How it works
Nothing here is a black box.
Four steps from notes to a finished piece, two gates before any model runs, and one honest number afterwards. If you want to know why a request cost what it cost, this page answers it.
The work
Brief, generate, work, ship.
You can stop after any of them. Nothing is a wizard you have to finish, and nothing expires because you walked away mid-draft.
- 01
Brief it
Paste your notes, a transcript, or a one-line angle.
Pick the layout the piece will run in and the mood it should carry before anything is generated. The model writes to the shape of the page rather than being poured into it afterwards.
- 02
Generate it
Gemini drafts the whole piece in the layout you chose.
You see the token cost of the request as it happens, split into what you sent, what the model thought, and what came back. Regenerate any single section without disturbing the rest.
- 03
Work it
Enhance the prose, pull headline variants, translate, spin social posts.
Every action is metered the same way and recorded in the organisation usage log, so at the end of the month the bill reconstructs from a list of things you actually did.
- 04
Ship it
Share it for review, take comments, keep the version history.
Share links go to people outside the organisation without giving them a seat — and never get indexed, in any environment. Every save is a snapshot you can read back.
Metering
A token is not a word, and most of them are thinking.
A reasoning model works the problem before it answers. Those thinking tokens never appear in the article and are billed at output rates. Every AI writing tool built this way charges you for them; this one shows you the split.
We meter the total, thinking included, read from the model’s own usage figures. Estimating it afterwards would be cheaper for us to build and impossible for you to check.
Routing
The saving is in choosing the right model, not the shorter prompt.
Asking a reasoning model to think less does not work — we tested it, and a zero thinking budget did not reduce the thinking. What does work is not sending it the jobs it is not needed for.
Everyday model
Headlines · translation · social posts · everything on the free plan
Answers directly. No thinking tokens, so a short request costs about what it looks like it should.
Frontier model
Long-form generation and investigative work on plans that carry priority routing
Deliberates before it writes. Better at holding an argument across 2,000 words, and priced accordingly.
A headline and a 2,000-word investigation are not the same request, so they do not go to the same model. Cheap actions run on the everyday model on every plan — including Pro, where you would otherwise be paying frontier prices to rename a story.
Before any model runs
Two gates, in this order, every time.
Both are checked before a request leaves the building, so a refusal never costs you tokens — and a refusal always names what would fix it.
If a plan lapses
Degrade, never destroy.
A lapsed plan is not a deletion event. Here is precisely what changes, so nobody has to find out by losing something.
- Articles past the free limit
- Stay readable and editable. You just cannot create new ones until you renew.
- Pro-only layouts
- Keep rendering exactly as written. They cannot be regenerated.
- Members past the free cap
- Keep their access. New invitations are blocked.
- Version history older than seven days
- Retained a further 30 days before pruning, so a renewal inside that window loses nothing.
- Unused top-up tokens
- Frozen, not forfeited. They come back on renewal, provided the pack has not reached its own expiry.
The only way to know is to run one.
Bring notes you already have and generate a piece. The cost lands in your usage log the moment it finishes, itemised.
No card. Prices in INR, GST included.