Relay.app Shutdown: Migration Options and Checklist

If your Relay.app workspace is still accessible, save its export before choosing a replacement. Then shortlist Make, n8n or Zapier by the workflow behavior you cannot afford to lose—not by a promise of one-click migration. If access has already ended, start with files you previously saved and records in your connected apps; do not plan around recovering the Relay account.

Relay.app shutdown dates: which path applies?

Relay's announced free-account deadline was August 15, 2026 at 23:59 Pacific Time. Paying customers retain access through September 14, 2026 at 23:59 Pacific Time. Its notice says accounts and associated data are permanently deleted after the applicable deadline. On September 8, the free deadline has passed and the paid window remains open. Relay's shutdown notice.

For a still-accessible workspace, request the export now. Relay documents up to 24 hours to prepare it and a 48-hour download window after the email arrives. Do not assume that download window extends account access beyond shutdown. Save and open the archive promptly. Official export instructions.

If you missed your window, inventory existing local exports, approved documentation and downstream business records. These may help reconstruct the process, but they cannot prove that you recovered every Relay setting or unfinished run. Pause consequential automation you cannot safely reconstruct and give a named person responsibility for manual continuity.

Preserve the files, then inventory the behavior

Relay's workspace ZIP includes Workflow, Sequence and MCP (Model Context Protocol) server definitions, per-run JSON, Markdown rebuild prompts, run-history CSV and applicable Table CSVs. Workflow screenshots are a separate export. The JSON is not a universal native-import format; a specialized converter is a separate route, discussed below. Relay export formats.

Keep an access-controlled original and work from a copy. Open representative files, reconcile the workflows and tables you expected, and inspect sensitive content before sharing anything with a builder or AI assistant. Never paste credentials into a build prompt. A generated description can help rebuild a flow; it is not proof that the replacement preserves its behavior.

Copy this inventory for each workflow. Retire obsolete automation instead of paying to migrate it, and prioritize active processes that send messages, change customer records or have pending human decisions.

Separate “exported” from “ready to run.” A Table CSV does not establish that relationships, field semantics or approval state will behave the same elsewhere. Likewise, an old run record is evidence for a test case, not an instruction to replay every action it contains.

Make, n8n or Zapier: choose a migration route

These are editorial starting points, not a performance ranking. Evaluate the route that fits your hardest required behavior and your future maintainer. Use the broader platform comparison if that decision is still unclear.

Make: selective conversion, followed by completion

The Relay → Make Migrator describes conversion of supported steps into paused scenarios and new Data Stores. Its listed coverage includes selected Gmail actions, Airtable record-change triggers, common schedules, HTTP requests, branches and fixed delays. Other app actions and AI steps, webhook/form/RSS triggers, Gmail filters and thread deduplication, loops, sequence calls and advanced delays require manual work. Unsupported steps remain visible as TODO modules.

That is a dedicated conversion service, not evidence that Make's ordinary importer accepts arbitrary Relay JSON. Keep each result paused while you inspect mappings, reconnect accounts and finish the TODOs. Make also documents AI-assisted rebuilding and app reconnection; generated configuration still needs acceptance tests.

Approval-heavy teams must check access before committing. Make's native Human in the Loop app is documented as an invited Enterprise closed beta. Do not assume a normal paid account includes it. If your design uses another review mechanism, prove reviewer identity, edited values, rejection and timeout behavior explicitly. An email notification alone is not an approval gate.

n8n: separate the builder from hosting responsibility

n8n documents managed Cloud and self-hosted options, with plan or edition choice separate from deployment. Choosing its builder does not require operating servers. If you do self-host, budget an owner for upgrades, backups, secrets, monitoring and recovery; hosting the workflow engine yourself also does not make external app or model calls local.

Its human-in-the-loop tool review can pause selected AI Agent tool calls for approval or denial. That specific capability does not by itself reproduce every Relay review-and-edit workflow. For a general business approval, design how proposed values, reviewer changes and the authorized action are connected, then test it. See n8n vs Make when hosting and custom logic are the deciding factors.

Zapier: test the reviewer's access and edited outputs

Zapier's Human in the Loop is a paid-plan tool for pausing a Zap for human input or review. Its Request Approval instructions require reviewers to have access to the shared Zap. Edited values appear in separate output fields, which must be mapped into later actions.

Crucially, decline behavior can be configured to continue or stop the run. For a workflow where rejection must prevent a send or write, configure that boundary deliberately and test it. Verify the selected plan supports your actual reviewer arrangement before buying. Use Make vs Zapier for the wider builder and purchase comparison.

Prove approval and duplicate handling with one workflow

The following is an illustrative acceptance scenario, not a migration we performed: an email triggers an AI draft, a human edits the reply and proposed CRM fields, and approval permits a CRM update plus an outbound message.

Rebuild this representative process with writes redirected to test destinations. Keep production credentials and live triggers out of the trial until its permissions and behavior have been reviewed. Retain expected values and downstream IDs, not just a green workflow status.

Use a stable event identity and a record of each completed side effect. Where a destination supports idempotency keys, use them; otherwise reconcile its records before retrying. A check-then-send sequence can still race under concurrent runs, so use an atomic claim or serialize that work. If delivery remains uncertain, stop and assign reconciliation instead of blindly sending again.

Reconnect apps under the intended production owner, with the least access the workflow needs. Make's migration guidance says connections must be re-established; Relay's notice says its stored credentials and tokens are deleted with the account. Make reconnection guidance, Relay data and privacy notice. Test expired credentials and the backup operator's access before calling the migration complete.

Cut over with one production writer

1. Record the event watermark, completed downstream actions and unresolved runs. Give each pending approval, delayed action and partial run an explicit disposition: finish, cancel, or reconstruct under a named owner.
2. Pause the old triggers while Relay remains accessible. Drain or deliberately stop in-flight work before handing over. If Relay access has ended, reconstruct the boundary from retained evidence and downstream records; treat unknown work as unresolved.
3. Move webhook delivery or enable the new schedule from the agreed boundary. Reconcile arrivals during the handoff; a pause must not silently discard them. Only one system should own each production side effect.
4. Monitor the first real outcomes against source events and destination records. Stop new writes on a mismatch and investigate before extending the migration to other workflows.

Rollback is conditional: Relay can only be a fallback while the account remains available and a safe handback has been reconciled. After shutdown, the fallback is containment and controlled manual continuity—not restarting an export. Preserve evidence, identify affected records and assign recovery. Sent messages cannot simply be undone; other corrections may require an app-specific compensating action.

Before committing to a plan, list required app actions, reviewer access, history retention, expected retries, external AI bills and operating time. Compare those requirements using the workflow pricing worksheet, rather than treating old Relay steps as interchangeable with another platform's units. If hosting remains undecided, use the builder-versus-hosting guide.

Choose only after the critical workflow passes its acceptance cases and a second operator can explain how to pause it, resolve an uncertain write and resume safely.