medit is an agentic AI coding editor that lives inside your running app. Click the thing you want changed, describe it in plain English — medit writes the code, shows you the result live, commits it, and opens the PR.
No editor. No terminal. No laptop.
One overlay. Everyone ships.
The person who spots the problem is almost never the person who can fix it. medit deletes the handoff — whoever sees it, ships it.
Exactly the change you pictured
Stop annotating screenshots and hoping it survives translation. Click the element, describe the tweak — spacing, color, type, motion — and watch the real app change under your eye. Pixel-perfect, because you're the one pointing at the pixels.
Send PRs, not paragraphs
That copy tweak you'd normally describe across three Slack messages and a Loom? Make it yourself. See it live, then hand the team a pull request containing the exact change — reviewed and mergeable, not interpretable.
Cursor, but the canvas is your app
Skip the file hunt — the element you click is the context. Every change lands on its own branch with a clean commit. Git worktrees, per-branch previews, a real terminal, and your choice of agent: Claude Code, Cursor CLI, Codex, or any API model.
Ship from your phone
medit serves over your network (LAN / Tailscale), so the overlay opens on whatever you're holding. Embedded at a customer site, mid-demo, in the room where it breaks — see it, fix it, ship it before the meeting ends. The ultimate forward-deployed engineer tool.
The precision of pointing at the thing. The power of an agent that writes real code. In the only place that's true — your running app, not a mockup of it.
Bring your subscription, not an API key
medit doesn't meter you per token. It drives Claude Code — the agent on the subscription you're already paying for — and points it at your running app. No API key required.
claude runs in your terminal, medit works right now — same login, same plan.medit lives inside your app — the overlay in the corner sends your selection down to your agent — Claude Code, Cursor, or OpenCode — and real code flows back up. On loop, all day.
Anywhere decisions are made
Your dev environment runs where your app runs. medit's overlay opens on any device on your network — so the moment you see it is the moment you fix it.
How it works
Batteries included
Beyond the chat box
The overlay carries the whole toolchain — so “almost done” never means “back to the laptop.”
A real terminal, in the overlay
Not a sandboxed toy — a full shell on the machine your app runs on. Tail the logs, run the migration, poke the database — from the same panel, even from your phone.
Guidance the agent actually follows
Write it once, attach it to any chat. Unlike skills — which the model picks — you choose what applies, and it's honored on every message in that chat, scheduled ones included.
Every task runs your ritual
Standing rules around every change: pull and lint before, test and tidy after. The agent runs your checklist on every single task — nothing gets skipped on a Friday afternoon.
git pull · npm run lintnpm test · update changelog“Email me about this” — done
Connect Gmail in one click, or bring a Resend key — then the agent can mail you: summaries, PR links, nightly-audit reports. The recipient is locked to your own address, so it's injection-safe.
Multiplayer, out of the box
The same session opens on every screen: the PM types, the designer watches the change land — live. Shared chats per branch, presence on the page, and git worktrees keeping parallel work from colliding. No “can you share your screen?”
Get medit
cd into your React app — or scaffold one with medit --create-new-canvas.MEDIT_OPENAI_API_KEY to .env.)medit --dev, open your app — the overlay appears bottom-right. Point at something.One self-contained binary — no Node needed for medit itself. Auto-updates. Free for personal use; Pro unlocks PR automation, multiple projects, production previews, and team access — pricing.