Relay.app Shut Down: What Happened and What Users Lost
Relay.app shut down in September 2026. Free accounts were deleted after August 15 and paid accounts after September 14, and the site now says all accounts, workflows and run history “have been permanently deleted”[1]. If you’re looking for it today, there’s nothing left to log in to and nothing left to export.
Most pages about the shutdown still tell you to export before the deadline. The deadline is gone. What’s left to sort out is what happened, what was lost, and what to check before client work goes on any platform again, ours included.
What happened to Relay.app?
Relay.app announced on July 16, 2026 that it was closing, turned off signups the same day, and deleted accounts in two waves: free on August 15 and paid on September 14. A month after the notice, founder Jacob Bank said he was joining Google as VP of Product for Chrome[2][3].
- July 16. Bank posts the notice. New signups and free-to-paid upgrades are turned off. Free users get 30 days. Paid customers get 60 days of free usage with extra steps and AI credits, and annual customers get a pro-rated refund[2].
- August 15, 11:59 p.m. Pacific. Free accounts end.
- August 17. Bank announces he’s “(re)joining Google as the VP of Product for Google Chrome” and that “Several of my colleagues from Relay.app are joining the team as well”[3]. TechCrunch reports it the same day[4].
- September 14, 11:59 p.m. Pacific. Paid accounts end.
- Today. relay.app is a single page that opens with “Relay.app has shut down.” The old docs address sends you to the same page[1].
Sixty days is short for this job. Pipedream gave its Workflows users about six months to do the same thing, which I covered in the Pipedream shutdown guide. Relay.app’s paying customers had two.
Why did Relay.app shut down?
Nobody outside the company knows, because Relay.app never said. The shutdown notice gave no reason, and Bank’s Google announcement a month later doesn’t mention the shutdown at all. Anything more specific than that, including every explanation you’ve read in an alternatives roundup, is a guess.
- Known. Bank and several colleagues joined Google’s Chrome team[3]. Google had bought his previous startup, Timeful, in 2015[4]. Relay raised a $5 million seed round led by Khosla Ventures in 2022[5].
- Not known. Whether Google paid anything for the company, the product or the technology. Neither Bank’s post nor TechCrunch’s report calls it an acquisition[3][4]. How many people moved, and what revenue or runway looked like, are also unstated.
I’d resist reading a failed business into it. A founder taking a senior job at Google and bringing part of the team ends a small company whether or not the product was working. That’s the uncomfortable part for customers: a tool you depend on can go away for reasons that have nothing to do with how good it is.
What Relay.app users lost
Everything stored in Relay.app: workflows, run history, tables and the saved connections to other apps. Relay offered an export during the wind-down, but it produced files for reading and rebuilding. No other tool could load them, so a workflow that wasn’t rebuilt somewhere else before the cutoff no longer exists.
The export was one download with a folder for every workflow, sequence, MCP server and table. Each folder held a plain-language build prompt, the run history as a CSV and the workflow’s structure as JSON. It could take up to 24 hours to generate, and the emailed download link expired after 48[6].
- The JSON had no importer. Zapier’s own migration guide says it “isn’t directly compatible with Zapier’s structure”[6]. The build prompt was the useful file, because you could paste it into another tool’s AI builder.
- Connections didn’t travel. App credentials and tokens were deleted at the end of the wind-down, so every account had to be reconnected by hand wherever you landed[6].
- Approval steps were the hard part. Relay.app let a workflow stop and wait for a person to approve something, fill in missing data or pick a path. Most tools don’t have a matching step for each of those, so they had to be redesigned, not copied.
- The record of past runs. If you needed to show what an automation did for a client in June, the run-history CSV in that export was the only copy.
Can you still recover anything from Relay.app?
No. Relay.app’s own page says the data was permanently deleted, and it offers no recovery window and no contact[1]. If you downloaded the export before your cutoff, you have build prompts to work from. If you didn’t, you’re rebuilding from what the automations left behind in other apps.
- Look for the zip. Search your downloads folder and anyone else’s on the team. The emailed link is long dead, but a saved file still has every build prompt.
- Rebuild from traces. Sent emails, Slack messages and the rows a workflow wrote to a spreadsheet or CRM show what it did, in what order, and how often.
