Merge branch 'upstream' into concedo_experimental

# Conflicts:
#	.github/workflows/ui-publish.yml
#	CODEOWNERS
#	docs/ops.md
#	ggml/CMakeLists.txt
#	ggml/src/CMakeLists.txt
#	ggml/src/ggml-hexagon/htp/CMakeLists.txt
#	ggml/src/ggml-hexagon/htp/argsort-ops.c
#	scripts/sync-ggml.last
#	tools/mtmd/tests/test-deepseek-ocr.py

tools/server/server-http.cpp

@@ -175,6 +175,15 @@ bool server_http_context::init(const common_params & params) {
     // Middlewares
     //
 
+    // Frontend paths - all embedded UI assets
+    static const std::unordered_set<std::string> frontend_paths = []() {
+        std::unordered_set<std::string> paths { "/" };
+        for (const llama_ui_asset & a : llama_ui_get_assets()) {
+            paths.insert("/" + a.name);
+        }
+        return paths;
+    }();
+
     // Public endpoints - API routes plus all embedded UI assets
     static const std::unordered_set<std::string> get_public_endpoints = []() {
         std::unordered_set<std::string> endpoints {
@@ -182,11 +191,8 @@ bool server_http_context::init(const common_params & params) {
             "/v1/health",
             "/models",
             "/v1/models",
-            "/",
         };
-        for (const llama_ui_asset & a : llama_ui_get_assets()) {
-            endpoints.insert("/" + a.name);
-        }
+        endpoints.insert(frontend_paths.begin(), frontend_paths.end());
         return endpoints;
     }();
 
@@ -239,18 +245,9 @@ bool server_http_context::init(const common_params & params) {
 
     auto middleware_server_state = [this](const httplib::Request & req, httplib::Response & res) {
         if (!is_ready.load()) {
-#if defined(LLAMA_UI_HAS_ASSETS)
-            if (const auto tmp = string_split<std::string>(req.path, '.');
-                req.path == "/" || (!tmp.empty() && tmp.back() == "html")) {
-                if (const llama_ui_asset * a = llama_ui_find_asset("loading.html")) {
-                    res.status = 503;
-                    res.set_content(reinterpret_cast<const char*>(a->data), a->size, "text/html; charset=utf-8");
-                    return false;
-                }
+            if (frontend_paths.count(req.path)) {
+                return true; // frontend asset, allow it to load and show "loading"
             }
-#else
-            (void)req;
-#endif
             // no endpoints are allowed to be accessed when the server is not ready
             // this is to prevent any data races or inconsistent states
             res.status = 503;

tools/server/server-schema.cpp

@@ -568,10 +568,16 @@ static void handle_with_catch(const char * name, std::function<void()> func) {
     }
 }
 
+// treat a null value as absent so clients can send null to request the server default
+static bool has_value(const json & data, const char * n) {
+    auto it = data.find(n);
+    return it != data.end() && !it->is_null();
+}
+
 template <typename T>
 void field_num<T>::eval(field_eval_context & ctx, const json & data) {
     for (const auto & n : name) {
-        if (data.contains(n)) {
+        if (has_value(data, n)) {
             handle_with_catch(n, [&]() {
                 if (custom_handler) {
                 custom_handler(ctx, data);
@@ -593,7 +599,7 @@ void field_num<T>::eval(field_eval_context & ctx, const json & data) {
 void field_str::eval(field_eval_context & ctx, const json & data) {
     GGML_ASSERT(custom_handler);
     for (const auto & n : name) {
-        if (data.contains(n)) {
+        if (has_value(data, n)) {
             handle_with_catch(n, [&]() {
                 custom_handler(ctx, data);
             });
@@ -604,7 +610,7 @@ void field_str::eval(field_eval_context & ctx, const json & data) {
 
 void field_bool::eval(field_eval_context & ctx, const json & data) {
     for (const auto & n : name) {
-        if (data.contains(n)) {
+        if (has_value(data, n)) {
             handle_with_catch(n, [&]() {
                 if (custom_handler) {
                     custom_handler(ctx, data);
@@ -620,7 +626,7 @@ void field_bool::eval(field_eval_context & ctx, const json & data) {
 void field_json::eval(field_eval_context & ctx, const json & data) {
     GGML_ASSERT(custom_handler);
     for (const auto & n : name) {
-        if (data.contains(n)) {
+        if (has_value(data, n)) {
             handle_with_catch(n, [&]() {
                 custom_handler(ctx, data);
             });

tools/server/server-tools.h

@@ -9,10 +9,10 @@ struct server_tool {
     bool permission_write = false;
 
     virtual ~server_tool() = default;
-    virtual json get_definition() = 0;
-    virtual json invoke(json params) = 0;
+    virtual json get_definition() const = 0;
+    virtual json invoke(json params) const = 0;
 
