Pricing

Priced by the run,
not by the token.

Replay calls no model. A workflow costs what a browser costs to drive it — the same this quarter as the next, whether it runs once a day or ten thousand times an hour. Everything that runs on your own machine is free, and stays that way.

Save 20%

What you can have today, plainly. The Local tier is installable now and free forever. The hosted tiers — the store, hosted scheduling, fleet-managed Chrome — are in development: the prices here are the committed shape of the offer, not a checkout that works yet. The waitlist is how you hold a place in it, and nobody is billed before the thing they are billed for exists.

Local

For one machine

The whole runtime, on your machine, with nothing reporting home.

$0forever

No account, no card, no seat count.

Install the CLI
  • Record, validate and replay unlimited workflows
  • Deterministic replay — zero model calls
  • Scheduled replays, compiled into launchd or cron
  • Similarity-based repair proposals
  • Hash-chained audit log, on disk
  • MCP server over stdio
  • Capture, render, mirror and browse
  • Pinned Chrome for Testing resolution

Workflows are files. Commit them next to the code they test.

Pro

For one builder

When the run history needs to outlive the laptop that produced it.

$29$23 per seat / month

Billed monthly. Cancel anytime.Billed annually — two months free.

Join the waitlist
  • Everything in Local
  • Hosted store — workflows, runs, heal history
  • 25,000 replayed steps a month
  • Scheduled replays and webhooks
  • Repair queue with reviewable diffs
  • 90-day run and audit retention
  • Email support

Extra steps at $2 per 1,000, or the run simply queues.

Team · Most chosen

For a fleet

A registry a team can trust, and an audit trail somebody will actually read.

$79$63 per seat / month

Billed monthly. Minimum 3 seats.Billed annually — two months free.

Join the waitlist
  • Everything in Pro
  • 250,000 replayed steps a month, pooled
  • Shared registry with two-key repair review
  • SSO (SAML, OIDC) and role-based access
  • Audit export — JSON, S3, or your SIEM
  • Chrome pin managed across the fleet
  • 1-year retention · priority support

Seats are for humans. Agents and CI runners are not billed as seats.

Enterprise

For your org

Your network, your browsers, your compliance calendar.

Let's talk

Annual agreement. Volume pricing on steps.

Contact sales
  • Everything in Team
  • Self-hosted control plane in your VPC
  • Unlimited steps, dedicated browser fleet
  • Bring your own KMS key for the audit chain
  • Custom retention and data residency
  • Security review, DPA, named engineer
  • 99.9% uptime SLA

The runtime is the same binary in every tier. Only where it stores things changes.

What you are billed for

Three meters. Two of them can't move.

Every other browser agent bills you for thinking. Replay does not think — it resolves, acts and verifies against conditions a human already approved. That makes the two meters below deterministic, and it makes the third one optional.

A replayed step

One resolve → act → verify cycle against a real browser. No tokens, no context window, no retry storm when a page gets chattier. The hundredth run of a workflow costs exactly what the second one did.

A proposed repair

Relocation is similarity scoring, not inference — free, deterministic, and not a prompt-injection surface. Proposing a repair costs nothing beyond the step it belongs to. Reviewing it costs a minute of your attention, which is the point.

A labelled capture

The only place a model appears, and it is off unless you turn it on. Labelling runs once per captured page at build time, billed per capture, and ships beside a free deterministic labeler that already scores 94.7% — so the model has to earn it.

Compare

Line by line.

Feature Local Pro Team Enterprise
Runtime
Workflows UnlimitedUnlimitedUnlimitedUnlimited
Replayed steps a month Local only25,000250,000 pooledUnlimited
Model calls during replay 0000
Static pre/post condition validation
Pinned Chrome for Testing Per machinePer machineFleet-managedFleet-managed
Scheduled replays & webhooks Via launchd/cron
Review & audit
Repair proposals with diffs
Heal history retained across runs
Two-key repair approval
Hash-chained audit log On diskHostedHosted + exportBYO KMS
Run & audit retention Yours90 days1 yearCustom
Team & access
Seats 1 machine13+Unlimited
Shared workflow registry
SSO (SAML, OIDC) & RBAC
MCP server for agents stdiostdiostdio + hostedstdio + hosted
Deployment Your machineCloudCloudYour VPC
Support
Channel GitHubEmailPriority, 1 business dayNamed engineer
Uptime SLA 99.5%99.9%
Security review & DPA On request

Questions

The ones worth asking.

Is the free tier a trial?

No. Local is the product, not a sample of it — the same runtime, the same replay engine, the same audit log, with no cap on workflows or runs. What the paid tiers add is a place to keep the results that isn't one person's laptop, and the review and access controls a team needs around them. If you never want that, you never need to pay.

What exactly is a “replayed step”?

One entry in a compiled workflow: verify the pre-condition, resolve the target by similarity, act, then poll until the post-condition holds. A login workflow is typically 4–8 steps. A step that fails closed still counts; a step that never ran because an earlier one stopped the run does not. fs stats shows the exact number for every workflow you have, locally and for free, before you decide on a plan.

Do you charge for model tokens?

There are none to charge for. Replay makes zero model calls by design, and relocation after drift uses similarity rather than inference. The single optional model — region labelling on captured pages — runs at build time, is billed per capture, and is disabled until you enable it. Nothing in the automation runtime gains a model dependency when you turn it on.

What happens when we go over the included steps?

Overage is $2 per 1,000 steps on Pro and $1.20 per 1,000 on Team, and you can cap it at zero — in which case runs queue until the next period instead of billing. Nothing is throttled silently and nothing is deleted; you get a notification at 80% and again at 100%, in the tool and by email.

Can we self-host?

The runtime is always self-hosted — it is a CLI on your machine or your runner. What Enterprise adds is a self-hosted control plane: the store, the registry, the repair queue and the audit chain running inside your VPC, with the browser fleet on your own instances and a KMS key you hold for the audit log.

Are agents and CI runners billed as seats?

No. A seat is a human who logs in — reviews repairs, approves workflows, reads audit trails. An MCP-connected agent replaying an approved workflow consumes steps, not a seat, and so does CI. Charging per agent would price the thing the product exists to make cheap.

How do we cancel?

One click, from billing, effective at the end of the period. Your workflows were always files in your repo, so there is nothing to export to keep running them — drop the hosted store and the same fs replay keeps working against your local one. We will export your run history and audit chain as JSON on the way out regardless.

Start where it's free

Install it before you price it.

Record one workflow, replay it twice, and read the audit log. Ten minutes, no account, nothing to uninstall but a directory.