What does a real export look like when a platform closes?
Relay.app is the current worked example, and it is a fairer one than most. Paid accounts close at 23:59 PT on 14 September 2026, after which the workspace, user account, and app credentials and tokens are deleted. The team was acquihired by Google and joined the Chrome organisation; the product was shut down rather than sold, so there is no successor to migrate to.
What makes Relay instructive is not that it handled this badly. Its migration documentation is unusually candid — more so than most vendors manage while they are still trading. Relay states plainly that different workflow builders use different JSON structures, so its JSON export is "not perfectly transferrable to other tools without AI interpretation of the structure first," and that the JSON "won't work through their native importers." Its actual recommended path was to export a prompt.md description of each workflow and ask an AI to rebuild it in Zapier, n8n or Make.
That is the point. Candour does not create portability. Relay told the truth about its export and the export still went nowhere. If a vendor honest about the limitation cannot hand you a portable workflow, a vendor vague about it certainly cannot. Treat disclosure and capability as separate things.
What are the three levels of portability that actually exist?
Strip away the marketing and there are three, not a spectrum. Most readers only ever have access to the first two.
| Level | What it is | What it gives you | What it loses |
|---|---|---|---|
| (a) Proprietary JSON | The platform's own export of its own structure | A faithful record of what you built. Keep it as the reference of record. | Imports nowhere. No mainstream platform's native importer accepts another's format. |
| (b) Natural-language description | A prose or Markdown summary of the workflow, of the prompt.md kind |
What an AI can genuinely rebuild from, in a different tool, at reasonable speed. | The exceptions. Retry behaviour, approval timeouts, what runs when a field is empty, escalation paths — a description flattens the happy path and drops the rest silently. |
| (c) Open notation | BPMN 2.0, and in principle the Open Workflow Specification | Real portability between engines that implement it. | Availability. BPMN is portable only inside the BPM product family — Camunda, Flowable, Appian, Pega. If you are on a SaaS integration platform, this level is not offered to you at all. |
The failure mode is rating (a) highly because it is machine-readable and (b) poorly because it is prose. In practice (b) is the migration path and (a) is the evidence you check it against. The gap between them — the exception handling — is what to write down separately, by hand, while the working system is still in front of you.
Why do the standards that exist not help you?
They exist, and they are real, governed and actively developed. They are simply in a different market from the one most readers are buying in.
The nearest thing to a standard is the Open Workflow Specification, currently at v1.0.3, formerly Serverless Workflow, governed under the CNCF and the Linux Foundation; serverlessworkflow.io now redirects to open-workflow-specification.org. Its implementations are Apache EventMesh, SonataFlow, Lemline, Quarkus Flow, Synapse and Zigflow. Every one of those is a cloud-native developer orchestration engine, not a SaaS integration platform. There is no Zapier adoption, no Make adoption, no n8n adoption. If your automations live in a connector-based tool, the specification is not something you can ask your vendor to support in a renewal negotiation — it is not in their category.
BPMN 2.0 is the only broadly implemented portable notation, and its reach is bounded the same way: it works within the BPM world, across Camunda, Flowable, Appian and Pega. That is genuine portability, and it comes attached to a much heavier class of platform than most teams are shopping for.
So the standards do not close the gap. Implying that a specification exists and portability is therefore solved is the mistake to watch for when someone waves a logo at you.
What should you ask before you sign?
Six questions, in the order they cause pain. Take them into the call and write down the answers.
- What exactly does the export contain? Ask for a sample file, not a description of one. Ask specifically whether it includes error branches, retry policy and timeout configuration, or only the primary path.
- Does it include execution history? Usually not. Definitions and runs are different exports and often only the first exists. If you have a retention obligation, this is the answer that decides the contract.
- Can I test a restore in the first month? Not "can I download an export" — can I take the export and stand the workflow up somewhere else, while I still have a functioning original to compare against. Do this in month one, not in the month you need it.
- What happens to credentials on termination? Relay deletes tokens at shutdown. That is tidy on their side and leaves standing authorisations on yours, in every system you connected. Ask what is deleted, when, and what you are expected to revoke yourself.
- Where does processing occur? See below. This is the question that eliminates options rather than ranking them.
- What is the notice period on discontinuation? Ask for it in writing. Frame it against the Relay timeline: signups closed 16 July 2026, paid accounts are gone 14 September 2026. That is roughly eight weeks from the first public signal to deletion of the workspace — and the public shutdown announcement came four weeks before the end, not eight.
Where does the workflow actually run?
This is the portability question people forget, and it is the one that bites hardest. A workflow you cannot execute inside your own boundary is not portable in the sense that matters when a regulator asks, a customer security review lands, or a data-residency requirement appears mid-contract. "We can export it" is no help if the only place it can run is someone else's cloud.
| Platform | Runs inside your boundary? | What is verified |
|---|---|---|
| Zapier | No — cloud-only | Zapier staff state on the official community that "Zapier is a cloud-based application and is not available on-premises." |
| Workato | No — cloud-only | The On-Prem Agent is not on-premises deployment. It creates "a secure connection from within your private IT environment to connect to Workato cloud" — an outbound TLS WebSocket only. Orchestration and processing remain in Workato's cloud. Data residency US and EU (Frankfurt). |
| Make | Unconfirmed | Ships on-prem agent installers for Windows Server and Linux, Java 11+. The page is a download hub with no architecture documentation. It looks like the same connectivity-bridge pattern, but that is not documented and we will not assert it — ask them. |
| n8n | Yes | Genuinely self-hostable. Note the licensing edge: n8n staff state on their own community forum that "for enterprise clients we can generate special license types that will work in an airgapped environment, but we do not offer these for the Community Edition." Air-gapped n8n therefore runs free without enterprise features, or needs an Enterprise contract. |
| Camunda | Yes | Publishes an official air-gapped installation guide: mirror Docker images to a private registry, host the Helm chart internally, override image repositories. More than ten images. Licensing in disconnected environments is not addressed in that document. |
If processing locality applies to you, work this table first. It shortens the shortlist faster than any feature comparison, and unlike features it is not negotiable later.