-    json to_json();
+    json to_json() const;
 };
 
 struct server_tools {

tools/server/tests/unit/test_tools_builtin.py

@@ -0,0 +1,125 @@
+import os
+
+import pytest
+from utils import *
+
+server: ServerProcess
+
+# project root, used as the search directory for grep_search/file_glob_search
+PROJECT_ROOT = os.path.abspath(os.path.join(os.path.dirname(__file__), "..", "..", "..", ".."))
+
+# marker for the grep_search test to find in this file
+GREP_MARKER = "llama_cpp_test_tools_builtin_marker_grep_search"
+
+
+@pytest.fixture(autouse=True)
+def create_server():
+    global server
+    server = ServerPreset.router()
+    server.server_tools = "all"
+
+
+def call_tool(name: str, params: dict) -> dict:
+    res = server.make_request("POST", "/tools", data={"tool": name, "params": params})
+    assert res.status_code == 200, res.body
+    assert "error" not in res.body, res.body
+    return res.body
+
+
+def call_tool_expect_error(name: str, params: dict) -> str:
+    res = server.make_request("POST", "/tools", data={"tool": name, "params": params})
+    assert res.status_code == 200, res.body
+    assert "error" in res.body, res.body
+    return res.body["error"]
+
+
+def test_tools_builtin_grep_search():
+    global server
+    server.start()
+
+    res = call_tool("grep_search", {
+        "path": PROJECT_ROOT,
+        "pattern": GREP_MARKER,
+        "include": "test_tools_builtin.py",  # bare pattern -> matches basename at any depth
+    })
+    text = res["plain_text_response"]
+    assert "test_tools_builtin.py" in text
+    assert GREP_MARKER in text
+    assert "Total matches: 1" in text
+
+
+def test_tools_builtin_read_file():
+    global server
+    server.start()
+
+    this_file = os.path.join(PROJECT_ROOT, "tools", "server", "tests", "unit", "test_tools_builtin.py")
+    res = call_tool("read_file", {"path": this_file})
+    text = res["plain_text_response"]
+    assert GREP_MARKER in text
+    assert "def test_tools_builtin_read_file" in text
+
+
+def test_tools_builtin_write_then_edit_file():
+    global server
+    server.start()
+
+    log_path = os.path.join(PROJECT_ROOT, "test.log")
+    try:
+        write_res = call_tool("write_file", {"path": log_path, "content": "line1\nline2\nline3\n"})
+        assert write_res["result"] == "file written successfully"
+
+        read_before = call_tool("read_file", {"path": log_path})
+        assert read_before["plain_text_response"] == "line1\nline2\nline3\n"
+
+        edit_res = call_tool("edit_file", {
+            "path": log_path,
+            "edits": [
+                {"old_text": "line2", "new_text": "line2-edited"},
+                {"old_text": "line3\n", "new_text": "line3\nline4\n"},
+            ],
+        })
+        assert edit_res["result"] == "file edited successfully"
+        assert edit_res["edits_applied"] == 2
+
+        read_after = call_tool("read_file", {"path": log_path})
+        assert read_after["plain_text_response"] == "line1\nline2-edited\nline3\nline4\n"
+    finally:
+        if os.path.exists(log_path):
+            os.remove(log_path)
+
+
+def test_tools_builtin_edit_file_rejects_non_unique_old_text():
+    global server
+    server.start()
+
+    log_path = os.path.join(PROJECT_ROOT, "test.log")
+    try:
+        call_tool("write_file", {"path": log_path, "content": "dup\ndup\n"})
+        err = call_tool_expect_error("edit_file", {
+            "path": log_path,
+            "edits": [{"old_text": "dup", "new_text": "changed"}],
+        })
+        assert "unique" in err
+    finally:
+        if os.path.exists(log_path):
+            os.remove(log_path)
+
+
+def test_tools_builtin_edit_file_rejects_overlapping_edits():
+    global server
+    server.start()
+
+    log_path = os.path.join(PROJECT_ROOT, "test.log")
+    try:
+        call_tool("write_file", {"path": log_path, "content": "line1\nline2\n"})
+        err = call_tool_expect_error("edit_file", {
+            "path": log_path,
+            "edits": [
+                {"old_text": "line1\nline2", "new_text": "a"},
+                {"old_text": "line2", "new_text": "b"},
+            ],
+        })
+        assert "overlap" in err
+    finally:
+        if os.path.exists(log_path):
+            os.remove(log_path)

tools/server/tests/utils.py

@@ -113,6 +113,7 @@ class ServerProcess:
     ui_mcp_proxy: bool = False
     backend_sampling: bool = False
     gcp_compat: bool = False
+    server_tools: str | None = None
 
     # session variables
     process: subprocess.Popen | None = None
@@ -256,6 +257,8 @@ class ServerProcess:
             server_args.append("--no-cache-idle-slots")
         if self.ui_mcp_proxy:
             server_args.append("--ui-mcp-proxy")
+        if self.server_tools:
+            server_args.extend(["--tools", self.server_tools])
         if self.backend_sampling:
             server_args.append("--backend_sampling")
         if self.gcp_compat: