Merge commit '34af94cd9ab277632e27caeec2d41de2fd091b31' into concedo_experimental

# Conflicts:
#	.github/workflows/docker.yml
#	.github/workflows/make-release.yml
#	.github/workflows/release.yml
#	.pi/gg/SYSTEM.md
#	CMakeLists.txt
#	build-xcframework.sh
#	docs/development/HOWTO-add-model.md
#	docs/ops.md
#	docs/ops/SYCL.csv
#	docs/speculative.md
#	examples/sycl/update-ops-doc.sh
#	ggml/CMakeLists.txt
#	ggml/src/ggml-sycl/cpy.cpp
#	ggml/src/ggml-sycl/ggml-sycl.cpp
#	scripts/make-release-checks.sh
#	scripts/sync-ggml.last
#	tests/test-chat-auto-parser.cpp
#	tests/test-chat.cpp
#	tests/test-jinja.cpp
#	tests/test-llama-archs.cpp
#	tests/testing.h
#	tools/llama-bench/llama-bench.cpp
#	tools/server/README.md

tools/mtmd/mtmd-image.cpp

@@ -1606,12 +1606,16 @@ mtmd_image_preproc_out mtmd_image_preprocessor_granite::preprocess(const clip_im
 
     const clip_image_size orig_size = img.get_size();
     const int             tile_size = hparams.image_size;
+    GGML_ASSERT(tile_size > 0);
 
     // llava-next always encodes an overview plus a grid of tiles, even for small images
     const clip_image_size refined_size = select_best_resolution(orig_size, hparams.image_res_candidates);
     const int             grid_x       = refined_size.width  / tile_size;
     const int             grid_y       = refined_size.height / tile_size;
 
+    // the tiles are stacked on the Y axis, a big grid overflows the stacked image height
+    GGML_ASSERT(grid_x >= 0 && grid_x <= 1024 && grid_y >= 0 && grid_y <= 1024);
+
     clip_image_u8 overview;
     img_tool::resize(img, overview, {tile_size, tile_size}, hparams.image_resize_algo_ov,
                         hparams.image_pad_ov, hparams.image_pad_color_ov);

tools/server/server-context.cpp

@@ -2947,8 +2947,10 @@ private:
         });
 
         // generate the actual drafts (if any)
-        {
-            common_speculative_draft(spec.get());
+        if (!drafting.empty()) {
+            queue_tasks.yield_to_queue([&]() {
+                common_speculative_draft(spec.get());
+            });
         }
 
         // make checkpoints if needed
@@ -3578,8 +3580,8 @@ private:
             has_output |= batch.tokens[i].output;
         }
 
-        // decode on the worker thread, so we can still handle metrics tasks while waiting
-        // note: the sync is done here too, so that the wait also happens off the main thread
+        // yield to the queue, so we can still handle metrics tasks while decoding
+        // note: the sync is done here too, so that the wait is also covered by the yield
         int ret = 0;
         queue_tasks.yield_to_queue([&]() {
             ret = llama_decode(ctx_tgt, batch_view);
@@ -3644,11 +3646,18 @@ private:
         // TODO: avoid restoring the draft context and re-evaluating the drafted tokens when not needed [TAG_SPEC_AVOID_DRAFT_REEVAL]
         //       for now, always re-evaluate for simplicity
         //       ref: https://github.com/ggml-org/llama.cpp/pull/22728#issuecomment-4400925384
-        if (!common_speculative_process(spec.get(), batch_view)) {
-            SRV_ERR("%s", "failed to process speculative batch\n");
+        if (spec) {
+            bool ok = true;
+            queue_tasks.yield_to_queue([&]() {
+                ok = common_speculative_process(spec.get(), batch_view);
+            });
 
-            // TODO: handle error
-            throw std::runtime_error("failed to process speculative batch");
+            if (!ok) {
+                SRV_ERR("%s", "failed to process speculative batch\n");
+
+                // TODO: handle error
+                throw std::runtime_error("failed to process speculative batch");
+            }
         }
 
