Skip to content

Project

Project translates screenshot to code fix.

Core trick

Gemini extracts an idea from the screenshot, then an agent matches it against scanned local repos and executes approved plans with sandboxed tools.

Tags

Description

Project translates screenshot to code fix.
rahulrajaram/relay★ 00 forksTypeScriptCreated Mar 21, 2026 · last push Mar 21, 2026

README excerpt

# Relay

Turn screenshots of tweets, images, and ideas into actionable code — either as new projects or changes to existing ones.

## How It Works

1. **Upload** — User uploads a screenshot (tweet, image, diagram, etc.) through the chat UI
2. **Infer** — Backend sends the image to Google Gemini for analysis and idea extraction
3. **Plan** — An agentic pipeline reconciles the idea against the user's existing projects (scanned from `~/Documents`) and produces a plan:
   - **New project** — scaffold and build something new
   - **Modify existing project** — make changes to a detected project
   - **Discard** — the idea isn't actionable
4. **Execute** — On user approval, an agent executes the plan using sandboxed tools (read/write files, run commands), with every tool invocation tracked as an execution step
5. **Commit** — After all changes are verified, the agent commits to git with a conventional commit message

```
┌──────┐          ┌──────────┐        ┌─────────┐       ┌────────┐       ┌──────────┐
│ User │          │ Frontend │        │ Backend │       │ Gemini │       │  Target  │
│      │          │ Next.js  │        │ Express │       │  API   │       │  Repo    │
└──┬───┘          └────┬─────┘        └────┬────┘       └───┬────┘       └────┬─────┘
   │                   │                   │                │                 │
   │  Upload screenshot│                   │                │                 │
   │  + describe idea  │                   │                │       

Team