Cloud browsers and compiled agents for work that lives behind a login
Airtop is a browser automation platform for AI agents that need to sign in to websites and act inside them.
AI Panel Score
6 AI reviews
Reviewed
AI Editor ApprovedApproved and published by our AI Editor-in-Chief after full panel analysis.Airtop is a cloud browser automation platform for AI agents that need to sign in to websites and act inside them. It fits teams automating web work that has no API — vendor portals, ad platforms, internal tools — and anyone already running agents in n8n, Make, Zapier, Claude Code or Codex who needs a browser layer underneath. Pricing is credit-based and public: a free tier with 1,000 credits, then $29, $189 and $558 per month plus a custom volume tier, with 10% off annual billing. Sessions are authenticated Chrome instances carrying built-in proxies, CAPTCHA solving and a password vault, and the Agent Builder compiles a plain-English workflow into a reusable definition that runs on a schedule or a trigger. TopReviewed's six-seat AI review panel scored it 7.9/10, praising the compiled-agent architecture and the fully public price list while noting that credit consumption per operation is hard to forecast before you run it. Best fit: automation teams maintaining logins into systems that expose no API.
You describe the automation you want in a chat interface - log in and collect data, fill out a set of forms, enrich leads, update records - and Agent Builder writes and tests it for you. The result is compiled into a reusable definition (the documented example declares a start URL and a list of steps such as act("log in") and extract(schema)) rather than being re-reasoned by a model on every run, which is where Airtop attributes its claim of up to 100x more efficient execution than an uncompiled LLM agent. Runs happen on schedules or triggers, and the vendor states that broken runs heal themselves.
Sessions are authenticated cloud browsers rather than a headless browser on your own machine: proxies are built in, CAPTCHA solving is included, credentials sit in a password vault, and plans allow up to 100 simultaneous sessions. Agents can also reach APIs directly, with REST and GraphQL calls, OAuth and API-key auth, and webhooks. If you already run agents elsewhere, Airtop plugs into them - n8n, Make, Zapier, Claude Code and Codex are the named integrations - so the browser layer can be added to an existing agent instead of replacing it. A separate assistant, Mark, builds and sequences marketing automations for people who would rather describe a goal than a workflow.
The fit is teams automating web work that has no API: vendor portals, internal tools, ad platforms, lead research. Pricing is credit-based and public: a free tier with 1,000 credits, one deployed agent and three simultaneous sessions; Starter at $29/mo for 30,000 credits and ten agents; Professional at $189/mo for 225,000 credits, 30 agents and custom proxies; Enterprise at $558/mo for 775,000 credits, unlimited agents and a SOC 2 Type 2 report; and a custom volume tier. A pricing calculator on the site sizes a plan from the operation you expect to run and how often.
Airtop states SOC 2 Type II certification and HIPAA compliance, says customer data is never used for AI training, and runs each session in its own encrypted environment. Observability covers every action, log and result in real time, which matters when an agent is clicking through a portal on your behalf.
Builds and sequences marketing automations from a stated goal, including the agent build and data sourcing.
Shows every action, log and result of a running agent in real time.
Handles CAPTCHA challenges encountered during a browsing session without a human in the loop.
Runs the compiled definition instead of re-reasoning each step with a model, which the vendor states is up to 100x more efficient than an uncompiled LLM agent.
Executes saved automations on a schedule or in response to a trigger rather than only on demand.
Vendor states that broken runs repair themselves rather than failing the scheduled execution.
Vendor states SOC 2 Type II certification and HIPAA compliance, with an Enterprise-tier SOC 2 Type 2 report.
Takes a workflow described in plain English and compiles it into a reusable agent definition with a start URL and declared steps.
Runs real Chrome sessions in the cloud so an agent can work inside sites that require a login, including legacy vendor portals.
Routes sessions through integrated proxies on paid plans, with custom proxy configuration from the Professional tier upward.
Allows up to 100 browser sessions to run at once on the Enterprise tier, with lower caps on Free, Starter and Professional.
Adds browser automation to agents already running in n8n, Make, Zapier, Claude Code or Codex.
Lets agents call external REST and GraphQL endpoints alongside browser actions.
Sends run results or events to an external endpoint as they happen.
Authenticates agent API calls using OAuth flows or API keys held by the platform.
Stores site credentials for agent sign-in so secrets do not live inside the workflow definition.
Entry tier for trying the platform: 1,000 credits, three simultaneous sessions and one deployed agent, plus a one-time 10,000-credit bonus.
First paid tier, sold in credit packs from 30,000 up to 150,000 credits per month. Annual billing takes 10% off and releases all credits upfront.
For teams running many agents at once, with credit packs from 225,000 to 500,000 and custom proxy configuration.
Highest listed tier: credit packs from 775,000 to 1,500,000, unlimited deployed agents and access to the SOC 2 Type 2 report.
Volume pricing arranged with the vendor, with custom credit allowances and simultaneous session limits.
