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-code | AI coding agent for the terminal |
| redrob-cowork | Desktop and MCP app for working with agents on your own files |
| redrob-office | Docs, Sheets, Slides, PDF, Markdown and Hangul editors in one window |
| redrob-design | Prompt-driven design editor for .fig and .pen files |
| redrob-canvas | Agentic raster graphics editor in Rust and Qt Quick |
| redrob-query | JVM-free database client for PostgreSQL, MySQL, SQLite and MongoDB |
| redrob-recall | Local-first file search with grounded answers |
| redrob-studio | Image and audio tools backed by ComfyUI on your own GPU |
| redrob-eval | LLM eval workbench: verifiable prompts, blind preference, GEPA, self-hosted vLLM |
| redrob-verify | Open harness for document OCR, forgery detection and face matching |
| redrob-image | Open-weight text-to-image model, free for commercial use |
Outside contributions are welcome, and most of these repositories are forks, which changes the rules more than anything else:
- CONTRIBUTING.md (한국어): what gets merged quickly, and what needs a discussion first
- FORKS.md (한국어): 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.