Skip to content
ConsultEvo

Relay.app Alternatives After Shutdown: Recover and Rebuild

Relay.app is no longer available for export or recovery. Relay’s current homepage says the service shut down in September 2026 and that accounts, workflows, and run history were permanently deleted. Its former export documentation now redirects to the shutdown notice.

If you downloaded a ZIP, build prompt, screenshot, table file, or run-history record before shutdown, those files can still provide evidence for a reconstruction. They do not restore the original workflow, connections, credentials, or deleted account. If you have no saved files, begin with business-process records and the process owner’s knowledge, and treat the result as a new specification rather than a recovered Relay configuration.

This guide explains what saved evidence is useful, how to choose a replacement platform, how to turn a prompt or screenshot into a tested workflow design, and how to control validation, duplicate writes, permissions, and cutover.

Relay.app is shut down: what can you still recover?

The current Relay shutdown notice is the most relevant source for present status. It says that Relay shut down in September 2026 and that accounts, workflows, and run history were permanently deleted. No current export interface, recovery API, or other verified route to generate a new Relay export was found.

That limitation is different from recovering files you already possess. Check local downloads, shared drives, team handover folders, password-protected archives, and email attachments. Preserve the original files before editing them, record who can access them, and assume that run data or table exports may contain customer information or other sensitive values.

For historical context, a July 20, 2026 Zapier article described account-tier shutdown dates, transfer credits, export timing, and the contents of a completed export. Those details describe the former migration period and are not current export instructions. The article reported August 15 for free accounts and September 14 for paid accounts, but the exact account-tier dates cannot now be independently confirmed from Relay’s live site.

A shut-down service cannot create a new export. Reconstruction starts with evidence already saved or with a newly confirmed description of the business process.

Inventory the files saved before shutdown

The historical Zapier article said a completed Relay workspace export could contain workflow folders with Markdown build prompts, workflow JSON, run-history CSV files, per-run JSON, and table CSV files. It also said screenshots were saved separately. These are historical descriptions of files that may exist in an export already downloaded; they are not steps that can be performed in Relay today.

Create an access-controlled inventory for every workflow. Record the original workflow name or ID, each filename, the apps involved, whether representative successful and failed runs are available, which connections must be reauthorized, and which details remain unknown. Keep credentials out of working notes and do not paste sensitive run data into an assistant unless the organization has approved that use.

  • Build prompt: Use it as the starting description of the intended outcome, trigger, actions, conditions, and branches.
  • Workflow JSON and screenshots: Inspect them for sequence, branches, formulas, and field mappings. Do not assume either is a portable workflow definition.
  • Run history and per-run records: Use them to compare real inputs, outputs, errors, and branch outcomes. A summary report is not a substitute for event-level records.
  • Table CSVs: Treat them as candidate source data for a separately planned import. Confirm column meaning, sensitivity, destination field types, and duplicate behavior first.

For each workflow, document the trigger event, stable source-record or event identifier, destination action, required fields, conditions, formulas, human review steps, and exception behavior. Mark every inference from incomplete evidence as unresolved until a process owner confirms it.

Choose a Relay.app alternative by workflow requirements

Choose the replacement from the workflow outward, not from a platform directory inward. Confirm the exact trigger, action, fields, authentication scope, branch behavior, plan requirement, access policy, and person responsible for maintenance. Directory size indicates broad platform coverage, not compatibility with a particular Relay workflow.

Platform Consider it when Verify before choosing Operational question
Zapier The exact app event and action fit the workflow and the team prefers a managed workflow service. Trigger, action, fields, authentication, plan access, filters, paths, and connection ownership. Zapier currently advertises 9,000+ apps. Who owns failed runs, app access, edits, and production approvals?
Make The team needs a visual scenario-building approach and its current modules fit the process. Exact app modules, mappings, limits, permissions, and operating model. Make currently documents more than 3,000 apps. Who will maintain scenarios, credentials, error routes, and changes?
n8n The team can evaluate its current integrations and deployment and maintenance responsibilities. Exact integration, hosting or deployment model, code requirements, security ownership, and operational support. n8n currently advertises more than 2,000 integrations. Who operates the deployment and responds when infrastructure or integrations fail?

The counts come from current vendor information: Zapier’s app documentation, Make’s app documentation, Make’s product overview, and n8n’s integrations directory. Counts can change and do not establish that a specific trigger, action, field, authentication method, or plan is available.

If implementation support would help while evaluating Zapier, see Zapier automation consulting. For a Make evaluation, see Make automation consulting. These are optional service resources, not proof of platform compatibility.

