The practical promise of Relay.app is straightforward: you can automate a process end to end, but you decide exactly where a human has to sign off before it continues. That single design choice is what separates it from automation tools that run everything unattended. If a workflow is about to send a customer email, issue a refund, or push data into a CRM, Relay.app lets you insert a checkpoint where a person reviews the work first. The automation does the tedious parts; a human keeps authority over the consequential ones.
This is a workflow automation product in the same broad family as other tools you'll find under AI workflow automation, but the emphasis is different. Rather than optimizing purely for lights-out, no-touch execution, Relay.app treats human judgment as a first-class step type alongside its AI and app actions.
What Relay.app actually is
Relay.app is a web-based platform for building automations that connect apps, run AI steps, and pause for human input when needed. The vendor describes it as an AI automation platform where you "describe what you want to automate" and the system helps assemble the workflow. In practice you get two ways to build: a chat-first interface (which the site notes is powered by Claude Opus 4.8) for drafting a workflow from a description, and a visual drag-and-drop editor for refining it step by step. Both feed the same underlying workflow, so you're not locked into one mode.
There is no mobile app documented; it runs in the browser. That's normal for this category, where workflows execute on the server regardless of what device you started them from.
The features that matter, and when you'd reach for each
Human approval and review gates
This is the reason to look at Relay.app in the first place. You can drop an approval checkpoint into any workflow so that an action, the vendor cites examples like payments and outbound communications, waits for a person to confirm before it fires. There are also data-input forms for cases where a human needs to add or correct information mid-flow, not just approve or reject. The benefit is control over irreversible or reputation-sensitive actions. You reach for this whenever the cost of a wrong automated action is high: sending money, contacting a customer, changing a record that other systems depend on. If your process has any step where you'd normally say "let me just double-check this before it goes out," that step becomes a native part of the automation instead of a manual detour.
AI steps with a choice of model providers
Relay.app includes AI steps for common jobs, extraction, classification, and writing, plus custom AI steps where you can pick the provider. The site names Anthropic, OpenAI, and Google Gemini as supported providers. The practical value here is twofold. First, you can pull structured information out of messy inputs (parse an email into fields, classify an incoming request, draft a reply) without writing code. Second, provider choice matters because different models have different strengths and costs; being able to select one per step means you're not forced into a single vendor's tradeoffs. You'd use these steps when the input is unstructured language, an email body, a form free-text field, a document, and you need it turned into something a downstream app can consume.
One detail worth calling out: the platform lets you "control what the model can see" with access limits on AI steps. For teams handling sensitive data, being able to scope what an AI step is exposed to is a meaningful guardrail rather than a cosmetic one. There is also an AI-powered review feature that can assist a human approval step, though the human validation is still required, so the AI assists the reviewer rather than replacing them.
A flow-control model built for real processes
The builder supports conditional paths, loops, dynamic waiting, and path merging, along with sequences, tables, and MCP servers. In plain terms: your workflow can branch based on data, repeat over a list of items, wait for something to happen before resuming, and then bring separate branches back together. This matters because real business processes are rarely a straight line. An onboarding flow might branch on plan type; a lead-routing flow might loop over multiple records; an approval flow might need to wait days for a response. The support for MCP servers signals that Relay.app can connect to tools and data sources through that emerging standard, which is relevant if you're building AI-driven steps that need external context.
Triggers and integrations
Workflows can start from a range of triggers: app events (the site gives Gmail, HubSpot, and Stripe as examples), scheduled/time-based runs, manual button clicks, webhooks, mailhooks, form submissions, RSS feeds, and batch triggers. On the connectivity side, Relay.app advertises integrations with 200+ apps. The variety of trigger types is what makes it usable across departments, a scheduled trigger suits a recurring report, a webhook suits an event from a custom system, a mailhook suits an inbound-email process. The breadth of connectors determines whether the tool fits your existing stack, so if your critical apps are on the list, the human-in-the-loop model becomes genuinely usable; if they aren't, you'll be leaning on webhooks to bridge the gap.