         // handle `n_cmpl > 1` tasks - when the main prompt is processed, activate all child tasks too

tools/server/server-queue.cpp

@@ -150,31 +150,46 @@ bool server_queue::process_new_tasks(bool is_yielding) {
 
 void server_queue::worker_loop() {
     while (true) {
-        std::function<void()> work;
         {
             std::unique_lock<std::mutex> lock(mutex_tasks);
+            // wait on busy instead of yielding - busy stays set even when the yield already ended
             worker.cv.wait(lock, [&]{
-                return worker.stop || worker.work != nullptr;
+                return worker.stop || worker.busy;
             });
             if (worker.stop) {
                 return;
             }
-            work = std::move(worker.work);
-            worker.work = nullptr;
         }
 
-        // note: do not hold any lock here, work() may post new tasks
-        std::exception_ptr exception;
-        try {
-            work();
-        } catch (...) {
-            exception = std::current_exception();
+        // process tasks while the yield is active
+        while (true) {
+            bool terminated = false;
+            try {
+                // note: do not hold any lock here, the callback may post new tasks
+                terminated = process_new_tasks(true);
+            } catch (...) {
+                std::unique_lock<std::mutex> lock(mutex_tasks);
+                worker.exception = std::current_exception();
+                break;
+            }
+
+            std::unique_lock<std::mutex> lock(mutex_tasks);
+            if (terminated || worker.stop || !worker.yielding) {
+                break;
+            }
+            if (!queue_tasks.empty()) {
+                continue; // a new task arrived in the meantime
+            }
+            condition_tasks.wait(lock, [&]{
+                return worker.stop || !running || !worker.yielding || !queue_tasks.empty();
+            });
         }
 
-        // signal completion to yield_to_queue()
-        std::unique_lock<std::mutex> lock(mutex_tasks);
-        worker.exception = std::move(exception);
-        worker.busy = false;
+        // signal to yield_to_queue() that no more tasks will be processed
+        {
+            std::unique_lock<std::mutex> lock(mutex_tasks);
+            worker.busy = false;
+        }
         condition_tasks.notify_all();
     }
 }
@@ -188,6 +203,7 @@ void server_queue::worker_stop() {
         worker.stop = true;
     }
     worker.cv.notify_one();
+    condition_tasks.notify_all();
     worker.thread.join();
 }
 
@@ -199,29 +215,29 @@ void server_queue::yield_to_queue(std::function<void()> && work) {
     {
         std::unique_lock<std::mutex> lock(mutex_tasks);
         GGML_ASSERT(!worker.busy && "yield_to_queue() cannot be nested");
-        worker.busy = true;
-        worker.work = std::move(work);
+        worker.busy     = true;
+        worker.yielding = true;
     }
     worker.cv.notify_one();
 
-    while (true) {
-        // note: on terminate this is a no-op, but we still wait for the work to finish
-        process_new_tasks(true);
-
-        std::unique_lock<std::mutex> lock(mutex_tasks);
-        // declined tasks are moved to queue_tasks_unhandled, so a non-empty queue always has something new
-        condition_tasks.wait(lock, [&]{
-            return !worker.busy || (running && !queue_tasks.empty());
-        });
-        if (!worker.busy) {
-            break;
-        }
+    // run the work on the current thread, so that all ggml compute stays on the same thread
+    std::exception_ptr exception;
+    try {
+        work();
+    } catch (...) {
+        exception = std::current_exception();
     }
 
-    std::exception_ptr exception;
     {
         std::unique_lock<std::mutex> lock(mutex_tasks);
 
+        // the yield is over, wait for the worker to finish its current task
+        worker.yielding = false;
+        condition_tasks.notify_all();
+        condition_tasks.wait(lock, [&]{
+            return !worker.busy;
+        });
+
         // put the declined tasks back, keeping their order
         while (!queue_tasks_unhandled.empty()) {
             queue_tasks.push_front(std::move(queue_tasks_unhandled.back()));
@@ -231,8 +247,12 @@ void server_queue::yield_to_queue(std::function<void()> && work) {
         // make sure to avoid idle timeout here
         time_last_task = ggml_time_ms();
 