Rebuild from a saved prompt, or reconstruct the prompt first

Zapier’s July 2026 guidance said Relay’s workflow JSON was not directly compatible with Zapier and recommended rebuilding from a plain-language build prompt. That is historical migration guidance, not a currently tested import path. A saved prompt can describe intended behavior, but it cannot supply deleted credentials, undocumented formulas, missing branches, or the difference between a create and an update.

Zapier’s current Copilot guidance describes Copilot as open beta and warns that proposed workflows may require premium apps, higher plan tiers, or features unavailable to the user. Use it to draft and refine a candidate, then inspect every proposed app, event, action, mapping, condition, connection, and write.

A useful specification should contain:

  • Outcome: What business result should occur, and what must never happen?
  • Trigger: Which app, event, object, and source identifier start processing?
  • Inputs: Which fields are required, and how are values normalized?
  • Sequence: Which actions happen in order, and which depend on earlier results?
  • Rules: Which exact conditions, thresholds, allowed values, and branch outcomes apply?
  • Destination: Is the operation a create, update, upsert, notification, stop, or review?
  • Exceptions: What happens when data is missing, ambiguous, unauthorized, duplicated, or rejected?

For a complex prompt, separate stages with headings or XML-style boundaries, as recommended in Zapier’s Copilot guidance. If only screenshots remain, transcribe the visible trigger, ordered steps, branches, and field mappings first. Leave unknown behavior explicitly marked for owner review instead of allowing an assistant to fill it in silently.

Three reconstruction patterns with concrete validation gates

The following patterns are proposed implementation designs, not vendor-provided templates. They show how to turn different kinds of saved evidence into a testable workflow. Confirm the actual app modules and actions in the selected platform before building.

Trigger and input AI or processing job Validation gate Action and fallback
New lead record with source_object_id, email, message, and source timestamp. Classify message intent and return an owner-approved label, confidence where supported, extracted entities, and a review indicator. Require the source ID and email; reject labels outside the approved list; normalize the email; check for an existing record using a stable external ID; apply the owner-approved review threshold. Create or update the CRM record only after validation. Missing IDs, conflicting matches, unsupported labels, or uncertain results go to a named review queue.
Saved screenshot of a form submission workflow with visible branch conditions but no prompt. Transcribe the screenshot into an ordered specification. An assistant may draft the specification, but inferred steps remain unresolved. The process owner confirms whether each destination step creates, updates, searches, notifies, or stops. Test both matched and unmatched branch outcomes. Configure verified app actions against a test destination. Contradictory or invisible behavior becomes a review task rather than a production write.
Saved run-history CSV or per-run JSON containing source and destination observations. No AI is required. If used for summaries, constrain it to describing differences between historical and replacement runs. Compare one source event at a time. Separate the source event ID from each execution attempt and check for missing mappings, duplicate writes, and changed branch outcomes. Store the comparison in an access-controlled migration log. The maintainer investigates technical differences and the process owner approves acceptable behavior changes.

A lead-intake example illustrates the required order: verify the source ID and required fields; classify the message using an approved set of labels; validate the returned structure and label; then create or update the CRM record using the stable source ID. An unknown label, missing ID, ambiguous match, or uncertain classification should reach a human reviewer before any system-of-record write.

{
  "source_object_id": "lead-4821",
  "classification": "product_question",
  "confidence": 0.88,
  "extracted_entities": [],
  "requires_review": false,
  "destination_operation": "update_or_create_by_external_id"
}

The JSON is an illustrative output contract. Whether the selected AI step returns confidence, rationale, or extracted entities depends on the tool and configuration. The workflow owner must approve the allowed labels, required fields, review threshold, and destination operation before the output can authorize a write.

01Inventory evidenceThe maintainer preserves files, records filenames and sensitive content, and lists unknown behavior.
02Approve the specificationThe process owner confirms the trigger, intended outcomes, fields, branches, exceptions, and write operations.
03Draft the replacementThe maintainer uses the selected platform or assistant to propose verified app events, actions, mappings, and rules.
04Validate access and data controlsThe maintainer reconnects apps, checks write permissions and plan limits, validates fields, and configures duplicate and review handling.
05Test and approve cutoverThe owner reviews representative outcomes, assigns failure ownership, and approves production writes only after the test gate passes.

Translate branches and AI steps into operating rules

Use deterministic rules for exact checks such as a required field being present, a status equaling an approved value, a domain matching an allowlist, or an amount exceeding a threshold. Zapier documents Filters for stopping a Zap when conditions are not met and Paths for routing data to different outcomes. Filters are available on paid plans, so verify the selected plan.

