Ralph: A Minimal Autonomous Coding Agent Using File and Git Memory
tom_doerr · x · 2026-08-13
Ralph is a minimal, file-based agent loop designed for autonomous coding. Instead of relying on model context, it treats files and Git as its memory.
Core Workflow:
- Uses a JSON-based PRD to define stories, gates, and status.
- Each iteration starts fresh, reads the latest on-disk state, executes one story, and commits the work.
Usage:
- Install globally via npm i -g @iannuttall/ralph.
- Features an interactive prompt to generate PRDs.
- Supports flexible project-level template overrides, with state and logs stored in the .ralph/ directory.
More from coding & agent
- DeepSeek Agent Iteratively Generates Animation Using Vision Model Feedback — PandaBearFred · 2026-08-13
- Testing LLMs Before Real Orders: The Biggest Error Source Was the Exam Author — ActiveStriking2719 · 2026-08-13
- Ditching Electron for Rust: Building Cross-Platform Apps with Multiple AI Models — dotey · 2026-08-13
- Ex-AWS Exec Runs 34 AI Agents With a 3-Word Prompt: "Do Smart Things" — alliekmiller · 2026-08-13
- ComfyUI MiniMax H3 Node Update: Multi-Shot 60s Timelines & External LLM Support — scooglecops · 2026-08-13
- SmolVM: Open-Source MicroVM Infrastructure Booting in Milliseconds for AI Agents — tom_doerr · 2026-08-13