Relay.app Shut Down and Deleted Every Customer Workflow: The Vendor-Continuity Lesson Nobody Plans For

On 14 September 2026, Relay.app stopped existing. Not paused, not wound down gradually. Deleted. The company's homepage now carries a single paragraph: "Relay.app has shut down. All accounts, workflows, and run history have been permanently deleted." For the ops teams who had wired their invoicing, their onboarding and their support routing through it, that sentence is what's left of their infrastructure. Free accounts were cut off on 15 August, paid accounts on 14 September, and once those dates passed, the data went with them. There is no archive to download now, no read-only mode to revisit.

TL;DR

• Relay.app, an AI workflow automation tool founded in 2021, shut down in September 2026 after raising about $8.1M and shipping real product through early 2026.
• Free accounts were deleted on 15 August 2026; paid accounts followed on 14 September. All workflows and run history are permanently gone.
• The company never stated a public cause, and we won't invent one. The lesson is structural: a venture-backed tool can delete your production automations overnight.
• The vendor-continuity checklist at the end is the part to save.

What actually happened, and when?

Relay.app launched in 2021 out of San Francisco, founded by Jacob Bank, with a pitch that should sound familiar to anyone in this space: workflow automation with a human in the loop and AI handling the fiddly bits. It raised about $8.1M across two rounds, the last a $3.1M seed in October 2023, with Khosla Ventures and Andreessen Horowitz among its backers. Public profile data pegged revenue at roughly $2.5M a year and traffic around 235,000 monthly visits.

The product was real. Through early 2026, Relay shipped native Tables, MCP server support, more than 200 integrations, human-in-the-loop approval steps, and an affiliate and certified-partner programme. This was not a landing page that never launched. It was a working platform with paying customers and an ecosystem forming around it.

Then, in the summer of 2026, the lights went out. Free accounts lost access on 15 August at 23:59 PT. Paid accounts got a short extension to 14 September at 23:59 PT, with access at no charge through that window and a pro-rated refund on unused annual time, according to the migration guides competitors published while helping people move. After those dates, accounts and their data were permanently deleted. The homepage is explicit about it, and even points to an archived July 2026 version so you can see what the product looked like while it was still alive.

Who actually gets hurt when a platform deletes its data?

The ops teams who treated a startup as infrastructure. A workflow that auto-routes a sales lead, generates an invoice, or escalates a support ticket stops being a nice-to-have the moment it's been running for a year. It's a dependency. And the person who built it is often not the person now asked to replace it. The original builder moved on, or the reason step four had that one weird filter lives only in a Slack thread from 2024.

If you build with no-code tools for a living, this one is aimed directly at you. You are exactly the person who wires a workflow into a startup's tool, forgets about it, and only reopens it when something breaks.

This is the specific failure Relay exposes. The shutdown itself is sad but ordinary; startups close all the time. What's not ordinary is the combination: a venture-backed tool, shipping real product, building an ecosystem, that still vanished in a way that left customers holding nothing. There was no open-source fallback, no self-host option, no export of workflow definitions into a portable format. When the company went, so did the configuration.

Honestly, the most damning detail isn't that Relay closed. It's that the company pointed customers to an archived homepage instead of an export tool.

What do we actually know about why it shut down?

Not much, and I'm not going to dress that up. Relay never published a cause. The announcement was short and final. TechCrunch reported in August 2026 that members of the team were joining Google's Chrome team, with Bank hinting at "ambitious plans to help you work with AI in Chrome." That tells you where some people went, not why the company closed. A talent move, a shift in direction, a market squeeze; any of those could be true and none are confirmed.

The cause doesn't change the lesson anyway. The point for anyone reading this is structural: the thing you built your business on can disappear for reasons you never get to see, and you don't get to negotiate with a liquidation.

What should a vendor-continuity checklist look like?

This is the part to save. Before you commit a workflow to any platform, run it through these checks.

• Export before you commit. On day one, not day 300, find out whether the tool can hand back your data and your workflow definitions. If the answer is "we can do a one-off export if you email support," that's a favour, not an export plan. Test it for real: export an actual workflow, re-import it somewhere, confirm it runs. A CSV of your table rows is not a backup of your automations.
• Read the exit clause like it's part of the product. A "no liability, no guarantee of data retention" line in the terms is the vendor telling you in advance what they'll do if they wind down. If there's no stated retention commitment, assume the worst. Ask in writing what happens to your data on acquisition, shutdown or bankruptcy. Most small vendors will agree to a clause that returns your data in a usable format within a defined window if the service is discontinued. If they won't, that's information.
• Keep the workflow portable. Document the logic outside the tool. A workflow's value is rarely the drag-and-drop canvas; it's the decision logic, the field mappings, the edge cases. Write those down somewhere you control. A well-maintained Notion page beats a perfectly configured Zap you can no longer read.
• Respect the rebuild cost. "We can rebuild it" is the most expensive sentence in ops. A build that took two days originally carried weeks of accumulated tweaks, exceptions and fixes. Rebuilding from memory, under a deadline, against a tool whose syntax you don't know yet, runs to a multiple of the original effort in my experience. Price that in when you pick a vendor.
• Check the funding reality. A company with $8M raised and $2.5M in revenue that's still shipping is a going concern until the day it isn't. Venture-backed tools live or die by their next round. That's not a reason to avoid them. It's a reason to treat them as what they are: rented infrastructure with a shorter average lifespan than your mortgage.

Where does this sit in the wider agent-builder consolidation?

Relay isn't a one-off. The agent-builder category has been consolidating hard over the past year. We've watched tools get acquired, pivot, and quietly stop shipping. The pattern is familiar: a category gets hot, money flows in, dozens of near-identical products compete on "one-click AI," and then the weaker ones get bought for the team or run out of road.

What's new in Relay's case is the clean cut. Most shut-down products hand customers a downloadable export or a read-only period. Relay handed customers a paragraph and an Internet Archive link. That's a lower bar than the industry had set, and it's a fair warning: nothing about "venture-backed" implies "your data will outlive us."

So here's the blunt version. Don't mistake shipping cadence for stability. Don't mistake an ecosystem page for a commitment. And before you wire anything you can't afford to lose into a tool with a runway, find out how you'd get it out. The moment to ask is while the vendor is still answering emails, not after the homepage has been replaced with a goodbye.