gridwright
Turning a Figma design into a working component eats an afternoon of pixel-pushing.
- TypeScript
- Figma
- Claude Code
Open source
We build with AI agents every day. What we make along the way, we open up: repos you can use, prompts to copy and what we've learned.
01 · Code you can use
Each one solves something that was slowing us down. Signed by whoever built it.
With five agents working at once, you can't tell who's doing what.
Your coding agents, their branches and the system they're building, as a 3D scene you can move through. The visual face of ORCA.
Every Claude Code session starts from zero, and you explain the project all over again.
Persistent memory for Claude Code: it remembers what you decided, where you left off and how each project works. We use it on ours.
A model's first answer is almost never the good one.
An MCP server that generates, evaluates and fixes until the result clears the quality bar you set. With Claude or OpenAI.
Reviewing every LLM output by hand doesn't scale.
The same generate-and-evaluate loop, built in LangGraph so you can drop it into your own pipeline.
Turning a Figma design into a working component eats an afternoon of pixel-pushing.
A Figma node goes in, a verified component comes out. All driven from Claude Code.
Your agents can't read or update Jira.
A Jira Cloud CLI for issues, worklogs, boards and sprints, with JSON output a person, a script or an agent can read.
Eight hours coding in a chair.
Drive Claude Code with hand gestures and code standing up, no keyboard needed.
There are four RAG frameworks and you don't know which one to pick.
All four compared in TypeScript —Vercel AI SDK, LangChain, LlamaIndex and Mastra— with a demo that runs entirely in the browser.
AI writes correctly, but it doesn't write like you.
Local RAG that retrieves examples of an author's style and method to write the same way. Reads PDFs and images, runs on your machine with Ollama.
The declassified files exist, but nobody can read ten thousand documents.
An open chatbot over 10,000+ officially declassified documents from Argentina's last dictatorship, running at US$0 a month.
You want a personal assistant that doesn't send your data anywhere.
An agent that runs locally with Ollama, talks to you on Telegram, reads your Gmail and keeps its memory in SQLite.
The launch video ends up as a screen recording with text on top.
Turns any web app into a Linear- or Vercel-style launch video from a config. An LLM can produce it.
Three projects open and two fighting over port 3000.
Run several Laravel, Django, Next.js, Vite or Docker projects at once, with no port collisions and zero config.
Filling out the same job application twenty times.
A Chrome extension that fills job applications in one click. No backend: nothing leaves your computer.
Walking into an interview without having reviewed anything.
A CLI that combines your profile and the job post into a context block ready to paste into Claude and practice.
Everyone talks about AI; nobody tells you what works in production.
A bilingual logbook of applied-AI practices: a 12-module course and a radar that stays current.
You want to practice technical English and never find the time.
Turns the code you shipped today into an English lesson: vocabulary, reading and drills, from Claude Code.
OS memory management doesn't click on a whiteboard.
An interactive simulator for memory management and CPU scheduling, in English, Spanish and Portuguese.