Where this fits in day-to-day work
The clearest fit is any repetitive process that includes a judgment call. A few realistic patterns:
- Inbound triage: an email or form submission triggers a workflow, an AI step classifies and summarizes it, and a person approves the routing or the drafted response before anything is sent.
- Finance and billing sign-off: a Stripe or CRM event kicks off a flow that prepares an action (a refund, a communication), then holds at an approval gate so someone with authority confirms it.
- Data enrichment with a check: a new record triggers AI extraction to fill in fields, and a human reviews the enriched data before it's written back to the source of truth.
- Scheduled operational tasks: a time-based trigger runs a recurring process, with a review step only on the exceptions that branch off via conditional logic.
The common thread is that automation handles volume and consistency while a person handles the small number of decisions that carry real risk.
What it costs
Relay.app uses a freemium model. The pricing published on the site is organized around two usage meters, monthly steps and AI credits, plus the number of users. The free tier is a genuine working plan rather than a trial, with lower limits on the same capabilities.
| Plan | Price (annual billing) | Users | Monthly steps | AI credits/month |
|---|---|---|---|---|
| Free | $0 | 1 | 200 | 500 |
| Professional | $19/month | 1 | 750 | 2,000 |
| Team | $59/month | 10 included | 1,500 | 2,000 |
| Enterprise | Custom | Custom | Custom | Custom |
The free plan includes access to the 200+ connectors, the visual editor, and free testing of automations, which makes it realistic to build and validate a workflow before paying. The Team plan adds shared workflows and shared connections, the collaboration features that matter once more than one person maintains automations. Enterprise adds custom usage limits, custom integrations, priority support, training and agent workshops, and SOC2 and GDPR compliance. The site indicates annual billing carries roughly a 50% saving versus monthly; the figures above are the annual rates. Because the plans meter both steps and AI credits, the right way to size a plan is to estimate how many step executions and AI operations your workflows will consume per month rather than counting seats alone.
Who should consider it, and who shouldn't
Relay.app suits operations teams, founders, and business teams that want automation but can't hand over full control of certain actions. If your processes are compliance-sensitive, customer-facing, or involve money, the native approval gates are the differentiator worth paying for. It also suits people who prefer describing a workflow in plain language and then refining it visually.
It's a weaker fit if you need pure, unattended high-volume throughput with no human touchpoints, in that case the approval model is overhead you won't use, and the step-based limits on lower tiers may bind sooner than you'd like. It's also worth checking the 200+ integration list against your specific stack before committing, since the value of the human-in-the-loop design depends on being connected to the systems where those decisions actually happen. You can compare options across the tools directory or read more on the blog before deciding.
Questions people ask about Relay.app
Does Relay.app have a free plan?
Yes. The free plan costs $0, supports one user, and includes 200 monthly steps and 500 AI credits per month. It offers the same core capabilities as the paid plans with lower limits, including access to the 200+ app connectors, the visual workflow editor, and free automation testing.
What makes Relay.app different from a fully automated workflow tool?
Its defining feature is built-in human-in-the-loop control. You can insert approval checkpoints and data-input forms into any workflow so a person reviews, confirms, or enriches information, for example before a payment or an outbound communication, before the automation proceeds.
Which AI models can Relay.app use?
Its custom AI steps support multiple providers, with the site naming Anthropic, OpenAI, and Google Gemini. The chat-based workflow builder is described as being powered by Claude Opus 4.8. There are also templated AI steps for extraction, classification, and writing.
How many apps does Relay.app integrate with?
The vendor states integrations with more than 200 apps, with Gmail, HubSpot, and Stripe cited as examples. Workflows can also be triggered by schedules, manual buttons, webhooks, mailhooks, form submissions, RSS feeds, and batch triggers.
How is Relay.app priced beyond the free tier?
Paid plans are Professional at $19/month and Team at $59/month (annual billing), with an Enterprise tier at custom pricing. Plans are metered by monthly steps and AI credits, and the Team plan adds shared workflows and connections for up to 10 users.