-        // the worker is idle now, take the exception it may have left behind
-        std::swap(exception, worker.exception);
+        // an exception from work() takes precedence over the one from the worker
+        if (!exception) {
+            std::swap(exception, worker.exception);
+        } else {
+            worker.exception = nullptr;
+        }
     }
 
     QUE_DBG("%s", "done yielding to queue\n");
@@ -249,7 +269,9 @@ void server_queue::start_loop(int64_t idle_sleep_ms) {
 
     // spawn the worker thread used by yield_to_queue()
     GGML_ASSERT(!worker.thread.joinable() && "start_loop() is already running");
-    worker.stop = false;
+    worker.stop     = false;
+    worker.busy     = false;
+    worker.yielding = false;
     worker.thread = std::thread([this]() { worker_loop(); });
 
     constexpr auto max_wait_time = std::chrono::seconds(1);

tools/server/server-queue.h

@@ -33,11 +33,11 @@ private:
     // used by yield_to_queue, all fields are guarded by mutex_tasks
     struct worker_t {
         std::thread             thread;
-        std::condition_variable cv;   // the worker sleeps on this until there is work
-        std::function<void()>   work; // pending work, picked up by the thread
-        std::exception_ptr      exception; // exception thrown by work(), if any
-        bool stop = false;
-        bool busy = false;
+        std::condition_variable cv;        // the worker sleeps on this until a yield starts
+        std::exception_ptr      exception; // exception thrown while processing tasks, if any
+        bool stop     = false;
+        bool busy     = false; // set by yield_to_queue(), cleared by the worker once it is done processing tasks
+        bool yielding = false; // work() is still running on the start_loop() thread
     };
     worker_t worker;
 
@@ -93,7 +93,7 @@ public:
      */
     void start_loop(int64_t idle_sleep_ms = -1);
 
-    // run work() on a separate thread, while the current thread calls process_new_tasks
+    // while waiting for work() to finish, run process_new_tasks on the worker thread
     // returns once work() is done (may throw exceptions)
     // must be called from start_loop() thread (ideally inside callback_update_slots)
     // use case: return metrics while encode/decode is running
@@ -116,6 +116,7 @@ public:
     // the second argument tells whether the queue is currently yielding (see yield_to_queue)
     // only then may the callback return false to decline the task, and it must leave it
     // untouched, so that it can be put back in the queue later
+    // note: while yielding, the callback runs on worker thread, not main thread
     void on_new_task(std::function<bool(server_task &&, bool)> callback) {
         callback_new_task = std::move(callback);
     }

tools/server/server-tools.cpp

@@ -8,7 +8,6 @@
 #include <regex>
 #include <thread>
 #include <chrono>
-#include <ctime>
 #include <atomic>
 #include <cstring>
 #include <cctype>
@@ -1692,61 +1691,6 @@ private:
     }
 };
 
-//
-// get_datetime: returns the current date and time
-//
-
-struct server_tool_get_datetime : server_tool {
-    server_tool_get_datetime() {
-        name = "get_datetime";
-        display_name = "Get Date & Time";
-        permission_write = false;
-    }
-
-    json get_definition() const override {
-        return {
-            {"type", "function"},
-            {"function", {
-                {"name", name},
-                {"description", "Returns the current date and time in UTC"},
-                {"parameters", {
-                    {"type", "object"},
-                    {"properties", {
-                        {"format", {
-                            {"type", "string"},
-                            {"description",
-                                "strftime()-style format string for the output (default: \"%Y-%m-%dT%H:%M:%SZ\", "
-                                "e.g. ISO 8601). Choose your own format if you need something else, "
-                                "e.g. \"%A, %B %d %Y\" for a human-readable date."},
-                        }},
-                    }},
-                }},
-            }},
-        };
-    }
-
-    json invoke(json params, server_tool::stream *) const override {
-        std::string format = json_value(params, "format", std::string("%Y-%m-%dT%H:%M:%SZ"));
-
-        auto now  = std::chrono::system_clock::now();
-        auto time = std::chrono::system_clock::to_time_t(now);
-        std::tm tm_utc;
-#ifdef _WIN32
-        gmtime_s(&tm_utc, &time);
-#else
-        gmtime_r(&time, &tm_utc);
-#endif
-
-        char buf[256];
-        size_t len = std::strftime(buf, sizeof(buf), format.c_str(), &tm_utc);
-        if (len == 0) {
-            return {{"error", "invalid format string"}};
-        }
-
-        return {{"result", std::string(buf, len)}};
-    }
-};
-
 //
 // get_info: returns runtime info (OS name/version and cwd)
 //
@@ -2005,6 +1949,10 @@ static server_tool & find_tool(std::vector<std::unique_ptr<server_tool>> & tools
 //
 
