Relay.app Shuts Down 14 September: 7 Alternatives Compared on Price

Relay.app goes dark on Monday 14 September 2026. Paid accounts keep free access until 14 September 2026, 23:59 PT, and after that the workspace and everything inside it is deleted.

Free accounts have already gone, deleted after 15 August 2026, 23:59 PT. Relay announced the wind-down on 16 July. New signups and upgrades stopped the same day.

So there is one decision to make this week. It is not which canvas you like the look of. It is where the human approval step goes. Everything else follows.

Relay had that pause built in.

Seven places to go, ordered by who they suit. Every price came off the vendor's own pricing page on 11 September 2026. Every approval fact came off their own documentation the same day. Mine is last, because I run it.

Reading this after 14 September 2026? Relay is gone and any data you did not export went with it. Skip to the table; the comparison holds.

What Relay.app's pause did

Relay's trick was the pause. An AI step drafted the email, the reply, the summary. Then it stopped and waited. A person approved it, edited it or killed it, and the run carried on.

Around that sat the usual machinery, per Relay's docs. AI steps with agentic tool use, MCP servers, sequences, tables, scheduled triggers, webhooks. All of that is available in five places on this page.

The pause is what made Relay safe to point at a customer.

The flows worth rebuilding first are the ones pointed at a customer: outbound, collections, the inbox. That is exactly the work you cannot leave running unwatched, which is why the pause is the first thing to check in a replacement.

Good news, mostly. Every tool below can stop and wait for a person. The terms are where they differ, and the terms are the whole decision: a premium app in one place, an Enterprise plan in another, a piece marked legacy in a third.

How I compared them

Five questions, asked of every tool. They sort the list faster than a feature grid:

• Is there a human approval step, and on which plan? AI is table stakes now. The question is whether a run can stop, show a person what it wrote, and wait for an answer.
• What does the meter count? Per step, per run, per seat or per month. That one line explains most of the surprise invoices in this category.
• What is the starting price? Read off the vendor's own page in September 2026, in the currency they publish it in.
• Who does the rebuilding? You, on a canvas, for the rest of the year. Or somebody who already built and tested the thing.
• How much of your export survives? Relay hands you JSON and CSV. Nothing here reads either.

I read pricing pages, product documentation and Relay's shutdown notice on 11 September 2026. I ran no tests, and I am not going to pretend otherwise. Every approval claim below is sourced to the page it came from, listed at the end.

Quick comparison

Prices are what each vendor's pricing page showed on 11 September 2026, per month. n8n and Activepieces publish annual-billing rates. n8n prices in euros and no conversion is applied. The approval column records what each vendor's pricing page or documentation states, read 11 September 2026. Gravity's row is my own product. Nothing here was tested.

1. Zapier: best for the connector nobody else has

Zapier counts a task for every step in a Zap. AI steps, code and MCP calls draw from the same pool, at higher rates. Relay metered you in steps too, so the arithmetic you already did carries over.

The catalogue is the argument for coming here.

Zapier lists more apps than anything else on this page. If your Relay flow touched something obscure, this is the route most likely to still have it on the shelf.

What the per-step meter does to a Relay rebuild

Count the steps in your longest flow. Multiply by runs a month. Then read the task slider before you pick a plan. It steps 100, 2K, 10K and 100K. An eight-step flow running fifty times a day lands above the 10K mark, at 12,000 tasks.

Zapier's approval step has two strings attached

Zapier ships a Human in the Loop tool. Its Request Approval action pauses the Zap and puts the data in front of one or more reviewers. They approve it, decline it, or change it, and the Zap continues from there. Notifications go by email, Slack or another Zap. Timeouts send reminders. A Decision field lets later steps filter on the answer.

Two conditions come with it, per Zapier's help page. It is a premium app. Reviewers need Zapier accounts, with the Zap shared to them. So the person signing off needs a login first.

Best for

A Relay flow that touched one app nothing else connects to.

Pricing

Free, 100 tasks a month. Professional from $19.99 a month, Team from $69, Enterprise on request. As of September 2026, and Zapier pricing explained unpacks what a task really is.

Free tier

Yes, 100 tasks a month on two-step Zaps.

Verdict

Closest match for a Relay flow that lived on one rare connector. You will pay for it a step at a time, and your approvers will need logins.

2. Make: best if the bill was the whole complaint

Make runs one self-serve paid plan sized by a slider. Nine dollars a month buys 5,000 credits. For a scenario that fires a few times an hour, that is the smallest bill on this page.

Make's own pitch is the visual one, and it holds up. Routers, filters and iterators are drawn as shapes, so branching reads at a glance. The same drawing is harder to hand over than a flat list of steps.

The pause costs more than the plan

Make sells a Human in the loop app. Its Create a review request module stops the scenario until somebody approves or rejects on a review page Make hosts. That app is on the Enterprise plan.

Not on the $9 plan, then.

