What Tasklet AI costs, beside an open‑source bill

What Tasklet charges, line by line

The plans start at $25 a month for casual use and $100 a month for power use, with a $250 tier and any custom amount up to $10,000. Billing is at the organization level with unlimited team members and one shared credit pool, as its pricing page states, so adding a colleague never changes the bill. Starter at $25 a month carries 10,000 credits, Pro at $100 carries 40,000, and the $250 tier carries 100,000.

There is a second way to buy the same thing. One-time credits start at $25 at a fixed rate of 320 credits per dollar. They are valid for a year from purchase and are spent only after the plan’s own credits are gone. A monthly plan can also be set to any amount from $25 to $10,000 in $5 steps through Stripe, and each step adds 2,000 credits.

What a credit pays for is the part that matters a year later. Tasklet’s own list of what moves the number: task complexity, how much context and data are involved, which tools and connections are active, automation frequency, and the intelligence level you choose. Every item on that list is a property of the request. None of them is a headcount, a server or a contract. The platform that runs the request rides inside the rate, priced into the same credit as the model call. The figures here are as published in October 2026.

What the free tier actually buys

Tasklet’s free tier asks for no payment card. It carries limited trial credits and lets you create your first agent and connect your tools. Its page states that everyone can create an agent and start for free. Tasklet’s own advice on the same page is to upgrade or buy credits once usage grows.

The free tier answers one question well: do these agents do useful work for this company? It answers a second question badly. A trial allowance is a fixed pot, not a rate, so it cannot be multiplied out into what a working month will cost. The paid page is where that number lives, and it is quoted in credits rather than hours.

What Kortix charges for the same job

Kortix’s free plan is a published rate rather than a trial. Free is $0 with 500 credits per month, every feature and 1 project, on the user’s own provider keys or ChatGPT subscription. The paid plan is credits: 25 tiers from $50 a month (5,000 credits) to $50,000 a month (5,000,000 credits), at 100 credits per dollar. Members are unlimited and free, and the account shares one credit pool.

Self-host is free. The platform, the config in your repo and the history of every change run on a machine you already pay for. The install comes from Kortix on GitHub. Any model, your own keys, on Kortix Cloud, in your VPC or on your own on-prem network.

A Kortix credit buys something narrower, and that is why it can be counted. Kortix’s pricing page puts 1 credit at $0.01. The default 2 vCPU / 4 GB / 20 GB machine runs at about 11 credits an hour, billed per second, and costs nothing while it is stopped. Model calls are a separate line. Your own API key is free for the first $1,000 of model usage a month, then 5% is paid from credits. A connected ChatGPT subscription spends no credits. Kortix models cost the provider’s base price plus a 5.5% platform fee, paid from credits.

The difference, once both bills sit together

Two products, two meanings for the word credit. Tasklet’s credit buys a finished request on a platform Tasklet operates. Kortix’s credit buys the machine the work runs on, and the platform itself can be hosted, so its cost can leave the bill entirely.

Tasklet sells one-time credits at 320 per dollar. That price carries a bundle: the approval layer, the credential vault, the isolated environments and the vendor’s on-call rota. Kortix sells at 100 credits per dollar for compute, and quotes the model rate on a separate line. The control plane sits in a repo that self-hosts for nothing. Neither rate is better in the abstract. They price different things, and only one of them can be removed from the monthly total.

A small team doing a small amount of work is cheaper on Tasklet. One person on Starter gets 10,000 credits a month, plus 600 bonus credits on days the app is opened. That spends less than the smallest Kortix paid tier at $50 a month, and less than an afternoon spent standing up a server.

Kortix’s free plan is where that argument gets tested. It carries 500 credits a month. At the published machine rate that is a bit over 45 hours of the default computer, enough for real work before a card appears. Tasklet’s 10,000 credits do not convert into hours at all, because they buy requests, priced by the five factors its page lists. One side meters the work and the other meters the computer, so the comparison comes out in requests against hours, not credits against credits.

How to estimate your own month on either

Start with the shape of the work. If your agents take a few tasks a day and stop, you are buying requests. Tasklet’s model fits that shape: a fixed plan, one shared pool, and a credit rate set by the intelligence level and the context each request needs. If your agents hold a computer open for hours, editing files, running scripts, browsing, you are buying machine time. The Kortix rate converts straight into hours: about 11 credits each, a cent apiece, nothing charged while the machine is stopped.

Then count the people, and notice that the answer changes neither bill. Both products make members free, so a comparison built on headcount is measuring the wrong line.

Then choose the model, which is the largest variable on one side and a fixed menu on the other. Tasklet "gives you access to leading models from Anthropic, OpenAI, Google, and open-source labs in one plan", priced into the same credit. On Kortix the model is a decision: bring a key, connect a ChatGPT subscription, or use managed models at the provider’s price plus the platform fee. A team that already pays for model access twice is paying for it twice.

Finally, price the platform as work rather than as free. Self-hosting removes the control plane from the bill. It adds a machine, a model key, and somebody who answers when the server is unhappy. That cost is real and it appears on no pricing page. Build the estimate from published rates and your own count of hours. Treat each vendor’s monthly headline as the starting point.