Questions teams actually ask.

Kanten is a workspace where a QA team keeps a function list per feature, one shared viewpoint catalog, and a case repo. AI may draft updates to the lists and write cases into the repo. It is not a chatbot that dumps steps — the design lives in the lists, not in a conversation.

A private prompt does not unify design. Different modules get different ideas of “enough,” and the reasoning disappears with the chat. Lists are how the team shares classes, boundaries, and what it chose not to test.

The artifacts are the same idea — 機能一覧 and 観点一覧. The process is not waterfall. Lists live across sprints and you update only what moved. AI writes inside those lists instead of replacing them.

No. This version does not run the browser and does not generate scripts. It is analysis, a repo of functions, viewpoints and cases, and Excel export.

The inventory of behaviors for one feature — Apply coupon, Pay, re-validate cart. One row is one behavior, not one validation rule. IDs stay stable (FUNC-xxx), so a case written in March still points at the same row in September.

One team list of questions coverage must answer: empty, expired, unauthorized, decline must not consume quota. Features map each viewpoint Must / Should / Drop. Features use the catalog; they do not fork it.

Any owner or editor in the workspace. After a production miss, add the viewpoint that would have caught it, with a note on where it applies. Every later feature in that workspace sees it, so the lesson becomes the next person’s starting point instead of a chat remnant.

No formal PRD is needed, but you do need something written: a ticket, an acceptance criterion, a paragraph of notes. Typing functions by hand is free and always available. AI drafting needs material to read — an empty box is refused rather than guessed at, and nothing is charged for a refused run.

In the case repo, each case tracing to a function ID and a viewpoint ID. They persist across sprints and are not stuck in a chat log or a disposable project.

A case is stale when its function or mapped viewpoint changed after the case was written. The case stays visible until you regenerate or retire it, so a changed requirement shows up as work to review rather than a rewrite you did not ask for.

Only model work: drafting a function list from the material you provide, the AI coverage review, marking the function × viewpoint matrix, and writing cases. Kanten shows a planned estimate before a run and charges the usage the run actually reports — the estimate is a plan, not a guaranteed maximum. Reviewing, editing by hand, applying a draft and exporting cost nothing.

No. Drafts are drafts — you accept what “pass” means. Rows marked assumed or ask-product stay pending until a person confirms them. The oracle is human.

No. That is the one thing Kanten refuses. Cases are a product of the two lists, so coverage always traces back to a decision you can review.

Editing requirements, functions and viewpoints, deciding Must/Should/Drop by hand, applying or discarding a draft, confirming and retiring rows, inviting people, and exporting the Excel workbook. You pay only for AI runs, never for your own work in the lists.

Free $0: a one-time trial of 100 AI credits, issued after you confirm your email address and not renewed each month, one account. Solo $29/month: 900 AI credits, one account. Team $79/month: 4,000 AI credits shared by everyone in the workspace, at one flat price with no seat count — this is the plan that lets you invite colleagues into a shared workspace. Every plan supports feature-scoped requirements, reviewed functions, a maintained shared viewpoint library, a coverage matrix and traceable case drafts. On a paid plan, credits follow your subscription's billing period from Stripe rather than the calendar month; the Free trial allowance is issued once and does not renew, so reviewing, editing by hand and Excel export stay free after it is used up.

Hand-editing a requirement, function or case. Adding a viewpoint. Reviewing or applying a draft an AI run already produced. Reading the repo. Exporting the workbook. Metered means model runs, not your own keystrokes — but note that the run which produced a draft was charged when it ran, even if you later discard the draft.

No. The monthly allowance resets with your billing cycle. Purchased top-up packs, when available, sit on the workspace until they are used.

Export does not spend AI credits on any plan, including Free. Free may label the file “Kanten Free”; the sheets themselves are complete.

No. Each step explains what it is for and what a good row looks like, and the shared library already holds the perspectives experienced testers reach for — boundaries, permissions, interrupted flows — so you start from questions rather than a blank page. If you are experienced, you can override every AI decision, write your own viewpoints, and keep several concrete variations per function and viewpoint.

Yes. AI proposes; it does not certify, and it can be wrong, incomplete, or label a row wrongly. Kanten is built around that: prompts ask the model to mark source statements, assumptions and open questions, the function list and the viewpoint review cannot be skipped, and what the workbook records is your decisions and the questions still unresolved. Treat every draft as something to review, not as evidence of coverage.

That is the point of the shared library. One library serves every project in the workspace, each viewpoint carrying where it applies, so the next feature inherits the question instead of rediscovering it. You can edit or add viewpoints at any time, free of charge.

Yes, on Team. Free and Solo are single-account plans: the person who pays is the only person in the workspace. On Team you invite people as editors or viewers — editors author, review and can run AI; viewers read and export only; the paying account stays the single owner and manages people, invitations and billing. Invitations are single-use and expire. All AI runs come out of the workspace's one 4,000-credit monthly allowance, so there is no per-person charge and adding people never adds credits. If a workspace later leaves Team, shared access pauses and nothing is deleted: members, roles and work stay, and access returns when Team is active again. Libraries and projects are never shared between separate workspaces.

Kanten splits the work into sequential batches on its own and shows the progress, so no single call is truncated. Caps still apply per batch: 30 functions, 80 viewpoints, and 80 cases. The estimate you confirm covers every batch.

English and Japanese. You choose the interface language, and you choose separately which language AI writes in, so a Japanese team can keep an English interface or the reverse. Export headers follow the interface language; your own text, IDs and codes are never translated for you.

When you start an AI run, the material that run needs is sent through our AI gateway to a third-party model provider, which processes it to return the draft. That provider's own processing and retention policies apply and we do not restate them for you. Only paste material you are permitted to share. See the Privacy page and the AI notice.

You can delete projects, features and rows from inside the app; an empty project can be removed outright. Deleting a whole account is not yet something you can request from these pages, and no removal timescale is promised here. Billing records may be retained where law requires.

Not in this version. Export the workbook and import it where you need it. Integrations are deliberately out of scope until the lists themselves prove their value.

Several projects live in one workspace and share its viewpoint library, which suits several products in one team. Separate workspaces per company are not offered yet, so a library cannot be split between unrelated products.