Peekable lets agents show work to humans and turn comments into the next iteration. Share an HTML mockup or live localhost app, collect feedback directly on the page, then pull structured notes back into your coding loop.
2 items · ships in 3–5 days
↑ click any element to annotate · widget stays in the corner
Peekable runs from your shell or your agent. No accounts to wire up, no dashboards to babysit, no build pipeline to thread. Just a URL.
Static HTML, a brainstorm screen, slides, a report. Run peekable create for a share ID, then peekable push for each new version. Reviewers always see the latest.
Running a dev server? Peekable proxies your local port to a public URL so a reviewer on a phone — or you on the couch — can poke around without deploying.
Reviewers click on elements and drop notes. Your next agent loop pulls them in as structured JSON — selector, element text, DOM path, route, version, reviewer — not vibes. Resolve and push the next version.
When a reviewer leaves a note, Peekable captures the element, its CSS selector, computed text, route, version, and viewport. Your agent gets a JSON payload it can actually act on.
Slack threads die. Screenshots get stale. Peekable comments stick to the element — even when the agent rewrites the DOM, they still point at .cart-item--headphones or whichever heuristic survives the rewrite.
The whole point: you don't have to be in front of the machine that's building. Be on the train, on a phone, on someone else's couch — Peekable links go anywhere, and feedback comes back the way your agent likes it.
Whatever your loop looks like — solo with Claude, a tiny team, or a designer reviewing an agent's PRs — Peekable slots in without changing your stack.
Claude is building locally on your desktop. You're at lunch. Peekable lets you peek at the result on your phone, drop a few notes, and walk back into your office to a queued patch.
Send the live thing instead. They click the actual button that's wrong, leave a note, and your agent gets a selector — not a "make this more punchy" Slack DM.
Clients click around, leave sticky notes on the actual thing. No Loom recordings, no "can you make the logo bigger" Slack threads from screenshots — just structured feedback that you (or your agent) can triage in minutes.
QA finds a broken element on Safari iOS. They click it. Peekable captures the element, the browser, the route, the version. Your agent pulls the report and reproduces it on the first try.
Designed to be called by an agent, not just a human. Predictable flags, stable JSON output, no interactive prompts unless you ask. Works with Claude Code, Cursor, Codex, and anything that can shell out.
Run peekable proxy <port> and your local dev server gets a public URL routed through Peekable. No deploy, no SSH gymnastics — just send the link.
Reviewers click any element on the page and leave a note. Comments stick to selectors, not pixels.
Every peekable push bumps a version. Reviewers always see the latest, and feedback is pinned to the version it was left on.
Mark comments resolved from the CLI, then push the next version. The loop closes where you live — the terminal — not in another tab.
Each comment carries selector, DOM path, element text, computed styles, bounding box, route, push version, and reviewer name.
Tail new comments as they come in, while your agent is still working. Pipe them straight into the next prompt.
Drop-in instructions so Claude Code or Codex can create, push, and pull feedback as part of its loop — no orchestration code to write.
The hosted beta runs on Peekable's managed service. We're focusing the first batch on the CLI and support loop, so hosted peekable.fyi is the fastest way in.
We're keeping pricing simple while we ship. Get an early-access account today; we'll keep limits clear as the beta grows.
Join the early access waitlist. We'll send you an account, the beta guide, CLI install steps, and support notes as the beta evolves.