Is platform risk a real consideration or a hypothetical one?
Two data points suggest it is live rather than theoretical, and they point the same way.
Gartner published its inaugural Magic Quadrant for Business Orchestration and Automation Technologies (BOAT) on 24 October 2025, defining the category as platforms that "unify process orchestration, connectivity and agentic features to enable enterprisewide automation." Appian was named a Leader; Camunda, Flowable and Pega also appear. An inaugural quadrant means a market has consolidated enough to be named — and named markets are where acquisitions, repositioning and discontinuations happen.
At the same time, Forrester's 2026 automation predictions hold that "less than 15% of firms will turn on the agentic features in intelligent automation suites," attributed to governance and ROI concerns. The capability everyone is selling is largely switched off while the market underneath it reshapes — which is precisely the condition under which products get consolidated away.
Meanwhile, cross-platform migration is still served only by third-party converters such as migromat and data.page, plus manual and AI-assisted rebuild services. No native importers exist, and there is no sign that is about to change.
Frequently asked questions
Is there a standard format for workflow definitions?
Not across mainstream automation platforms. Every platform is proprietary and no interchange format exists between them. The Open Workflow Specification and BPMN 2.0 are real, but neither is implemented by the SaaS integration platforms most teams use.
Can I move workflows from one platform to another?
Only by rebuilding. No mainstream platform ships an importer that accepts a competitor's export. Third-party converters and AI-assisted rebuild from a description are the working paths.
Does the Open Workflow Specification solve this?
Not for most teams. It is genuinely CNCF-governed at v1.0.3, but its implementations are cloud-native developer orchestrators. Zapier, Make and n8n have not adopted it.
What is a "workflow manifest"?
Not a recognised industry term. A targeted search returns no definitional sources. Where a vendor uses it — including us — it is that vendor's own coinage, and the right response is to ask what it contains.
Which platforms can run inside my own network?
n8n and Camunda. Zapier and Workato are cloud-only, whatever the On-Prem Agent naming suggests. Make's on-prem installers are undocumented architecturally.
Where this leaves you
Portability is not a checkbox anyone can currently tick, so stop shopping for it and start testing for it. Get a sample export before you sign. Restore from it in the first month. Write the exception handling down separately, because no export format captures it well. And decide early whether processing locality is a requirement, because it is the one constraint you cannot retrofit. Everything else in a migration is work; that one is a wall.
Related
- Governed workflow automation — the complete guide
- Relay.app closes on 14 September 2026: what to do with the time left
- Running client workflows: isolation, credentials and blast radius
- How to build an audit trail for AI agents
BarzelOps approaches this from the deployment-locality end: cross-system workflow automation with durable state, human approval checkpoints, tenant isolation, and a portable Governed Workflow Manifest intended to run in cloud, VPC, on-premises or air-gapped deployments. To be explicit about that term, since this article has just spent several hundred words on vendor coinages: Governed Workflow Manifest is Real Biz Digital's own name for a portable definition format. It is not an industry standard. No such standard exists — which is the whole argument above, and applies to us exactly as it applies to everyone else.
In practice
Automation you can leave, with the approvals and the evidence intact.
A workflow platform is only as safe as your ability to leave it. BarzelOps runs cross-system workflows with durable state, human checkpoints and signed evidence, exports them as a portable manifest, and migrates Relay workflows with a dry run before cutover.
In forceRelay.app closes at 23:59 PT on 14 September 2026
BarzelOps
Governed workflow automation across the systems that run the business.
- Durable, idempotent execution: a timeout is retried once, never filed twice.
- Human approval checkpoints that pause the workflow and resume it.
- Isolation per entity or client, signed evidence receipts and a portable manifest; HubSpot, Xero, Gmail, Google Drive and Slack.
Free tier: 100 calls a dayPaid plans from $19 a monthLive on MCPize
BarzelVault
The AI action firewall: decide what an agent may do before it does it.
- Approval thresholds and policy checks enforced before execution; human approvals that expire and escalate.
- Cryptographically signed audit receipts: trigger, inputs, policy version, approver, outcome.
- Credential isolation, spend and action limits, and an emergency kill switch.
Free tier: 10,000 calls a monthPaid plans from $199 a monthLive on MCPize
Enterprise: written quote by email within two business days. No sales call.
Sources
- docs.relay.app — migration guidance and JSON export limitations, as published; relay.app shutdown notice.
- TechCrunch, AI automation startup Relay shuts down, staff joins Google's Chrome team, 17 August 2026.
- Open Workflow Specification v1.0.3 (formerly Serverless Workflow), open-workflow-specification.org; CNCF / Linux Foundation project pages and implementation list.
- OMG, Business Process Model and Notation (BPMN) Version 2.0.
- Gartner, Magic Quadrant for Business Orchestration and Automation Technologies, 24 October 2025 (inaugural).
- Forrester, Predictions 2026: Automation.
- Zapier Community — official staff response on on-premises availability.
- Workato documentation, On-prem agent and Security; Make, On-premise agents download page.
- n8n community forum — staff statement on air-gapped licensing and Community Edition.
- Camunda, Air-gapped installation guide.
- migromat and data.page — third-party workflow conversion tooling.
This article is for information and does not constitute legal or procurement advice. Product details, deployment options and dates are as published at 2 September 2026; verify each against the vendor's own current documentation before relying on it in a contract.