frontpage.
newsnewestaskshowjobs

Made with ♥ by @iamnishanth

Open Source @Github

fp.

Open in hackernews

Show HN: Lukan – An open-source agentic workstation in a single Rust binary

https://lukan.ai
2•enzovn•1h ago
Hi HN, I've been building Lukan, an open-source (MIT) agentic workstation that runs entirely as a single Rust binary with zero runtime dependencies. I started this because I wanted a unified workstation optimized for my own productivity. My goal was to build an environment where I could securely remote into my machine from anywhere, seamlessly view and modify local files, and run AI agents or drop into a terminal side-by-side, all integrated with a rich set of built-in tools. Here is what makes it different: Multi-agent sessions: Run multiple agents in parallel, each with its own isolated context and history. You can switch between them using tabs, resume after disconnects, and even rewind to previous checkpoints. Embedded terminal: A full tmux-backed terminal lives inside the Web and Desktop UI. Sessions survive page reloads and network drops. You can run any CLI tool alongside your agent (including Claude Code, Codex CLI, Lukan Cli, etc.). Pipelines & Background workers: DAG-based multi-agent workflows with a visual editor, human-in-the-loop approval gates, and cron-style autonomous tasks that keep running while you're away. E2E encrypted remote access: Access your workstation from any browser without port forwarding. It uses X25519 key exchange + AES-GCM, meaning the relay server never sees your data. Bring your own model: Native support for Anthropic, OpenAI, GitHub Copilot (via OAuth, no API key needed), Gemini, Ollama, and any OpenAI-compatible endpoint (vLLM, LM Studio). You can swap models mid-session using /model. Browser automation: Built-in Chrome DevTools Protocol support. The agent can navigate, screenshot, click, type, eval JS, and export PDFs. Sandboxed execution: OS-level isolation using bubblewrap (bwrap) on Linux and sandbox-exec (Seatbelt profiles) on macOS. You can configure allowed paths, and it automatically detects and protects sensitive files like .env, .ssh/, and .aws/. Plugin ecosystem: WhatsApp, Telegram, Slack, Discord, Gmail, Workspace, Docker monitor, and local audio transcription via Whisper (no data leaves your machine). The backend is built entirely in Rust with a React/TypeScript frontend. Quick start: curl -fsSL https://get.lukan.ai/install.sh | bash (Or via Docker: docker run -d -p 3000:3000 lukanlabs/lukan) Repo: https://github.com/lukanlabs/lukan Docs: https://docs.lukan.ai I'm not a professional software engineer — this started as a personal project to scratch my own itch. The codebase has grown a lot and I'm sure there are things that could be done better. Feedback, criticism, and suggestions are very welcome. I'll be around to answer any questions about the Rust architecture, the E2E encryption design, our cross-platform sandboxing, or anything else!