 static std::vector<std::unique_ptr<server_tool>> build_tools() {
+    // IMPORTANT: for contributors, please keep this array of tools as minimal as possible
+    //            we only accept minimal i/o and shell command tools here
+    //            for example, do not add: web search, get date time, etc.
+    //            high-level functionality should be added either via MCP or web UI
     std::vector<std::unique_ptr<server_tool>> tools;
     tools.push_back(std::make_unique<server_tool_read_file>());
     tools.push_back(std::make_unique<server_tool_file_glob_search>());
@@ -2012,7 +1960,6 @@ static std::vector<std::unique_ptr<server_tool>> build_tools() {
     tools.push_back(std::make_unique<server_tool_exec_shell_command>());
     tools.push_back(std::make_unique<server_tool_write_file>());
     tools.push_back(std::make_unique<server_tool_edit_file>());
-    tools.push_back(std::make_unique<server_tool_get_datetime>());
     tools.push_back(std::make_unique<server_tool_get_info>());
     return tools;
 }

tools/ui/src/lib/components/app/chat/ChatMessages/ChatMessage/ChatMessageToolCall/ChatMessageToolCallBlockGetDatetime.svelte

@@ -33,7 +33,7 @@
 				if (typeof obj.result === 'string') return { dateString: obj.result.trim() };
 			}
 		} catch {
-			return { dateString: toolResultString.trim() };
+			// not JSON - nothing to show
 		}
 
 		return {};

tools/ui/src/lib/components/app/server/ServerErrorSplash.svelte

@@ -158,6 +158,8 @@
 					<div class="relative">
 						<Input
 							id="api-key-input"
+							type="password"
+							autocomplete="new-password"
 							placeholder="Enter your API key..."
 							bind:value={apiKeyInput}
 							onkeydown={handleApiKeyKeydown}

tools/ui/src/lib/components/app/settings/SettingsChat/SettingsChatFields.svelte

@@ -81,7 +81,8 @@
 				<div class="relative w-full">
 					<Input
 						id={field.key}
-						type={field.isPositiveInteger ? 'number' : 'text'}
+						type={field.isPrivate ? 'password' : field.isPositiveInteger ? 'number' : 'text'}
+						autocomplete={field.isPrivate ? 'new-password' : undefined}
 						{...field.isPositiveInteger
 							? {
 									min: String(field.min ?? 1),

tools/ui/src/lib/constants/browser-info.ts

@@ -0,0 +1,36 @@
+import { CLI_FLAGS } from './cli-flags.constants';
+import { BuiltInTool, JsonSchemaType, ToolCallType } from '$lib/enums';
+import type { OpenAIToolDefinition } from '$lib/types';
+
+export const BROWSER_INFO_TOOL_NAME = BuiltInTool.GET_INFO;
+
+/** UA token to OS name, first match wins - Android and iOS UAs also carry the Linux / Mac OS X tokens */
+export const BROWSER_INFO_OS_UA_PATTERNS: readonly [RegExp, string][] = [
+	[/Windows NT/, 'Windows'],
+	[/Android/, 'Android'],
+	[/iPhone|iPad|iPod/, 'iOS'],
+	[/CrOS/, 'ChromeOS'],
+	[/Mac OS X/, 'macOS'],
+	[/Linux/, 'Linux']
+];
+
+export const BROWSER_INFO_OS_UNKNOWN = 'unknown';
+
+/** Sent to the model as the `note` field of the tool result, next to the OS name */
+export const BROWSER_INFO_NOTE = `This environment is browser-only, it cannot read or modify local files, and it cannot run shell commands. To get local file access, tell user to launch llama-server with the ${CLI_FLAGS.AGENT} argument.`;
+
+export function buildBrowserInfoToolDefinition(): OpenAIToolDefinition {
+	return {
+		function: {
+			description:
+				'Get runtime info (OS name), may call when user asks about local files or shell commands',
+			name: BROWSER_INFO_TOOL_NAME,
+			parameters: {
+				properties: {},
+				required: [],
+				type: JsonSchemaType.OBJECT
+			}
+		},
+		type: ToolCallType.FUNCTION
+	};
+}

