OllamaChat

Native macOS app for chatting with local Ollama models. Built with SwiftUI, runs entirely on-device.

Features

Chat

• Stream responses from local Ollama models (Gemma 4 26B default)
• Projects with separate chats and custom system instructions
• Thinking mode — toggle model reasoning (show/hide thought process)
• Web search — DuckDuckGo integration, results injected as context
• Markdown rendering — code blocks, tables, headers, links, blockquotes
• Image understanding — drag & drop or attach photos (multimodal)
• Cancel generation — stop mid-response

Voice

• Speech-to-Text — WhisperKit (Whisper large-v3-turbo), 100+ languages including Russian
• Text-to-Speech — Apple AVSpeechSynthesizer or Qwen3-TTS 0.6B
• Sound classification — Apple MLSoundClassifier (300+ categories)
• Hands-free mode — auto-detect speech, transcribe, send to model, speak response, repeat
• Auto language detection — model tags response language, TTS picks matching voice

Telegram Bot

• Full Telegram bot with streaming responses (live message updates)
• Inline mode — use @botname query in any chat
• Voice messages — transcribed via Whisper, answered by model
• Photo messages — analyzed by multimodal model
• Threaded chats — proper forum/topic support
• User authorization — approve/deny users from the app
• Per-user projects — each Telegram user gets their own project with chat history
• Commands: /thinking, /web, /cancel, /clear

Settings

• Model download management (Whisper, Qwen3-TTS)
• TTS engine selection (Apple vs Qwen3) with voice picker and speed control
• Telegram bot configuration and user management

Requirements

• macOS 14.0+ (Sonoma or later)
• Apple Silicon (M1/M2/M3/M4)
• Ollama running locally with a model pulled
• ffmpeg (for Telegram voice messages): brew install ffmpeg

Quick Start

1. Install and start Ollama:

   ollama pull gemma4:26b

2. Build & run:

   git clone https://github.com/timskap/OllamaChat.git
   cd OllamaChat
   open OllamaChat.xcodeproj

   Or build from command line:

   xcodebuild -project OllamaChat.xcodeproj -scheme OllamaChat -configuration Release build

3. Open Settings (gear icon) to download voice models (optional):

   • Whisper (~630 MB) — for speech-to-text
   • Qwen3-TTS (~600 MB) — for AI voice synthesis

Telegram Bot Setup

1. Create a bot via @BotFather
2. Enable inline mode: /setinline and /setinlinefeedback (set to Enabled)
3. Open Settings in the app → Telegram tab
4. Paste bot token and enable
5. Approve users as they request access

Architecture

OllamaChat/
├── OllamaChatApp.swift        # App entry point
├── Models.swift               # Data models (Project, Chat, Message)
├── ProjectStore.swift         # JSON persistence
├── OllamaService.swift        # Ollama API client with streaming
├── AudioService.swift         # WhisperKit STT + mic recording
├── TTSService.swift           # Apple TTS + Qwen3-TTS
├── SoundClassifierService.swift # Apple MLSoundClassifier
├── WebSearchService.swift     # DuckDuckGo search
├── TelegramService.swift      # Telegram Bot API
├── ContentView.swift          # Main layout (NavigationSplitView)
├── SidebarView.swift          # Projects & chats sidebar
├── ChatView.swift             # Chat UI with toggles
├── InstructionsView.swift     # Project instructions editor
└── SettingsView.swift         # Global settings & model management

Dependencies

• WhisperKit (includes TTSKit) — speech recognition & synthesis
• MarkdownUI — markdown rendering
• Apple frameworks: SoundAnalysis, AVFoundation, AppKit

License

MIT