Artux defaults to a ready-to-use managed plan. Standard is ¥1,699/month including a dedicated instance, managed AI, 10 seats, and 10M tokens. Pro is ¥4,999/month including 30 seats and 40M tokens. Teams that already have a Claude plan or API account can instead choose the ¥999/month Standard or ¥2,499/month Pro platform-only option; model charges are paid separately to the provider. Annual billing includes two months free. Registration is free, and the seven-day trial is manually approved. The live pricing page is the source of truth.
The complete price structure
One subscription covers one digital employee: its dedicated runtime, workspace, operational control plane, included AI usage, and the number of people allowed to interact with it.
| Plan | Managed AI included | Platform only, with your own AI | Instance | User seats | Managed quota |
|---|---|---|---|---|---|
| Trial | ¥0 for 7 days | — | t3.medium | 5 | 3M |
| Standard | ¥1,699/mo · ¥16,990/yr | ¥999/mo · ¥9,990/yr, model charges separate | t3.medium | 10 | 10M/month |
| Pro | ¥4,999/mo · ¥49,990/yr | ¥2,499/mo · ¥24,990/yr, model charges separate | t3.medium | 30 | 40M/month |
Enterprise arrangements can change instance size, seats, quotas, and commercial terms. They are not represented by a fabricated public price.
Seats and managed-AI usage are included in the published plans. As a team approaches a plan limit, the console warns the administrator to confirm a plan upgrade or Enterprise sizing. Artux does not automatically append per-seat or token overage charges.
What does the monthly price pay for?
The price combines two cost categories.
First is the dedicated environment: a continuously available employee instance, persistent workspace, isolation, monitoring, configuration delivery, upgrades, backups, and recovery work. A cheaper shared database can reduce the vendor’s infrastructure bill, but it changes the isolation model. Artux does not describe those two architectures as equivalent.
Production tenant instances currently run in a private Seoul subnet with no public IP and use NAT for outbound traffic. The observed monthly cost, including a 30 GB gp3 volume and allocated NAT cost, is about US$55 (roughly ¥400) for t3.medium. Trial, Standard, and Pro all use t3.medium today; their commercial differences are seats and managed-token allowance. t3.large is reserved for manually sized Enterprise deployments and currently costs about US$93 (roughly ¥670) per tenant.
The current fleet is bursty rather than capacity-bound: average CPU is 1–2%, seven-day peaks are around 58–70%, CPU credits remain full, and the tightest t3.medium still has roughly 2.7 GB available memory without swap use. There is no evidence-based reason to increase the public-plan instance size yet. The sample is only one to three days old, so this conclusion will be revisited after sustained paid usage.
Second is AI usage. Claude Sonnet list pricing is US$3 per million input tokens and US$15 per million output tokens. At an 80/20 input-output planning mix, 10M raw tokens cost roughly US$54 upstream and 40M roughly US$216. Output-heavy reports cost more. The control plane makes usage visible so a company can distinguish productive work from a retry loop or poorly scoped task.
Will 10 million tokens be enough?
It depends more on task shape than message count. A short question can be expensive if it loads a large document set; a scheduled workflow can be economical if it reads only the changed rows.
Start with the included quota and observe real work for a week. Use bounded context, reusable artifacts, and explicit stop conditions. If the employee repeatedly rereads the same source material, fix the workflow before moving to a larger plan.
What is a seat?
A seat is a human user authorized to interact with that digital employee. It is not another employee instance. Ten people can work with one Standard employee; adding a second digital employee creates another runtime with its own role, workspace, permissions, and quota.
This makes seat comparisons with conventional assistants misleading. A product sold “per human user” and a product sold “per autonomous worker” count different things.
Why compare it with labor cost?
The intended comparison is the recurring work being delegated, not another chat subscription. A specialist may cost at least ¥6,000 per month before management time and overhead. Managed Standard remains below one third of that baseline, but the comparison is valid only when it produces usable work.
If every output must be rebuilt by a human, the lower subscription price does not make it economical.
When is it not worth it?
It is a poor fit for a task performed once a quarter, a workflow that cannot be given system access, or a team that only wants occasional drafting help. A copilot or general assistant is usually simpler in those cases.
It is also premature when nobody owns the process definition. AI cannot repair an organization that has no agreement on what “done” means.
What hidden costs should you expect?
Plan for app-permission approval, initial workflow definition, review during the pilot, and maintenance when a connected system changes. Artux absorbs instance operations and platform upgrades, but it cannot decide your approval policy or define the correct business result for you. Upstream Apify, SellerSprite, and similar data-service fees are not hidden inside the connector: bring your own account by default, or have the vendor plan itemized separately.
The lowest-cost way to validate
Choose one real recurring job. Define the expected artifact, forbidden actions, data sources, and reviewer. Use the seven-day trial, then measure time saved, correction rate, and token usage. Do not evaluate with a polished demo that nobody performs in normal work.
FAQ
Is the listed price per person?
No. It is per digital employee. Each plan includes a number of human user seats.
Is the cloud instance charged separately?
No for the published Standard and Pro plans; dedicated-instance cost is part of the subscription.
What happens above the included token quota?
The console warns the administrator as usage approaches the limit. At the limit, further managed-AI usage pauses until the administrator confirms a plan upgrade or custom capacity; Artux does not charge a silent overage.
Can we test without entering a credit card?
Registration is free and trial provisioning is manually approved, preventing anonymous users from starting paid infrastructure automatically.