tools/ui/src/lib/constants/built-in-tools.constants.ts

@@ -34,7 +34,7 @@ export const BUILTIN_TOOL_UI: Readonly<Record<BuiltInTool, BuiltinToolUiEntry>>
 		label: 'Search files',
 		source: ToolSource.BUILTIN
 	},
-	[BuiltInTool.GET_DATETIME]: { icon: Clock, label: 'Current time', source: ToolSource.BUILTIN },
+	[BuiltInTool.GET_DATETIME]: { icon: Clock, label: 'Current time', source: ToolSource.FRONTEND },
 	[BuiltInTool.GET_INFO]: { icon: Info, label: 'Runtime info', source: ToolSource.BUILTIN },
 	[BuiltInTool.GREP_SEARCH]: {
 		icon: SearchCode,

tools/ui/src/lib/constants/cli-flags.constants.ts

@@ -1,4 +1,5 @@
 export const CLI_FLAGS = {
+	AGENT: '--agent',
 	API_KEY: '--api-key',
 	MCP_PROXY: '--ui-mcp-proxy',
 	SLOTS: '--slots',

tools/ui/src/lib/constants/get-datetime.ts

@@ -0,0 +1,20 @@
+import { BuiltInTool, JsonSchemaType, ToolCallType } from '$lib/enums';
+import type { OpenAIToolDefinition } from '$lib/types';
+
+export const GET_DATETIME_TOOL_NAME = BuiltInTool.GET_DATETIME;
+
+export function buildGetDatetimeToolDefinition(): OpenAIToolDefinition {
+	return {
+		function: {
+			description:
+				'Returns the current local date and time in ISO 8601 format, with the IANA time zone name',
+			name: GET_DATETIME_TOOL_NAME,
+			parameters: {
+				properties: {},
+				required: [],
+				type: JsonSchemaType.OBJECT
+			}
+		},
+		type: ToolCallType.FUNCTION
+	};
+}

tools/ui/src/lib/constants/index.ts

@@ -59,3 +59,5 @@ export * from './uri-template.constants';
 export * from './url.constants';
 export * from './working-directory.constants';
 export * from './read-media';
+export * from './get-datetime';
+export * from './browser-info';

tools/ui/src/lib/constants/settings-registry.constants.ts

@@ -324,6 +324,7 @@ const SETTINGS_REGISTRY: Record<string, SettingsSectionEntry> = {
 			{
 				defaultValue: '',
 				help: `Set the API Key if you are using <code> ${CLI_FLAGS.API_KEY} </code> option for the server.`,
+				isPrivate: true,
 				key: SETTINGS_KEYS.API_KEY,
 				label: 'API Key',
 				section: SETTINGS_SECTION_SLUGS.GENERAL,
@@ -713,6 +714,7 @@ export const SETTINGS_CHAT_SECTIONS: SettingsSection[] = [
 			help: s.help,
 			isExperimental: s.isExperimental,
 			isPositiveInteger: s.isPositiveInteger,
+			isPrivate: s.isPrivate,
 			key: s.key,
 			label: s.label,
 			max: s.max,

tools/ui/src/lib/services/mcp.service.ts

@@ -18,7 +18,8 @@ import {
 	DEFAULT_CLIENT_VERSION,
 	DEFAULT_IMAGE_MIME_TYPE,
 	DEFAULT_MCP_CONFIG,
-	HEADERS
+	HEADERS,
+	NEWLINE
 } from '$lib/constants';
 import {
 	MCPConnectionPhase,
@@ -70,6 +71,7 @@ interface ToolResultContentItem {
 
 interface ToolCallResult {
 	content?: ToolResultContentItem[];
+	structuredContent?: Record<string, unknown>;
 	isError?: boolean;
 	_meta?: Record<string, unknown>;
 }
@@ -1012,10 +1014,20 @@ export class MCPService {
 
 		if (!Array.isArray(content)) return '';
 
