Next refactors

lib/domain/services/ai_process_manager.dart

@@ -1,88 +0,0 @@
-import 'dart:io';
-
-import 'package:path/path.dart' as p;
-import 'package:path_provider/path_provider.dart';
-
-/// Manages the two llama.cpp server processes that provide AI features.
-///
-/// Both processes are kept alive for the lifetime of the app and must be
-/// killed on shutdown to prevent zombie processes from consuming RAM.
-///
-/// - Qwen 2.5 7B  → port 8080  (chat / completions)
-/// - Nomic Embed  → port 8081  (embeddings)
-class AiProcessManager {
-  Process? _qwenProcess;
-  Process? _nomicProcess;
-  bool _running = false;
-
-  // -------------------------------------------------------------------------
-  // Public API
-  // -------------------------------------------------------------------------
-
-  bool get isRunning => _running;
-
-  /// Starts both inference servers.  No-ops if already running or if the
-  /// llama-server binary is not present on disk.
-  Future<void> startServers() async {
-    if (_running) return;
-
-    final dir = await getApplicationDocumentsDirectory();
-    final base = dir.path;
-    final serverBin = p.join(
-      base,
-      Platform.isWindows ? 'llama-server.exe' : 'llama-server',
-    );
-
-    if (!File(serverBin).existsSync()) return;
-
-    try {
-      // ── Qwen 2.5 7B chat server ──────────────────────────────────────────
-      _qwenProcess = await Process.start(
-        serverBin,
-        [
-          '-m', p.join(base, 'qwen2.5-7b-instruct-q4_k_m.gguf'),
-          '--port', '8080',
-          '--ctx-size', '4096',
-          '-ngl', '99', // offload all layers to GPU
-        ],
-        runInShell: false,
-      );
-      // Drain pipes so the process is never blocked by a full buffer.
-      _qwenProcess!.stdout.drain<List<int>>();
-      _qwenProcess!.stderr.drain<List<int>>();
-
-      // ── Nomic embedding server ───────────────────────────────────────────
-      _nomicProcess = await Process.start(
-        serverBin,
-        [
-          '-m', p.join(base, 'nomic-embed-text-v1.5.Q4_K_M.gguf'),
-          '--port', '8081',
-          '--ctx-size', '8192',
-          '--embedding',
-        ],
-        runInShell: false,
-      );
-      _nomicProcess!.stdout.drain<List<int>>();
-      _nomicProcess!.stderr.drain<List<int>>();
-
-      _running = true;
-    } catch (_) {
-      // Clean up any partially-started processes before rethrowing.
-      _qwenProcess?.kill();
-      _nomicProcess?.kill();
-      _qwenProcess = null;
-      _nomicProcess = null;
-      rethrow;
-    }
-  }
-
-  /// Kills both processes and resets the running flag.
-  /// Safe to call even if servers were never started.
-  Future<void> stopServers() async {
-    _qwenProcess?.kill();
-    _nomicProcess?.kill();
-    _qwenProcess = null;
-    _nomicProcess = null;
-    _running = false;
-  }
-}

lib/domain/services/embedding_service.dart

@@ -1,31 +0,0 @@
-import 'package:dio/dio.dart';
-
-/// Wraps the Nomic embedding server (llama.cpp, port 8081).
-/// Returns a 768-dimensional float vector for any input text.
-class EmbeddingService {
-  final _dio = Dio(
-    BaseOptions(
-      connectTimeout: const Duration(seconds: 10),
-      receiveTimeout: const Duration(seconds: 60),
-    ),
-  );
-
-  static const _url = 'http://localhost:8081/v1/embeddings';
-
-  /// Returns the embedding vector for [text].
-  /// Throws a [DioException] if the Nomic server is unreachable.
-  Future<List<double>> embed(String text) async {
-    final response = await _dio.post<Map<String, dynamic>>(
-      _url,
-      data: {
-        'input': text,
-        'model': 'nomic-embed-text-v1.5.Q4_K_M',
-      },
-    );
-
-    final raw =
-        (response.data!['data'] as List<dynamic>)[0]['embedding']
-            as List<dynamic>;
-    return raw.map((e) => (e as num).toDouble()).toList();
-  }
-}