Share a live AI agent preview URL.
You're mid-edit and someone wants to see the agent. Instead of screenshots or 'it works on my machine,' run `a2a dev` — your agent goes live at a public URL you can paste into Slack. It's the real runtime, it hot-reloads as you type, and the link stays current. Send it to a teammate, a tester, or a stakeholder and get feedback before you ship.
public url · hot reload · real runtime · no deploy
A screenshot isn't a demo, and localhost isn't shareable.
When your agent only runs on localhost, showing it to anyone means screenshots, a screen recording, or walking them through it on a call. None of that lets a reviewer actually use the agent — call its tools, poke its frontend, hit the edge case they're worried about. And the moment you make your next edit, whatever you shared is out of date. Spinning up a real staging box for every work-in-progress is its own chore, and it bills whether anyone's looking or not.
`a2a dev` gives your work-in-progress agent a public URL.
Run one command and your agent is served at https://<agent>-devbox.a2acloud.io — a live, hot-reloading preview of the real runtime on a scale-to-zero cloud dev box. Send the link, keep editing, and let people actually try the agent before you deploy.
A public URL, mid-edit
Run `a2a dev` and your agent goes live at https://<agent>-devbox.a2acloud.io — a real, public address you can paste into Slack or an email. No deploy, no release build, no waiting for a pipeline. You're still editing; the URL already works.
Hot-reloads as you type
Save a file and the preview reflects it in about two seconds. The link you shared an hour ago now shows the fix you just made — your reviewer refreshes and sees your latest work, not a stale snapshot from the last time you exported anything.
The real runtime, not a mock
This is the actual agent running, not a screenshot or a clickable prototype. Testers can call its tools, drive its frontend, and hit real behavior — the same code that runs in production, so the feedback you get is about the agent, not about a stand-in.
Scale-to-zero, cost-efficient
The dev box sleeps after about five minutes idle and wakes on the next request. A preview link you leave open in a channel doesn't burn compute while nobody's clicking — it costs when it's used, which is what a shared demo link should do.
One link, many reviewers
Send the same URL to a teammate, a QA tester, and a stakeholder at once. Everyone hits the same live agent; you keep editing; nobody needs the repo, a local Python setup, or a build of their own to try it.
Ship when the feedback's in
The preview is a dev URL — great for demos and sign-off before you commit to production. When it's ready, deploy the agent and it gets its own permanent URL, an MCP server, an API, and a signed receipt per run. The preview is the try-before-you-ship step.
Screenshots and localhost vs. a live shareable preview URL.
Frequently asked.
How do I share a preview URL for an AI agent I'm still building?
Run `a2a dev`. Your agent starts on a cloud dev box and is served at a public URL — https://<agent>-devbox.a2acloud.io — that you can send to a teammate, tester, or stakeholder. You don't have to deploy or cut a release: the URL is live while you're still editing, and it hot-reloads as you save, so whoever opens it always sees your latest work.
Is the preview the real agent or just a mock?
It's the real agent runtime — the same code that runs in production, running for real behind the preview URL. It is not a screenshot, a clickable prototype, or a mock. Reviewers can actually call the agent's tools and use its frontend, so the feedback you get is about how the agent really behaves, at parity with production.
Does the preview stay in sync as I keep editing?
Yes. `a2a dev` hot-reloads in about two seconds when you save, so the URL you already shared reflects your latest change without you re-sending anything. A reviewer can leave the link open, refresh, and see the fix you made moments ago.
Won't a shared preview link cost money while nobody's using it?
No — the dev box is scale-to-zero. It sleeps after roughly five minutes of inactivity and wakes on the next request, so a link sitting in a Slack channel doesn't burn compute while it's idle. A dev session can run up to eight hours; it just costs when someone is actually using it, which is what a shared demo link should do.
How is this different from a production deploy?
The `a2a dev` URL is a dev and preview address — ideal for demos and feedback before you ship. A production deploy is the permanent step: the agent gets its own stable URL, an MCP server, an API, and an Ed25519-signed receipt for every run. Use the preview to get sign-off, then deploy when it's ready. See /cloud-dev and /quickstart for the full flow.
Related guides.
All guides live in the guides index.
Trust the receipt.
Run `a2a dev` and your agent goes live at a public preview URL — the real runtime, hot-reloading as you edit, on a scale-to-zero dev box that sleeps when idle. Send it to a teammate, tester, or stakeholder and gather feedback while you're still building. When it's ready, deploy the agent for its own permanent URL, an MCP server, an API, and an Ed25519-signed receipt per run. Preview first, ship with proof.