-		return content
+		const formatted = content
 			.map((item) => this.formatSingleContent(item))
 			.filter(Boolean)
-			.join('\n');
+			.join(NEWLINE);
+
+		if (formatted !== '') {
+			return formatted;
+		}
+
+		if (result.structuredContent && typeof result.structuredContent === 'object') {
+			return JSON.stringify(result.structuredContent);
+		}
+
+		return '';
 	}
 
 	private static formatSingleContent(content: ToolResultContentItem): string {

tools/ui/src/lib/stores/agentic.svelte.ts

@@ -56,7 +56,8 @@ import type {
 	AgenticSession,
 	McpServerOverride,
 	MCPToolCall,
-	SettingsConfigType
+	SettingsConfigType,
+	ToolExecutionResult
 } from '$lib/types';
 import type {
 	AgenticFlowCallbacks,
@@ -83,7 +84,12 @@ import type {
 	DatabaseMessageExtraAudioFile,
 	DatabaseMessageExtraImageFile
 } from '$lib/types/database';
-import { getAudioInputFormat, isAbortError } from '$lib/utils';
+import {
+	executeBrowserInfoTool,
+	executeGetDatetimeTool,
+	getAudioInputFormat,
+	isAbortError
+} from '$lib/utils';
 import { SvelteMap } from 'svelte/reactivity';
 
 function createDefaultSession(): AgenticSession {
@@ -942,18 +948,26 @@ class AgenticStore {
 							if (executionResult.isError) toolSuccess = false;
 						} else if (toolSource === ToolSource.FRONTEND) {
 							const args = this.parseToolArguments(toolCall.function.arguments);
-							const executionResult =
-								toolName === BuiltInTool.READ_MEDIA
-									? await ReadMediaService.executeTool(
-											args,
-											{
-												audio: modelsStore.modelSupportsAudio(effectiveModel),
-												vision: modelsStore.modelSupportsVision(effectiveModel)
-											},
-											signal,
-											conversationsStore.activeConversation?.cwd
-										)
-									: await SandboxService.executeTool(toolName, args, signal);
+
+							let executionResult: ToolExecutionResult;
+
+							if (toolName === BuiltInTool.GET_DATETIME) {
+								executionResult = executeGetDatetimeTool();
+							} else if (toolName === BuiltInTool.GET_INFO) {
+								executionResult = executeBrowserInfoTool();
+							} else if (toolName === BuiltInTool.READ_MEDIA) {
+								executionResult = await ReadMediaService.executeTool(
+									args,
+									{
+										audio: modelsStore.modelSupportsAudio(effectiveModel),
+										vision: modelsStore.modelSupportsVision(effectiveModel)
+									},
+									signal,
+									conversationsStore.activeConversation?.cwd
+								);
+							} else {
+								executionResult = await SandboxService.executeTool(toolName, args, signal);
+							}
 
 							result = executionResult.content;
 

tools/ui/src/lib/stores/tools.svelte.ts

@@ -1,5 +1,7 @@
 import { browser } from '$app/environment';
 import {
+	buildBrowserInfoToolDefinition,
+	buildGetDatetimeToolDefinition,
 	buildReadMediaToolDefinition,
 	DISABLED_TOOL_KEYS_LOCALSTORAGE_KEY,
 	HOME_TILDE,
@@ -175,7 +177,7 @@ class ToolsStore {
 	}
 
 	get frontendTools(): OpenAIToolDefinition[] {
-		const tools: OpenAIToolDefinition[] = [];
+		const tools: OpenAIToolDefinition[] = [buildGetDatetimeToolDefinition()];
 
 		if (settingsStore.config.jsSandboxEnabled) {
 			tools.push(buildSandboxToolDefinition(!!settingsStore.config.symbolicMathEnabled));
@@ -185,9 +187,18 @@ class ToolsStore {
 
 		if (readMedia) tools.push(readMedia);
 