Lower down you build the pause yourself: park the row in a data store or in something like Airtable, wait for a human to mark it, and let a second scenario pick it up. It works. It is also a small project, and this week was supposed to go on the rebuild.

Free gives you 1,000 credits a month with that 15-minute floor attached. Paid opens at $9 and climbs the slider from there. Enterprise is quoted separately, and the review app lives up there with it. Those were Make's published numbers in September 2026.

3. n8n: best if you have a technical person and liked the AI steps

n8n is the closest thing here to Relay's ambition. AI steps, agentic tool use, MCP, and a canvas designed for long graphs. It assumes somebody on your side enjoys this work.

The meter is the fairest in the category. One execution per run, whatever the node count. Cloud Starter is 20 euros a month billed annually, for 2,500 executions. Pro is 50 euros for 10,000. Business is 667 euros for 40,000 self-hosted. Look hard at that last gap before you plan for growth.

Where the approval lives

n8n documents a Send and Wait for Response operation. It sits on the messaging and email nodes: Gmail, Slack, Outlook, Teams, Telegram, Discord, WhatsApp and n8n Chat. The workflow pauses until a person approves, disapproves, or fills in a form. There is a second human-in-the-loop path for AI tool calls. An agent can ask before it picks up a tool.

On paper that is the nearest copy of Relay's pause on this page, and it costs nothing extra. The trade is that you wire it yourself.

With a technical owner on hand, n8n gets you back to the same capabilities with the least detour. 9 best n8n alternatives covers the other direction, for the week you decide the canvas itself was the problem.

4. Activepieces: best open source without a server

Activepieces is n8n's model with a kinder meter. Open source, a visual builder, one price whether it runs on their servers or yours. A credit covers a whole run, so a twelve-step Relay rebuild costs the same as a two-step one.

Gravity's builders publish agents on Activepieces, so I read its docs more than most people do. What I would warn a Relay refugee about is the catalogue. It is smaller than Zapier's and Make's. You meet the gap on the day one specific app goes missing. The Plus plan also stops at five seats.

The Approval piece, and its small print

Activepieces has an Approval piece. You add Create Approval Links, then Wait for Approval, and the flow stops until somebody clicks. Paused time does not count against the run timeout, which matters when your approver is asleep.

The caveat is on the label. As of September 2026 the pieces page calls it Approval (Legacy), which is an odd word to meet on the step your customer emails depend on. It still works. I would read that label as a reason to check the piece again in six months, not a reason to skip it.

Best for

A Relay workspace that wants open source and no server to own.

Pricing

Free, 100 credits a day. Plus $16 a month billed yearly for 10,000 credits, then $0.007 a credit. Team $166 a month. Ultimate on request. As of September 2026.

Free tier

Yes, and it refills every day.

Verdict

Cheapest home for a long Relay flow, with the most legible bill on the page. Read the Approval piece's small print before you move the customer-facing one.

5. Gumloop: best when the AI steps were the point

Ask what was actually doing the work in your flows. If the answer is the model, and the wiring around it was just wiring, Gumloop is built that way round: any step can be an AI step, and that is where the dials live. Pick the model, watch the spend, cap each agent.

There is also a knowledge store your flows can read from, a command-line tool and an MCP server. So one developer can drive it from a terminal while everybody else stays on the canvas.

Approval happens at the agent, not the step

Gumloop documents Human in the Loop for its agents. You write the rule into the agent's Abilities: any email leaving your own domain, for example. The agent then halts, shows you the tool it wants to call, the arguments and its intent, and waits. You answer in the app, by Slack DM, or in the chat thread.

That is supervision of a colleague more than a step-level gate. For an outbound flow it lands in roughly the same place.

The 8% on top of your own tokens

Gumloop lets you plug in your own model keys. An 8% orchestration fee applies when you do, on top of what the model provider already bills you. Work that number out before you move a high-volume Relay flow across.

Pro is $37 a month and carries 20,000 credits. Enterprise is quoted on request. Both read in September 2026. If unlimited seats matter more than a free tier, start the trial today. The weekend before 14 September is too late to learn a new canvas. Best Gumloop alternatives has the wider field.

6. Lindy: best if the flows were really your inbox

Check whether your flows were ever really workflows. Some of them were one person's email wearing a workflow costume. Meeting follow-ups. The same three replies every morning. A Slack channel nobody else reads.

Lindy is an assistant rather than a canvas, which is why it ranks here. It sits in Gmail, Slack and your meetings, and it drafts in something close to your voice. Nothing about it asks you to draw a graph first.

The confirmation switch is per action

Lindy's documentation is specific about the pause. Any action with side effects (sending an email, updating a record, creating a calendar event) can have Ask for Confirmation turned on. The agent then stops, and the action waits as a draft in the task view until you approve it there or from the email Lindy sends you. Safe mode applies the same treatment to every side-effect action at once.

That is the least wiring of anything on this page.