- Close the access you granted. In Google, Slack, HubSpot and anything else you connected, remove Relay.app from the list of authorized apps. Relay deleted its side. Yours is still worth tidying.
- Hand a build prompt to an assistant. Paste one into an AI assistant connected to your new tool. On TaskJuice that’s Claude or ChatGPT connected to our MCP server, which builds a draft you review before anything goes live. The MCP setup guide covers the connection.
What to check before client work goes on any platform
Ask five things: whose accounts hold the connections, where client payments land, where the data lives, whether workflow definitions can leave, and how much notice you’d get. The answers decide how much of a shutdown you’d have to rebuild and how much simply stays where it already was.
I build TaskJuice, an automation platform for agencies, so here’s how we answer our own list, including the answers I’m not proud of.
- Connections. Each client gets a separate workspace, and on our Growth plan and up the client signs in to their own apps from a portal. The accounts are theirs. If we disappeared, they’d reconnect them somewhere else and nobody would be chasing a password.
- Client payments. On the Scale plan you bill clients through your own Stripe account. The money and the customer records sit in Stripe under your name.
- Data. Tables export to CSV. A client database is the client’s own Postgres database, which we connect to and don’t hold.
- Workflow definitions. This is our gap. There’s no export file for a workflow today. Every version is kept in a history you can switch back to, and an assistant connected over MCP can read a workflow’s steps back to you, but neither is a file on your disk.
- Run history. Kept for 3 to 90 days depending on your plan, then deleted. If a client needs a longer record, have the workflow write it to a table or a spreadsheet as it runs.
- Notice. No vendor’s promise here is worth much, mine included. Relay.app’s 60 days was decided the day it was announced. Plan as if you’ll get 30.
The contract side, who owns the account, the credentials and the logic, belongs in your statement of work. I went through the clauses in Who should own client automations.
Frequently asked questions
Is Relay.app still working?
No. Free accounts ended on August 15, 2026 and paid accounts ended on September 14, 2026. The website is a single shutdown notice, and the old documentation address redirects to it.
Did Google buy Relay.app?
Not as far as anyone has said. Jacob Bank joined Google as VP of Product for Chrome and several colleagues went with him. Neither his announcement nor TechCrunch’s report describes an acquisition of the company or the product.
Did Relay.app customers get refunds?
Annual customers were promised pro-rated refunds. Paid customers also got 60 days of free usage with extra steps and AI credits during the wind-down[2].
Can I import a Relay.app export into another tool?
No tool loads it directly. The JSON describes the workflow, and the build prompt is written to be pasted into another platform’s AI builder, which then rebuilds it step by step.
Relay.app was a good product, and the people behind it handled the ending decently: a clear notice, refunds, an export. Its customers still lost every workflow they didn’t rebuild in 60 days.
What I take from it is to keep the pieces that are expensive to recreate, the connections, the billing and the data, on accounts that belong to you or your client, and to treat the workflow itself as the part you might have to rebuild. The same question is live right now for anyone on OpenAI’s canvas, which I wrote up in Agent Builder Shutdown: You Built a Workflow, Not an Agent.
[1] Relay.app has shut down, Relay.app: relay.app
[2] Jacob Bank, shutdown announcement, LinkedIn: www.linkedin.com/posts/jacobbank_i-have-some-sad-news-to-share-relayapp-activity-7483561834340380672-2dJ8
[3] Jacob Bank, “I have a new job”, LinkedIn: www.linkedin.com/posts/jacobbank_update-i-have-a-new-job-and-im-very-excited-activity-7495092819376656384-xg0x
[4] AI automation startup Relay shuts down, staff joins Google’s Chrome team, TechCrunch: techcrunch.com/2026/08/17/ai-automation-startup-relay-shuts-down-staff-joins-googles-chrome-team/
[5] After selling his last startup to Google, this founder now wants to automate mundane tasks with Relay, TechCrunch: techcrunch.com/2022/10/12/after-selling-his-last-startup-to-google-this-founder-now-wants-to-automate-mundane-tasks-with-relay
[6] Relay.app is shutting down: How to export your workflows and move to Zapier, Zapier: zapier.com/blog/relay-alternatives/