A local-first AI desktop application integrating conversational AI, visual generation, vector search, and multi-channel connectivity, featuring deep research modes and local knowledge bases.
Seline is a desktop AI agent platform built on Electron 39 and Next.js 16, emphasizing local-first approach and privacy protection.
Core Features#
Multi-Channel Connectivity#
- WhatsApp (via Baileys): QR code pairing, persistent auth, text and image messages
- Telegram (via Grammy): Bot token auth, message thread support, auto-chunking
- Slack (via Bolt SDK): Socket mode, channel/DM/thread messages, file uploads
- MCP: External AI service connection with built-in Node.js runtime
Intelligence & Research#
- Deep Research Mode: 6-stage workflow (Plan → Search → Analyze → Draft → Refine → Finalize)
- Local Web Browsing: Puppeteer integration with headless Chromium
- Intelligent Tool Discovery: 40+ on-demand tools, ~70% token savings
Local Knowledge & Privacy#
- Vector Search: LanceDB hybrid retrieval (dense + lexical), local or API embeddings
- Document RAG: Instant file attachment indexing and search
- Local grep: Fast pattern search using ripgrep
Visual & Creative Tools#
- Image Generation: Flux.2, GPT-5, Gemini, Z-Image, FLUX.2 Klein 4B/9B (local), WAN 2.2
- Video Assembly: AI-driven scene planning, professional transitions, Ken Burns effects
- ComfyUI Workflows: JSON import, auto input/output detection, real-time WebSocket progress
Automation & Agents#
- Task Scheduler: Recursive cron jobs, preset templates, pause/resume
- Persistent Memory: Cross-session preference and workflow retention
Technical Architecture#
- Desktop Framework: Electron 39
- Frontend: Next.js 16 + React 19 + Tailwind CSS + Radix UI
- Database: Better-sqlite3 (local SQLite) + Drizzle ORM + LanceDB + Qdrant
- AI SDK: Vercel AI SDK + Anthropic Claude Agent SDK + ONNX Runtime
LLM Provider Support#
| Provider | Model Examples | Prompt Caching |
|---|---|---|
| Anthropic | Claude series | Explicit cache breakpoints, configurable TTL |
| OpenRouter | Claude, Gemini, OpenAI, Grok, DeepSeek | Provider-side automatic |
| Kimi/Moonshot | Kimi K2.5 (256K ctx) | Provider-side automatic |
| Ollama | Local models | Not supported |
Installation & Deployment#
End Users#
- macOS: DMG installer (signing issues noted, may require workaround)
- Windows: NSIS installer + portable version
- Linux: Not tested
Development Environment#
npm install
npm run electron:dev # Start development mode
Prerequisites: Node.js 20+ (22 recommended), npm 9+
Pending Verification#
- Architecture and development docs referenced in README return 404
- Specific HuggingFace model download links not explicitly provided
- Linux version compatibility pending verification
- macOS signing issue requires attention