server: add WebUI config via CLI and env var

- Add --webui-config-file and LLAMA_WEBUI_CONFIG to configure
- WebUI defaults from server side. Exposes webui_settings in
- /props for both router and child modes

Completes TODO placeholder in server-models.cpp

Author: ServeurpersoCom

common/arg.cpp

@@ -2493,6 +2493,13 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
             params.api_prefix = value;
         }
     ).set_examples({LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_API_PREFIX"));
+    add_opt(common_arg(
+        {"--webui-config-file"}, "PATH",
+        "JSON file that provides default WebUI settings (overrides WebUI defaults; can be overridden by LLAMA_WEBUI_CONFIG)",
+        [](common_params & params, const std::string & value) {
+            params.webui_config_file = value;
+        }
+    ).set_examples({LLAMA_EXAMPLE_SERVER}));
     add_opt(common_arg(
         {"--webui"},
         {"--no-webui"},

common/common.h

@@ -12,6 +12,8 @@
 #include <vector>
 #include <map>
 
+#include <nlohmann/json.hpp>
+
 #if defined(_WIN32) && !defined(_WIN32_WINNT)
 #define _WIN32_WINNT 0x0A00
 #endif
@@ -482,6 +484,8 @@ struct common_params {
 
     // "advanced" endpoints are disabled by default for better security
     bool webui            = true;
+    std::string webui_config_file;
+    nlohmann::ordered_json webui_config = nlohmann::ordered_json::object();
     bool endpoint_slots   = true;
     bool endpoint_props   = false; // only control POST requests, not GET
     bool endpoint_metrics = false;

tools/server/server-context.cpp

@@ -3117,6 +3117,7 @@ void server_routes::init_routes() {
             { "endpoint_props",              params.endpoint_props },
             { "endpoint_metrics",            params.endpoint_metrics },
             { "webui",                       params.webui },
+            { "webui_settings",              params.webui_config },
             { "chat_template",               common_chat_templates_source(ctx_server.chat_templates.get()) },
             { "bos_token",                   common_token_to_piece(ctx_server.ctx, llama_vocab_bos(ctx_server.vocab), /* special= */ true)},
             { "eos_token",                   common_token_to_piece(ctx_server.ctx, llama_vocab_eos(ctx_server.vocab), /* special= */ true)},

tools/server/server-models.cpp

@@ -822,6 +822,7 @@ void server_models_routes::init_routes() {
                     {"params", json{}},
                     {"n_ctx",  0},
                 }},
+                {"webui_settings", params.webui_config},
             });
             return res;
         }

tools/server/server.cpp

@@ -2,19 +2,24 @@
 #include "server-http.h"
 #include "server-models.h"
 
+#include "server-common.h"
 #include "arg.h"
 #include "common.h"
 #include "llama.h"
 #include "log.h"
 
 #include <atomic>
+#include <cstdlib>
+#include <fstream>
 #include <signal.h>
 #include <thread> // for std::thread::hardware_concurrency
 
 #if defined(_WIN32)
 #include <windows.h>
 #endif
 
+using json = nlohmann::ordered_json;
+
 static std::function<void(int)> shutdown_handler;
 static std::atomic_flag is_terminating = ATOMIC_FLAG_INIT;
 
@@ -29,6 +34,67 @@ static inline void signal_handler(int signal) {
     shutdown_handler(signal);
 }
 
+static bool is_valid_webui_setting_value(const json & value) {
+    return value.is_boolean() || value.is_number() || value.is_string();
+}
+
+static bool merge_webui_settings(const json & overrides, json & target, const std::string & source) {
+    if (overrides.is_null()) {
+        return true;
+    }
+
+    if (!overrides.is_object()) {
+        LOG_ERR("%s must be a JSON object of key/value pairs\n", source.c_str());
+        return false;
+    }
+
+    for (const auto & [key, value] : overrides.items()) {
+        if (!is_valid_webui_setting_value(value)) {
+            LOG_WRN("%s: ignoring '%s' because the value type is not supported (expected string/number/boolean)\n", source.c_str(), key.c_str());
+            continue;
+        }
+        target[key] = value;
+    }
+
+    return true;
+}
+
+static bool load_webui_config_from_file(const std::string & path, json & target) {
+    if (path.empty()) {
+        return true;
+    }
+
+    std::ifstream file(path);
+    if (!file.is_open()) {
+        LOG_WRN("failed to open webui config file '%s', continuing with defaults\n", path.c_str());
+        return true;
+    }
+
+    try {
+        json parsed = json::parse(file);
+        return merge_webui_settings(parsed, target, string_format("webui config file '%s'", path.c_str()));
+    } catch (const std::exception & e) {
+        LOG_ERR("failed to parse webui config file '%s' as JSON: %s\n", path.c_str(), e.what());
+        return false;
+    }
+}
+
+static void merge_webui_config_from_env(json & target) {
+    const char * env_value = std::getenv("LLAMA_WEBUI_CONFIG");
+    if (env_value == nullptr) {
+        return;
+    }
+
+    try {
+        json parsed = json::parse(env_value);
+        if (!merge_webui_settings(parsed, target, "LLAMA_WEBUI_CONFIG")) {
+            LOG_ERR("ignoring LLAMA_WEBUI_CONFIG because it is not a JSON object\n");
+        }
+    } catch (const std::exception & e) {
+        LOG_ERR("failed to parse LLAMA_WEBUI_CONFIG as JSON: %s\n", e.what());
+    }
+}
+
 // wrapper function that handles exceptions and logs errors
 // this is to make sure handler_t never throws exceptions; instead, it returns an error response
 static server_http_context::handler_t ex_wrapper(server_http_context::handler_t func) {
@@ -73,6 +139,14 @@ int main(int argc, char ** argv, char ** envp) {
         return 1;
     }
 
+    // Merge priority: WebUI built-in defaults -> config file -> LLAMA_WEBUI_CONFIG -> user localStorage
+    json webui_settings = json::object();
+    if (!load_webui_config_from_file(params.webui_config_file, webui_settings)) {
+        return 1;
+    }
+    merge_webui_config_from_env(webui_settings);
+    params.webui_config = std::move(webui_settings);
+
     // TODO: should we have a separate n_parallel parameter for the server?
     //       https://github.com/ggml-org/llama.cpp/pull/16736#discussion_r2483763177
     // TODO: this is a common configuration that is suitable for most local use cases