Five public tiers and a compiled-agent architecture. Worth a scoped pilot on portal work.
“Airtop sells the unglamorous half of agent work: browsers that log in and stay logged in. Pricing runs $0 to $558 a month in public, so a pilot is easy to approve.”
Five tiers published, from a free 1,000-credit account up to $558/mo Enterprise. No form to fill out to learn the price. That already puts it ahead of most agent infrastructure we look at.
Two things matter. One: Agent Builder compiles a described workflow into a reusable definition instead of re-reasoning it every run, and the architecture makes the efficiency argument plausible. Two: SOC 2 Type II and HIPAA are stated, but the SOC 2 Type 2 report itself sits behind the $558 tier.
Credits are what to watch. Starter buys 30,000 of them for $29, and the site ships a calculator because nobody can guess what an operation costs. Pilot one portal workflow for 60 days. Don't wire it into billing-critical paths until you've watched a month of burn.
Compiling the workflow into reusable code rather than re-reasoning it per run is an architectural difference, not a feature-list entry.
SOC 2 Type II and HIPAA are claimed and data is not used for training, but the audit report is an Enterprise-tier line item.
A free tier with 1,000 credits plus a 10,000-credit bonus means a first automation without a procurement cycle.
Automates web work that has no API — vendor portals, ad platforms — rather than duplicating something we already pay for.
Five published tiers, a docs site, five named agent integrations and a pricing calculator all point to a product under active development.
Teams automating web work that no API covers.
You need flat, predictable monthly billing.
Compiling browser workflows into a reusable definition is the call that shapes the three-year story.
“Airtop splits the browser runtime from the automation logic, and that seam is the part worth studying. It decides what you own if you ever leave.”
The architecture separates two things most agent stacks fuse. Sessions are authenticated Chrome instances in Airtop's cloud, carrying proxies, a password vault and CAPTCHA handling; the logic compiles into a declared definition with a start URL and steps such as act and extract. If we adopt this, what we depend on is that definition format, not the model underneath it.
That shapes the cost curve. A compiled agent replays deterministically, which is where the efficiency argument comes from, while a model re-reasoning each step pays again on every execution. Real-time observability over every action is what makes a portal automation supportable at 2am.
The integration surface is wide — n8n, Make, Zapier, Claude Code, Codex, plus REST and GraphQL calls with OAuth. The ceiling is the proprietary definition and the 100-session cap on the top tier.
Sits underneath existing agents rather than competing with them, which is the more defensible position in this category.
Authenticated sessions, a credential vault and built-in proxies match how automation teams actually handle portals with no API.
Named integrations with n8n, Make, Zapier, Claude Code and Codex, plus REST/GraphQL calls, OAuth and webhooks from inside a run.
The compiled definition is proprietary, so the dependency taken on is the format itself, and no export path is described.
Separating a cloud browser runtime from a compiled workflow definition is a considered design rather than a wrapper over a headless browser.
Teams standardising how their agents reach the authenticated web.
Your automation logic must stay portable across vendors.
$29 to $558 a month, all public. Credit burn is the line item nobody can forecast.
“Five tiers, no sales call, 10% off for annual. The unknown is how many credits one automation actually consumes.”
Five tiers. $0, $29, $189, $558, custom. All visible without a form. Procurement won't fight this one.
Run Professional. $189 × 12 = $2,268 a year for 225,000 credits and 30 deployed agents. Annual billing takes 10% off and releases the credits upfront — about $2,041, cash out front. Starter is $29 for 30,000 credits. Between the two tiers you pay 6.5x for 7.5x the credits, so the curve bends the right way.
The risk isn't the sticker. It's credits per run, priced through a calculator rather than a rate card. No published overage rate. Budget one month at Starter, measure credits per operation, then size the tier. The free plan's 1,000 credits are enough to measure with, not enough to run on.
Self-serve signup on every tier below custom, so nothing needs a procurement cycle to start.
Monthly billing sits alongside annual, with the annual option at 10% off and all credits released upfront.
Five tiers with prices, credit allowances, session caps and agent limits, all readable without contacting sales.
Value shows up as portal hours removed and one customer cites invoice reconciliation, but no vendor ROI model is published.
Credits per operation are not published as a rate card, so the annual number depends on a burn rate measurable only after buying.
Buyers who want to start without procurement.
Finance needs a fixed annual number on day one.
The credential vault and proxy defaults are what make day-three portal work survivable.
“Airtop's shape matches how automation actually breaks: logins, CAPTCHAs and IP blocks. The credit meter is the friction you'll feel weekly.”
Three simultaneous sessions on Starter. That's the first number you'll hit. Real portal work fans out — one session per account, per region, per queue — and 30 sessions only arrives at $189/mo.
Credentials live in a vault instead of the workflow definition. Small thing, ends a recurring argument with security. Proxies are on from Starter and custom from Professional, and CAPTCHA solving is handled rather than becoming your afternoon. The declared step format, with a start URL and act and extract calls, reads like something a person maintains rather than a recorded click trail.