+		// provide browser's get_info tool if server doesn't provide one
+		if (!this.hasBuiltinTool(BuiltInTool.GET_INFO)) {
+			tools.push(buildBrowserInfoToolDefinition());
+		}
+
 		return tools;
 	}
 
+	private hasBuiltinTool(name: BuiltInTool): boolean {
+		return this._builtinTools.some((def) => def.function.name === name);
+	}
+
 	/**
 	 * `read_media` runs in the frontend on top of the server's `read_file`, so it
 	 * exists only when that tool is served and the active model can perceive the
@@ -195,11 +206,7 @@ class ToolsStore {
 	 * conversation uses.
 	 */
 	private readMediaTool(): OpenAIToolDefinition | null {
-		const hasReadFile = this._builtinTools.some(
-			(def) => def.function.name === BuiltInTool.READ_FILE
-		);
-
-		if (!hasReadFile) return null;
+		if (!this.hasBuiltinTool(BuiltInTool.READ_FILE)) return null;
 
 		const model = modelsStore.selectedModelName ?? modelsStore.models[0]?.model ?? '';
 

tools/ui/src/lib/types/settings.d.ts

@@ -31,6 +31,7 @@ export interface SettingsEntry {
 	radioOptions?: Array<{ value: string; label: string; key: string; isExperimental?: boolean }>;
 	isExperimental?: boolean;
 	isPositiveInteger?: boolean;
+	isPrivate?: boolean;
 	placeholder?: string;
 	min?: number;
 	max?: number;
@@ -55,6 +56,7 @@ export interface SettingsFieldConfig {
 	type: SettingsFieldType;
 	isExperimental?: boolean;
 	isPositiveInteger?: boolean;
+	isPrivate?: boolean;
 	placeholder?: string;
 	min?: number;
 	max?: number;

tools/ui/src/lib/utils/browser-info.ts

@@ -0,0 +1,39 @@
+/**
+ * Browser fallback for the server's `get_info` tool, offered only when the
+ * server does not serve one (llama-server without --agent). It tells the model
+ * which OS the browser runs on and that there is no local file or shell access,
+ * so it does not plan around tools that are not there.
+ *
+ * @see server_tool_get_info in tools/server/server-tools.cpp - the served variant
+ * @see buildBrowserInfoToolDefinition in constants/browser-info.ts - tool schema sent to the LLM
+ */
+
+import { browser } from '$app/environment';
+import {
+	BROWSER_INFO_NOTE,
+	BROWSER_INFO_OS_UA_PATTERNS,
+	BROWSER_INFO_OS_UNKNOWN
+} from '$lib/constants';
+import type { ToolExecutionResult } from '$lib/types';
+
+function detectOs(userAgent: string): string {
+	for (const [pattern, os] of BROWSER_INFO_OS_UA_PATTERNS) {
+		if (pattern.test(userAgent)) return os;
+	}
+
+	return BROWSER_INFO_OS_UNKNOWN;
+}
+
+/**
+ * Result shape mirrors the server tool's JSON so the `get_info` renderer reads
+ * `os` the same way, minus `cwd` - there is no working directory to report.
+ */
+export function executeBrowserInfoTool(): ToolExecutionResult {
+	return {
+		content: JSON.stringify({
+			note: BROWSER_INFO_NOTE,
+			os: browser ? detectOs(navigator.userAgent) : BROWSER_INFO_OS_UNKNOWN
+		}),
+		isError: false
+	};
+}