Reserve AI for interpretation, such as classifying free text or extracting entities. Put parsing, allowed-value checks, and confidence or review rules after the AI step and before a consequential write. If the approved labels are product_question, billing, and other, reject every other value. Route missing fields, conflicting matches, low-confidence results, and sensitive changes to a named person.

Use a clear result contract: a valid classification with required fields proceeds to the approved destination action; a missing source ID stops the write and creates a review item; an uncertain classification enters a human queue with the source record and proposed label; and a destination error is recorded and assigned to the workflow maintainer.

Credentials do not transfer with a prompt or screenshot. Reconnect each app in the chosen platform and confirm that the connected account has the required write permissions. Also check premium-app requirements, organizational app-access policies, rate limits, and any plan restriction that could prevent the proposed workflow from running.

Prevent duplicate writes and preserve useful run evidence

Define the row grain before building deduplication. A source event is one business change or incoming event. An execution observation is one attempt to process that event. A destination write records an attempted or completed business-system operation. A daily or monthly report is an aggregate and must not overwrite any of those event-level records.

Preserve provenance where available: source system, source object ID, source event ID, source workflow ID, source run ID, target run ID, migration version, timestamp, and a payload hash or appropriately redacted reference. A proposed per-event key is source_system + source_event_id + workflow_migration_id. A proposed per-run observation key is source_run_id + target_run_id. These are implementation examples, not vendor-provided schemas.

Do not identify an event only by workflow name and date. That value can collide when several events run on one day, when retries occur, or when multiple destination operations belong to the same event. If several citations, model variants, or evidence records relate to one run, give each its own record rather than replacing a run-level value.

Concurrency changes the design

A lookup followed by create can still duplicate a record when two runs look up the same event before either writes. For concurrent processing, use a destination-supported upsert keyed by a stable external ID, or a database with an enforced uniqueness constraint and transactional conflict handling. If the selected tool cannot provide either guarantee, route collisions to review and record the attempted key instead of treating a lookup as proof of uniqueness.

Zapier documents unique run IDs, statuses, timestamps, Zap versions, and step-level inputs and outputs, but its available logs may have limited retention. Preserve durable business audit data separately. Zapier Tables documents CSV imports, fields, records, and workflow connections; the reviewed guide does not establish database-enforced unique indexes or transactional upserts, so do not assume Tables alone provides race-safe deduplication.

Test the replacement before production cutover

Use a test set that includes a normal new event, a duplicate event, a missing required field, an ambiguous AI result, a permission failure, and a destination error. For each case, define the expected result before running it: one intended write, a deliberate stop, or a review item. Test at the event grain first, then inspect execution observations and any aggregate reporting separately.

If saved Relay run records exist, compare destination effects, field mappings, and branch decisions one source event at a time. If they do not, have the process owner approve expected behavior and representative fixtures before the replacement writes to production. Start with a low-risk workflow and, where feasible, a test destination or review route.

Zapier documents replay, Autoreplay for some failed tasks, and custom error-handling options. These features do not prove unlimited retries, idempotent writes, exactly-once processing, or equivalence to a former Relay workflow. Assign an operational owner for alerts, access changes, failed runs, and edits, and keep a manual fallback for high-impact actions.

Review before rollout
  • A representative successful event produces the intended destination result.
  • A duplicate source event uses the verified uniqueness or upsert control and does not create an unintended second record.
  • Missing, ambiguous, and unauthorized inputs stop or reach the named review owner.
  • Required app permissions, organization policies, and plan access are confirmed.
  • Failure alerts, run evidence, and manual fallback responsibilities have assigned owners.
  • The process owner and workflow maintainer approve production cutover.

Frequently asked questions

Can I export Relay now?

No currently verified export route exists. Relay’s official notice says the service shut down and accounts, workflows, and run history were permanently deleted. The former export-documentation URL redirects to that notice.

Can Zapier import Relay workflow JSON directly?

Zapier’s July 2026 article said Relay JSON was not directly compatible with Zapier and recommended rebuilding from a saved prompt. Treat that as historical guidance, not a currently tested import flow.

What if I only have screenshots?

Transcribe the visible trigger, steps, branches, and mappings into a specification. Ask the process owner to resolve unknown create-versus-update behavior and unmatched branches, then test the result against a safe destination or review route.

Do app credentials carry over?

Do not assume they do. Reconnect and authorize every destination app in the chosen platform, and confirm the account has the permissions required for the intended action.

Are Make and n8n equivalent Relay replacements?

No equivalence is established by their directory counts. Verify the exact trigger, action, fields, authentication, plan, deployment model, and maintenance responsibilities for the workflow you need.