refactor(backend): use systemd for daemon management

Replace manual process management with systemctl --user commands.
Uses ollama.service and llama-server.service units for proper lifecycle
management, VRAM cleanup, and integration with system services.

src/frontends/tui/minimal.ts

@@ -52,7 +52,6 @@ export class MinimalTui {
   private totalUsage: TokenUsage = { inputTokens: 0, outputTokens: 0 };
   private currentHint = '';
   private lastLine = '';
-  private backendPids: Map<string, number> = new Map();
 
   constructor(private config: MinimalTuiConfig) {}
 
@@ -323,61 +322,62 @@ export class MinimalTui {
   }
 
   private async stopBackend(provider: string): Promise<void> {
-    const pid = this.backendPids.get(provider);
-    if (!pid) {
-      return; // No tracked PID, process not started by us
-    }
-
     try {
-      process.kill(pid, 'SIGTERM');
-      // Wait up to 2 seconds for graceful shutdown
-      await new Promise<void>((resolve) => {
-        const timeout = setTimeout(resolve, 2000);
-        const checkInterval = setInterval(() => {
-          try {
-            process.kill(pid, 0); // Check if process exists
-          } catch {
-            clearInterval(checkInterval);
-            clearTimeout(timeout);
+      const { exec } = await import('child_process');
+      let serviceName: string;
+      switch (provider) {
+        case 'ollama':
+          serviceName = 'ollama.service';
+          break;
+        case 'llamacpp':
+          serviceName = 'llama-server.service';
+          break;
+        default:
+          return;
+      }
+      await new Promise<void>((resolve, reject) => {
+        exec(`systemctl --user stop ${serviceName}`, (error) => {
+          if (error) {
+            reject(error);
+          } else {
             resolve();
           }
-        }, 100);
+        });
       });
-      this.backendPids.delete(provider);
     } catch (error) {
-      // Process already dead or permission error
-      this.backendPids.delete(provider);
+      // Service might not exist or already stopped, ignore
+      console.log(`${colors.gray}Note: ${provider} service not managed by systemd${colors.reset}\n`);
     }
   }
 
   private async startBackend(provider: string, config: ModelConfig): Promise<void> {
     try {
-      const { spawn } = await import('child_process');
-      const args: string[] = [];
-
-      let proc: ReturnType<typeof spawn> | undefined;
-
+      const { exec } = await import('child_process');
+      let serviceName: string;
       switch (provider) {
         case 'ollama':
-          proc = spawn('ollama', ['serve'], { detached: true, stdio: 'ignore' });
+          serviceName = 'ollama.service';
           break;
         case 'llamacpp':
-          args.push('--model', config.model);
-          args.push('--port', new URL(config.endpoint ?? 'http://localhost:8080').port || '8080');
-          args.push('--host', '0.0.0.0');
-          proc = spawn('llama-server', args, { detached: true, stdio: 'ignore' });
+          serviceName = 'llama-server.service';
           break;
+        default:
+          return;
       }
+      await new Promise<void>((resolve, reject) => {
+        exec(`systemctl --user start ${serviceName}`, (error) => {
+          if (error) {
+            reject(error);
+          } else {
+            resolve();
+          }
+        });
+      });
 
-      if (proc && proc.pid) {
-        proc.unref();
-        this.backendPids.set(provider, proc.pid);
-      }
-
-      // Wait briefly for the daemon to start
+      // Wait briefly for daemon to start
       await new Promise(resolve => setTimeout(resolve, 500));
     } catch (error) {
-      console.log(`${colors.gray}Warning: Failed to start ${provider}: ${error instanceof Error ? error.message : String(error)}${colors.reset}\n`);
+      console.log(`${colors.gray}Warning: Failed to start ${provider} via systemd: ${error instanceof Error ? error.message : String(error)}${colors.reset}\n`);
     }
   }