Self-hosting an LLM interface usually requires a heavy Docker stack or a dedicated GPU, but Skrongle Mini proves you can run a full-featured learning platform on a Raspberry Pi Zero 2 W. Developed by Alex Beeken, this single-binary application combines adaptive quizzes, flashcards, and an AI tutor into a lightweight service designed for families and small study groups. The project, which hit Hacker News on November 8, 2026, targets the niche of private, offline-first education without the bloat of traditional LMS platforms.

Built for the Metal, Not the Cloud

The architecture is aggressively minimalist. Skrongle is written in Rust and compiles to a static binary roughly 15 MB in size, embedding SQLite, templates, and assets directly. On a Pi Zero 2 W, the server idles at just 16 MB of RAM and caps usage at 48 MB, ensuring it won’t choke the tiny board. It supports up to eight accounts, each with four active sessions, making it perfect for a household or a small classroom. The database is a single file, simplifying backups to a mere copy operation.

AI Without the Vendor Lock-in

Unlike many self-hosted tools that force you to rely on a specific provider, Skrongle acts as a unified interface for multiple AI backends. Users can plug in API keys for Anthropic, OpenAI, Google, DeepSeek, Mistral, xAI, or OpenRouter. For the true privacy maximalist, it also supports an Ollama-compatible server running locally on your network. The AI tutor feature allows users to chat about current lessons, while the course generator can build entire curricula from a simple text description, validating every generated file against a strict JSON schema before saving.

Course Creation and Security

Content is defined in JSON files, allowing users to hand-code lessons or let the AI write them. The app ships with Spanish and French courses but supports any subject that fits the unit-to-lesson structure. Security is handled via argon2id password hashing and same-origin request checks, with no email required for sign-up. Access is managed through invitation links generated by the owner. For remote access, the documentation strongly recommends Tailscale or a VPN rather than forwarding ports, keeping the service exposed only to trusted devices.

Key Takeaways

  • Runs on a Raspberry Pi Zero 2 W with 16 MB idle RAM and a 15 MB binary.
  • Supports eight user accounts with no email dependency, using invitation links.
  • Compatible with Anthropic, OpenAI, Google, DeepSeek, Mistral, xAI, OpenRouter, and local Ollama servers.
  • Courses are defined in JSON, allowing for manual creation, file uploads, or AI generation.
  • Licensed under AGPL-3.0, with a single SQLite file for all state.

The Bottom Line

Skrongle Mini is a breath of fresh air for developers tired of Kubernetes manifests for a personal wiki. It’s a practical, low-resource solution for private education that respects both hardware limits and data sovereignty.