Buy it once
There are no servers behind AgentCron, so there is nothing to fund with a subscription and nothing to churn out of. One payment, and it keeps working whether or not we do.
-
Free
Freeforever
One project, and the entire safety half of the product. No account, no card, no trial clock.
- One project, with as many agents on it as you like
- Every supported CLI, chosen per agent: Claude Code, GitHub Copilot CLI, agy, and plain shell jobs
- Scheduling, chaining and mail-triggered runs
- The review gate and the full diff review with explanation cards
- The fail-closed approval gate
- The command log, with per-CLI capture coverage
- Blocking questions and declared plans
- The control CLI agents talk back through
- The cost view, per agent per day, with its own coverage line
- Diagnostics export, so a bug report contains what is needed to answer it
The safety mechanisms are the reason anyone would tell a friend about this, so putting them behind a paywall would be self-defeating. Diagnostics export is here for the same reason turned around: it is what a bug report needs to be answerable, and charging for it would mean the reports that reach us are the least useful ones.
-
Pro
$89one-time
The same product at the scale where it starts paying for itself: every repo and client folder at once, each run in its own environment, and an agent that can reach you on your phone.
- Everything in Free — the entire safety half, unchanged
- Unlimited projects
- Isolated per-worktree environments — a collision-checked port block, database name and compose project each
- MCP both ways: connect the servers you already run, and publish AgentCron's own tools to your agents
- The phone bridge — answer an agent's question from anywhere, through a bot you own
- Every ready-made agent pack, and new packs as they ship
Start on Free — see release status Not on sale yetOne payment. Every 1.x update included. No subscription. Thirty days money back, no questions — and if the price ever drops, everyone who paid more gets the difference back.
The arithmetic, in the open
Against the bill you already pay. An agent CLI subscription — the thing you must already have for AgentCron to be any use at all — can cost more per month than this app costs once: at the tiers built for heavy agent use, Claude's Max plans and Google's Ultra both do, and of the three vendors only Copilot's heaviest tier comes in under it. $89 buys the layer that decides what those tokens produce while you are not watching.
Spread it the way you would a subscription. Kept for three years, $89 is about $2.50 a month. No hosted automation product can sell at that number, because hosted products carry servers. This one deliberately carries none — that is not a discount, it is the architecture.
Against your own evening. One reclaimed hour a week is four hours a month; $89 divided by four is about $22. If an hour of your time is worth more than that, the app has paid for itself before the second month — and the chores it runs are precisely the ones that were eating your evenings.
Why it is not a subscription
There is no hosted anything. No servers, no inference, no storage. A monthly fee would be renting software that runs entirely on your own machine, which invites the question what am I renewing? every month — correctly.
You are already paying for an agent CLI. This is an add-on to a budget that exists, not a new line item, and that caps what it can reasonably charge.
A one-time price filters as much as it earns. One person answers every support request, so the number of users this can survive is a real constraint.
The trade, stated once: a one-time licence caps what any customer is worth and makes a future major version an awkward conversation rather than an automatic one. That is real, and we would rather name it than have you discover it later.
What the price does not include
The gap between expectation and reality is where refunds come from, so here is the gap.
An agent CLI
You bring your own — Claude Code, GitHub Copilot CLI or agy — on your own subscription, and its tokens are billed to you, including the block of instructions AgentCron appends to every prompt, which is roughly a couple of thousand tokens a run. AgentCron contains no model and no inference of its own, which is also why it can be a one-time price.
A support contract
There is no response-time commitment and one person answers everything. The channel it will happen in — an open issue tracker, so that one answer serves the next person — arrives with the first release. Until then there is no support address at all, and that is worth weighing rather than glossing.
Anything shared
One machine, one local data directory, no server, no sync. Two people on two Macs share nothing, and no future version of this architecture changes that.
Windows or Linux
macOS on Apple Silicon. Not now, and not planned.
