
+                maxModelCalls: 3,
+            }),
+        ).toBe(3);
+    });
+});
diff --git a/apps/x/packages/core/src/runtime/turns/runtime.ts b/apps/x/packages/core/src/runtime/turns/runtime.ts
index 271f68ba..e118f2ec 100644
--- a/apps/x/packages/core/src/runtime/turns/runtime.ts
+++ b/apps/x/packages/core/src/runtime/turns/runtime.ts
@@ -163,7 +163,9 @@ export class TurnRuntime implements ITurnRuntime {
             config: {
                 autoPermission: input.config.autoPermission ?? false,
                 humanAvailable: input.config.humanAvailable,
-                maxModelCalls: input.config.maxModelCalls ?? DEFAULT_MAX_MODEL_CALLS,
+                maxModelCalls:
+                    input.config.maxModelCalls ??
+                    (await defaultMaxModelCalls()),
                 ...(input.config.reasoningEffort === undefined
                     ? {}
                     : { reasoningEffort: input.config.reasoningEffort }),
@@ -1359,6 +1361,21 @@ function parseToolAdditions(
     return parsed.success ? parsed.data.toolAdditions : undefined;
 }
 
+// The default budget for turns created without an explicit maxModelCalls:
+// the user's configured global limit (config/turn_limits.json). Loaded
+// lazily so this module doesn't statically pull in the fs-backed config;
+// any load problem falls back to the built-in default.
+async function defaultMaxModelCalls(): Promise<number> {
+    try {
+        const { loadTurnLimitsSettings } = await import(
+            "../../config/turn_limits.js"
+        );
+        return (await loadTurnLimitsSettings()).maxModelCalls;
+    } catch {
+        return DEFAULT_MAX_MODEL_CALLS;
+    }
+}
+
 function errorMessage(error: unknown): string {
     const message = error instanceof Error ? error.message : String(error);
     // Provider API errors carry the actual upstream failure in responseBody
diff --git a/apps/x/packages/shared/src/index.ts b/apps/x/packages/shared/src/index.ts
index a40a893a..6661420e 100644
--- a/apps/x/packages/shared/src/index.ts
+++ b/apps/x/packages/shared/src/index.ts
@@ -19,6 +19,7 @@ export * as browserControl from './browser-control.js';
 export * as billing from './billing.js';
 export * as credits from './credits.js';
 export * as notificationSettings from './notification-settings.js';
+export * as turnLimits from './turn-limits.js';
 export * as codeSessions from './code-sessions.js';
 export * as channels from './channels.js';
 export * as rowboatApp from './rowboat-app.js';
diff --git a/apps/x/packages/shared/src/ipc.ts b/apps/x/packages/shared/src/ipc.ts
index e2ce9a62..55daa37f 100644
--- a/apps/x/packages/shared/src/ipc.ts
+++ b/apps/x/packages/shared/src/ipc.ts
@@ -25,6 +25,7 @@ import { CreditActivatedEventSchema, CreditsStateSchema, ReferralClaimResultSche
 import { EmailBlockSchema, GmailThreadSchema } from './blocks.js';
 import { PermissionDecision, ApprovalPolicy, CodingAgent, type CodeRunFeedEvent } from './code-mode.js';
 import { NotificationSettingsSchema } from './notification-settings.js';
+import { TurnLimitsSettingsSchema } from './turn-limits.js';
 import { CodeProject, CodeSession, CodeSessionMode, CodeSessionStatus, GitRepoInfo, GitStatusFile, CodeAgentModelOptions } from './code-sessions.js';
 import { ChannelsConfig, ChannelsStatus } from './channels.js';
 
@@ -2455,6 +2456,17 @@ const ipcSchemas = {
       success: z.literal(true),
     }),
   },
+  // Model-call limit settings channels
+  'turnLimits:getSettings': {
+    req: z.null(),
+    res: TurnLimitsSettingsSchema,
+  },
+  'turnLimits:setSettings': {
+    req: TurnLimitsSettingsSchema,
+    res: z.object({
+      success: z.literal(true),
+    }),
+  },
 } as const;
 
 // ============================================================================
diff --git a/apps/x/packages/shared/src/turn-limits.ts b/apps/x/packages/shared/src/turn-limits.ts
new file mode 100644
index 00000000..347a1a76
--- /dev/null
+++ b/apps/x/packages/shared/src/turn-limits.ts
@@ -0,0 +1,34 @@
+import { z } from 'zod';
+import { DEFAULT_MAX_MODEL_CALLS } from './turns.js';
+
+/**
+ * User-configurable model-call budgets (see issue #768).
+ *
+ * - maxModelCalls: the global limit every turn inherits by default,
+ *   including headless/knowledge work and spawned sub-agents (it is also
+ *   the cap a parent can grant a sub-agent).
+ * - chatMaxModelCalls: optional override for interactive chat turns only;
+ *   when absent, chat uses the global limit.
+ *
+ * Changing these affects only newly created turns — each turn persists its
+ * resolved limit in turn_created.config.maxModelCalls.
+ */
+export const MIN_MODEL_CALL_LIMIT = 1;
+export const MAX_MODEL_CALL_LIMIT = 500;
+
+const limit = z
+  .number()
+  .int()
+  .min(MIN_MODEL_CALL_LIMIT)
+  .max(MAX_MODEL_CALL_LIMIT);
+
+export const TurnLimitsSettingsSchema = z.object({
+  maxModelCalls: limit,
+  chatMaxModelCalls: limit.optional(),
+});
+
+export const DEFAULT_TURN_LIMITS_SETTINGS: TurnLimitsSettings = {
+  maxModelCalls: DEFAULT_MAX_MODEL_CALLS,
+};
+
+export type TurnLimitsSettings = z.infer<typeof TurnLimitsSettingsSchema>;