The per-seat pool is where this gets decided

You pay per seat and the allowance pools across seats. The meter runs from 2 credits for a small ask up to 250 for a bigger one, and a research report can take 1,000. So one heavy week from one person can empty what six people share.

Best for

One person, or a handful, whose Relay flows all started in an inbox.

Pricing

Plus $29.99, Pro $99.99 and Max $199.99 per seat a month. Enterprise on request. As of September 2026.

Free tier

A trial only.

Verdict

Least work of any route here, as long as the inbox was the real workload. The per-seat maths decides everything after that, and best Lindy alternatives is the page to read when it breaks.

7. Gravity: best when the deadline leaves no time to rebuild

This is my product, so discount me accordingly.

It sits last for that reason. The argument for it is narrow: your workspace disappears on 14 September 2026, and everything else here starts with a build job you do not have time for.

You describe the job your Relay flow used to do, in a prompt box, in ordinary English. What comes back is a finished agent from the catalogue, published by an independent builder who keeps it working. Connect the accounts involved, answer its setup questions, then it runs on schedule. No canvas, and no weekend spent redrawing something Relay already knew how to do.

The pause you are about to lose is the default here. Customer-facing steps wait for you to approve them, so nothing reaches a client and no money moves without a person saying yes. Agents pass listing checks before they appear in the catalogue. A run that fails on a confirmed platform error, with no usable output, is refunded under the refund policy; third-party outages and agent quality are excluded.

The first agent is free and asks for no card, which is the part that matters this week. Hand over the flow you would miss most, watch one run, and Relay is still alive beside it for the comparison until the 14th.

For the head-to-head instead of the field, I wrote Gravity vs Relay on the narrower question: a pause you supervise against a job that comes back finished.

Who fixes it the week after you move

The builder who published the agent owns the repairs, not you. Every other route here hands you the rebuild this week and the maintenance for the rest of the year. Here that work belongs to the person whose income depends on the agent still running next month.

Best for

A Relay flow that existed to finish one recurring job: chasing the invoice, qualifying the lead, collecting the document nobody ever sends.

Pricing

First agent free, no card. Then Autopilot at $20 a month worldwide, ₹1,999 a month in India, with a lot of usage included. India also has Minipilot at ₹399 a month with a smaller usage allowance. Buy more usage if you run out. As of September 2026, on the pricing page.

Free tier

Yes. One agent, free.

Verdict

The only route on this page with no rebuild in it, which is worth more in a shutdown week than in a normal one. Cheapest AI agent platforms puts the number in context.

Moving your export: a 30-minute checklist

Relay's export covers workflows, sequences and MCP servers as JSON, your AI prompts, your run history, and tables as CSV, all documented at docs.relay.app. Nothing on this page imports any of it. Treat the export as documentation of what you built, not a migration file.

1. Export today, before the weekend. Paid access ends 14 September 2026 at 23:59 PT and the workspace goes with it. Pull the JSON, the CSV tables, the prompts and the run history in one sitting.
2. List the flows that cannot fail. Odds are only a handful actually matter. The rest were experiments, and a shutdown is a decent excuse to let them go.
3. Rebuild that shortlist, and nothing else. Start with the one that touches a customer. That is the one whose approval step you have to recreate on purpose, on the terms in the table above.
4. Run the rebuild in parallel. Relay still works until 14 September, so for a few days you can compare outputs side by side. After that, compare against the real thing: the invoice that got paid, the reply that actually went out.
5. Close the account properly. Revoke Relay's access to your Google, Slack and Notion accounts from those apps' own security settings, then delete the account. Access grants outlive the products that asked for them.

Thirty minutes covers the first two steps.

Step three is where the week goes, which is why the list in step two should be short.

Which one should you choose?

Start from the approval step. All seven can pause for a person now, so the terms are the decision. Zapier charges premium for it and wants every reviewer to hold an account. Make puts the review page behind Enterprise. n8n and Activepieces hand it over free and leave you the wiring. Gumloop supervises the agent instead of the step. Lindy turns it on per action, or across the board with safe mode. Gravity pauses the customer-facing steps for you by default.

A missing connector settles the question early. Zapier has the catalogue, and you pay for it a task at a time.

Was the bill the whole complaint? Then Make at $9 a month is the smallest honest number on this page, with the caveat that its review page is priced out of reach.

A technical owner who liked Relay's AI steps should look at n8n first. It gives the most capability per euro, on the fairest unit in the category. Activepieces is the same idea with a bill you can predict and no server to own. Between those two, the tiebreaker is who maintains the box.

Gumloop suits the team where the model did the real work, once you have modelled the 8% fee. Lindy suits the inbox in disguise. Watch its per-seat maths for a month before you add a third seat.

My bias, stated plainly. For a recurring business job with an outcome attached, I would pick the route where nobody rebuilds anything. That is Gravity. The first agent is free, and 14 September is closer than it looks. For the longer version of that argument, AI agent vs workflow automation and the glossary both go deeper than a comparison table can.

FAQ

Sources