Anything that has to happen at a specific minute
The scheduler runs inside the app on your Mac and fires nothing while it sleeps. This is not the tool for work that cannot be late.
Before you pay — every question with money in it
When can I actually buy it?
Not yet. Checkout is not wired up, so the Pro plan above says "Not on sale yet" and means it. Everything else on this page — the price, the refund window, the fairness clause — is written so that the day a Buy button appears, nothing about the deal changes except the button.
If you want to hear about that day rather than check back for it: the changelog announces the build and the buy button first — and it has a feed, so you can subscribe without handing anyone an address. There is no waiting list, on purpose.
Is there a refund?
Thirty days, all of it back, for any reason or none. No form, no argument, no proof of anything. The refund policy is published in full, including the awkward cases — day 31, EU withdrawal rights, VAT, chargebacks — because a policy written after the first complaint is worth nothing.
How do I pay?
By card, through a merchant of record who is the legal seller, handles VAT and sales tax for your country, and issues the invoice. Your card details never reach us — a one-person company should not be holding them, and this way it structurally cannot.
Why would I hand $89 to a developer I have never heard of?
You should not have to take anything on trust, so the deal is arranged to need as little of it as possible:
- Try the real thing first. The free tier is not a trial — one project, every safety mechanism, no account, no card, no clock. Run it for a month before you spend anything.
- The money is refundable for thirty days, no questions, processed by the merchant of record rather than by us.
- The one claim that matters is checkable. The privacy page lists every network destination in the app and shows you how to verify the no-telemetry claim yourself, in about two minutes.
- The licence outlives us. Validation is offline, against a signature — no activation server, nothing that stops working if the project does.
- The person is named. Who you are paying says who is behind this, how support actually works with a company of one, and what happens to your licence if that one person stops.
What this page cannot offer yet is a wall of testimonials. It is a new product; pretending otherwise would be the first lie.
Will overnight agents eat my CLI plan's limits?
They draw from the same allowance as your daytime sessions — there is no separate pool, and AgentCron adds roughly a couple of thousand instruction tokens per run on top of what your prompt costs. Anyone telling you to run agents around the clock is selling you a rate-limit problem.
So the honest advice is the boring one: start with one overnight agent, not five — and give it a weekly ceiling. Every agent takes an optional weekly spending cap, and at the cap its schedule pauses itself before your plan does, resuming as old spend rolls out of the trailing week. The cap counts the runs the cost view could price, which is its honest limit. The cost view shows per agent per day what each one actually cost — with a coverage line admitting which runs it could not price — and your CLI's own usage dashboard stays the number that counts. Scale up only when the morning diffs are worth more than the tokens.
How many machines?
Your own machines. A licence is for a person, not a seat — if you have a laptop and a desktop, that is one purchase.
What happens to my licence if the project stops?
It keeps working. Validation is designed to happen offline against a signature, with no activation server to go dark and nothing that starts failing when a domain lapses. That is an architectural choice, not a promise about the future.
Do I get the next major version?
Every update within this major version is included. A future major version would be a separate purchase, and it is better to say that now than to surprise you with it later.
Is the price final?
No. If it moves down, anyone who paid more gets the difference back without asking. If it moves up, everyone who already bought keeps everything they bought.
Is the free tier really free forever, or a trial with better manners?
Free, forever, and deliberately not crippled: the review gate, the approval gate, the command log, blocking questions — the entire safety half — are all in it, on one project. The line you eventually cross into Pro is scale, not safety: the second project, isolated per-worktree environments, the MCP bridge, the phone bridge and the ready-made packs. If one project is your whole world, the free tier is the product and you should not pay.
There is no checkout yet. Nothing on this page can take your money today, and this note is deleted the day that changes. The price on this page is the real intended one, published now so you can hold us to it: $89, once, thirty days money back from the moment paying becomes possible.
The free tier is the product on one project — no account, no card, no trial clock — and the download page says exactly what still stands between you and it.