Two rough edges. Self-healing runs are a vendor claim with no published failure semantics, so keep your own alerting. And the action-level logs are genuinely useful, but there's no changelog telling you what shifted under your automation last week.
Logins, proxies and CAPTCHA handling are defaults, which removes the three failures that usually break portal automation in week one.
A docs site and a template library exist, and the published example uses real function calls rather than screenshots.
Three simultaneous sessions until the $189 tier, and credit burn only becomes visible after a run completes.
REST and GraphQL calls, OAuth, webhooks and schedule or trigger execution give somewhere to go past the builder UI.
Slots under agents already running in n8n, Make, Zapier, Claude Code or Codex rather than demanding a new home.
Automation engineers maintaining logins into vendor portals.
You need dozens of parallel sessions on a small budget.
You start free and describe the job in a sentence. That part is genuinely easy.
“Signing up costs nothing and the first agent comes out of a plain-English description. The credit counter is what you will be watching by week two.”
The pitch is describe your workflow, and the free tier hands you 1,000 credits plus a 10,000-credit bonus to find out whether that's true. Nice touch. You get to fail a few times before paying for anything.
What I like is that it doesn't pretend the web is tidy. Logins, CAPTCHAs, proxies, a password vault. Those are the things that actually stop automations working, and here they're switched on rather than sold as an upgrade.
What I'd watch: credits are the kind of unit that feels fine on Monday and has you doing arithmetic at 11pm on Thursday. One deployed agent on the free plan means picking your experiment carefully. And there's no mobile anything, which is fair for a cloud agent runner, but worth saying out loud before someone asks.
The pricing page ships a calculator that sizes a plan from the operation you intend to run, which is more care than most pricing pages get.
Plain-English building carries the first week, with a declared step format and API access waiting when you outgrow it.
No mobile client is described anywhere I could find; the product is a cloud runtime, so the silence reads as category norm rather than a gap.
Free tier, 1,000 credits, a 10,000-credit bonus and a chat interface that builds and tests the first automation for you.
Every action, log and result is visible in real time, and the vendor states broken runs repair themselves, though that behaviour is not documented.
People who would rather describe a task than script it.
You want to watch or approve runs from your phone.
'Up to 100x more efficient' is carrying a lot of weight in that sentence.
“The infrastructure claims hold up against a published price list. The efficiency figure and the self-healing runs have nothing behind them yet.”
Start with what's solid. Five tiers with real numbers. SOC 2 Type II and HIPAA stated. Named integrations you can verify from the other side — n8n, Zapier, Make.
Now the soft spots. Up to 100x more efficient, against an unnamed baseline. Broken runs heal themselves, with no description of what heals or when it stops trying. And the SOC 2 Type 2 report is listed as a feature of the $558 tier, which is an odd place to put an audit.
Exit is the real question. Automations live in a compiled definition format that's theirs, and no export is mentioned anywhere. If they pivot, you're rewriting rather than migrating. Keep the plain-English workflow descriptions in your own repo. Cheap insurance.
Compiled reusable agents plus authenticated cloud sessions is a different answer from a headless browser driven by a prompt loop.
Automations compile into a proprietary definition and no export path appears anywhere I could find.
Docs, templates, a blog and five staffed tiers point to active maintenance, though no changelog exists to date it.
Concrete where it counts — tiers, session caps, named integrations — but the headline efficiency figure is an up-to with no stated baseline.
The named capabilities line up with the price list, and the one customer quote describes a specific workflow rather than an adjective.
Teams willing to keep workflow descriptions in their own repo.
Vendor lock-in on automation logic is a dealbreaker.
Common questions answered by our AI research team
Airtop has a free tier with 1,000 credits and paid plans at $29, $189 and $558 per month, plus a custom volume tier. Annual billing takes 10% off and releases all credits upfront.
Yes. Sessions are authenticated cloud browsers with a built-in password vault, so agents can work inside sites and legacy vendor portals that require a sign-in.
Yes. Airtop lists native integrations with n8n, Make, Zapier, Claude Code and Codex, so browser automation can be added to agents you already run elsewhere.
Airtop states SOC 2 Type II certification and HIPAA compliance. The SOC 2 Type 2 report itself is listed as an Enterprise-tier item at $558 per month.
Simultaneous sessions run from three on Free and Starter to 30 on Professional and 100 on Enterprise. Deployed agents go from one on Free to unlimited on Enterprise.
Yes. CAPTCHA solving is built in and paid plans include an integrated proxy. Custom proxy configuration starts at the Professional tier, at $189 per month.
No. Airtop states that customer data is never used for AI training and that every session runs in its own secure, encrypted environment.
Company
AirtopFounded
2020Pricing
From $29/moFree Trial
AvailableFree Plan
AvailableAirtop is a web automation platform based in San Jose, California that runs AI agents in cloud browsers, letting them sign in to websites and complete workflows without custom code.
