feat: Pass API key via environment variable instead of command line argument (#6225)

This change modifies how the API key is passed to the llama-server process. Previously, it was sent as a command line argument (--api-key). This approach has been updated to pass the key via an environment variable (LLAMA_API_KEY).

This improves security by preventing the API key from being visible in the process list (ps aux on Linux, Task Manager on Windows, etc.), where it could potentially be exposed to other users or processes on the same system.

The commit also updates the Rust backend to read the API key from the environment variable instead of parsing it from the command line arguments.

extensions/llamacpp-extension/src/index.ts

@@ -1176,6 +1176,7 @@ export default class llamacpp_extension extends AIEngine {
       }
     }
     const args: string[] = []
+    const envs: Record<string, string> = {}
     const cfg = { ...this.config, ...(overrideSettings ?? {}) }
     const [version, backend] = cfg.version_backend.split('/')
     if (!version || !backend) {
@@ -1202,7 +1203,7 @@ export default class llamacpp_extension extends AIEngine {
     // disable llama-server webui
     args.push('--no-webui')
     const api_key = await this.generateApiKey(modelId, String(port))
-    args.push('--api-key', api_key)
+    envs["LLAMA_API_KEY"] = api_key
 
     // model option is required
     // NOTE: model_path and mmproj_path can be either relative to Jan's data folder or absolute path
@@ -1287,14 +1288,12 @@ export default class llamacpp_extension extends AIEngine {
 
     try {
       // TODO: add LIBRARY_PATH
-      const sInfo = await invoke<SessionInfo>(
-        'plugin:llamacpp|load_llama_model',
-        {
+      const sInfo = await invoke<SessionInfo>('plugin:llamacpp|load_llama_model', {
         backendPath,
         libraryPath,
         args,
-        }
-      )
+        envs,
+      })
       return sInfo
     } catch (error) {
       logger.error('Error in load command:\n', error)

src-tauri/plugins/tauri-plugin-llamacpp/src/commands.rs

@@ -1,6 +1,7 @@
 use base64::{engine::general_purpose, Engine as _};
 use hmac::{Hmac, Mac};
 use sha2::Sha256;
+use std::collections::HashMap;
 use std::process::Stdio;
 use std::time::Duration;
 use tauri::{Manager, Runtime, State};
@@ -42,6 +43,7 @@ pub async fn load_llama_model<R: Runtime>(
     backend_path: &str,
     library_path: Option<&str>,
     mut args: Vec<String>,
+    envs: HashMap<String, String>,
 ) -> ServerResult<SessionInfo> {
     let state: State<LlamacppState> = app_handle.state();
     let mut process_map = state.llama_server_process.lock().await;
@@ -54,12 +56,21 @@ pub async fn load_llama_model<R: Runtime>(
     let port = parse_port_from_args(&args);
     let model_path_pb = validate_model_path(&mut args)?;
 
-    let api_key = extract_arg_value(&args, "--api-key");
+    let api_key: String;
+
+    if let Some(api_value) = envs.get("LLAMA_API_KEY") {
+        api_key = api_value.to_string();
+    } else {
+        log::warn!("API key not provided");
+        api_key = "".to_string();
+    }
+
     let model_id = extract_arg_value(&args, "-a");
 
     // Configure the command to run the server
     let mut command = Command::new(backend_path);
     command.args(args);
+    command.envs(envs);
 
     setup_library_path(library_path, &mut command);
     command.stdout(Stdio::piped());