Applied AI research at Redrob. Open weights, open evals, Apache-2.0. Inference under hardware constraints, evaluation for high-fertility languages.
Korean is a first-class locale in everything here, not a translation layer bolted on afterwards.
| redrob-os | Agentic operating system for Raspberry Pi 5 that runs a resident agent in a sandbox |
| redrob-code | AI coding agent for your terminal, with Korean as a first-class locale |
| redrob-cowork | Desktop and MCP app for doing work with AI agents on your own files |
| redrob-office | Docs, Sheets, Slides, PDF, Markdown and Hangul editors in one window |
| redrob-design | Prompt-driven design editor that opens .fig and .pen files |
| redrob-canvas | Agentic graphics editor built with Rust and Qt Quick |
| redrob-query | JVM-free, AI-native database client for PostgreSQL, MySQL, SQLite and MongoDB |
| redrob-recall | Local-first file search with grounded answers about your own files |
| redrob-studio | Local image and audio studio on your own GPU; nothing you create leaves your machine |
| redrob-eval | Open LLM evaluation workbench: verifiable prompts, blind preference, self-hosted on your GPU |
| redrob-verify | Open harness for document OCR, forgery detection, face matching and developer-identity APIs |
| redrob-image | Open-weight text-to-image model, free to use and free for commercial use |
| redrob-ui | Redrob Group Design System 2026 as typed React components, a React-free CSS layer and tokens for C++ and Kotlin |
Outside contributions are welcome:
- CONTRIBUTING.md (한국어): what gets merged quickly, and what needs a discussion first
- FORKS.md (한국어): the upstream rules that apply where a repository builds on one; read before your first change
- SECURITY.md: privately, never as a public issue
- CODE_OF_CONDUCT.md (한국어)
Ask in English or Korean, whichever you write more comfortably.
labs.redrob.ai and mckinley-and-rice, our other organization, which holds the client and platform work and publishes the engineering standards both organizations follow.