tools/ui/src/lib/utils/get-datetime.ts

@@ -0,0 +1,38 @@
+/**
+ * Frontend executor for the `get_datetime` tool. It runs in the browser, so it
+ * reports the user's own clock and time zone instead of the server's UTC time -
+ * a chat about "tomorrow" means the user's tomorrow, not the host's.
+ *
+ * @see buildGetDatetimeToolDefinition in constants/get-datetime.ts - tool schema sent to the LLM
+ */
+
+import type { ToolExecutionResult } from '$lib/types';
+
+function pad(value: number): string {
+	return String(value).padStart(2, '0');
+}
+
+/** ISO 8601 in local time, e.g. `2026-08-17T14:05:09+02:00` */
+function localIsoString(date: Date): string {
+	// getTimezoneOffset() counts minutes behind UTC, ISO 8601 counts them ahead
+	const offset = -date.getTimezoneOffset();
+	const sign = offset < 0 ? '-' : '+';
+	const absOffset = Math.abs(offset);
+	const day = `${date.getFullYear()}-${pad(date.getMonth() + 1)}-${pad(date.getDate())}`;
+	const time = `${pad(date.getHours())}:${pad(date.getMinutes())}:${pad(date.getSeconds())}`;
+
+	return `${day}T${time}${sign}${pad(Math.floor(absOffset / 60))}:${pad(absOffset % 60)}`;
+}
+
+/** The `result` field keeps the shape the `get_datetime` renderer already reads. */
+export function executeGetDatetimeTool(): ToolExecutionResult {
+	const now = new Date();
+
+	return {
+		content: JSON.stringify({
+			result: localIsoString(now),
+			timezone: Intl.DateTimeFormat().resolvedOptions().timeZone
+		}),
+		isError: false
+	};
+}

tools/ui/src/lib/utils/index.ts

@@ -331,6 +331,12 @@ export { getChatCommands } from './chat-commands';
 // SANDBOX_TOOL_DEFINITION is deprecated; kept for backward compatibility.
 export { buildSandboxToolDefinition, SANDBOX_TOOL_DEFINITION } from './sandbox-tool';
 
+// Frontend `get_datetime` executor (the browser clock, not the server's)
+export { executeGetDatetimeTool } from './get-datetime';
+
+// Browser fallback for the server's get_info tool
+export { executeBrowserInfoTool } from './browser-info';
+
 // Cryptography utilities
 
 export { uuid } from './uuid';

tools/ui/tests/unit/mcp-service.test.ts

@@ -2,7 +2,7 @@ import { Client } from '@modelcontextprotocol/sdk/client';
 import { CORS_PROXY } from '$lib/constants';
 import { MCPConnectionPhase, MCPTransportType } from '$lib/enums';
 import { MCPService } from '$lib/services/mcp.service';
-import type { MCPConnectionLog, MCPServerConfig } from '$lib/types';
+import type { MCPConnection, MCPConnectionLog, MCPServerConfig } from '$lib/types';
 import { afterEach, describe, expect, it, vi } from 'vitest';
 
 type DiagnosticFetchFactory = (
@@ -329,4 +329,21 @@ describe('MCPService', () => {
 			)
 		).toHaveLength(0);
 	});
+
+	it('falls back to structuredContent when content array is empty', async () => {
+		const connection = {
+			client: {
+				callTool: vi.fn().mockResolvedValue({
+					content: [],
+					structuredContent: { accounts: [{ id: 1 }], total: 1 }
+				})
+			},
+			requestTimeoutMs: 9000,
+			serverName: 'test-server'
+		} as unknown as MCPConnection;
+		const result = await MCPService.callTool(connection, { arguments: {}, name: 'tool' });
+
+		expect(result.isError).toBe(false);
+		expect(result.content).toBe('{"accounts":[{"id":1}],"total":1}');
+	});
 });

tools/ui/tests/unit/settings-private-fields.test.ts

@@ -0,0 +1,12 @@
+import { SETTINGS_CHAT_SECTIONS, SETTINGS_KEYS } from '$lib/constants';
+import { describe, expect, it } from 'vitest';
+
+describe('checkApiKeyField', () => {
+	it('should have isPrivate set to true', () => {
+		const fields = SETTINGS_CHAT_SECTIONS.flatMap((section) => section.fields);
+		const apiKeyField = fields.find((field) => field?.key === SETTINGS_KEYS.API_KEY);
+
+		expect(apiKeyField).toBeDefined();
+		expect(apiKeyField?.isPrivate).toBe(true);
+	});
+});