4 min read

How the AI studio works

Real text documents, bounded spending, and an honest record of what ran.

Build on the last good decision

The text worker builds from your saved brief and recent documents. It moves through positioning, visual identity guidance, product specifications, website copy, customer validation plans, and useful revisions. Each result is a document you can review; a proposed logo, program, or website is not an image, executed program, or deployed site.

The ongoing work loop

  1. Read the brief and history

    Use the creator’s project description and excerpts from recent saved documents to choose the next writing task.

  2. Check the model and budget

    Recheck the exact selected model, its current prices, and supported limits. Reserve a conservative maximum from the brand’s available balance before sending a paid request.

  3. Create one bounded result

    Send one text request with a fixed output limit. The worker does not browse, run code, generate images, contact people, place orders, or access treasury signing keys.

  4. Save and account for it

    Record the generation ID, usable document, model, and provider-reported cost. Unknown charges retain their reservation while the receipt is checked.

  5. Continue when ready

    Use the saved result in later work. Scheduled checks happen every minute, with at most one new brand run per check and a five-minute delay after a settled attempt. Funding, demand, and provider availability determine when a brand actually runs.

What counts as useful work

WorkReviewable output
Identity developmentA revised voice guide, colour direction, or identity proposal with a clear rationale.
Product planningA focused offer, product specification, or artwork brief for review.
Content productionDraft product descriptions, campaign concepts, or a content calendar.
IterationA useful revision based on the saved brief and earlier documents, with assumptions stated.

A text model can help with reasoning and copy. Images, production mockups, and manufacturing files need tools suited to those outputs. Each tool must be evaluated and funded separately.

Follow actual work in the activity feed

The confirmed brand workspace reads actual job records: status, time, model, saved documents, and recorded cost. It also shows verified earned and claimed creator fees with relevant transaction links. Pending charges stay pending; an interrupted request is not silently sent again. An empty history means no completed work has been recorded.

Saved text documents are public on the brand page. Review them before relying on claims, using proposed code, or publishing material elsewhere. The feed displays results and task summaries, never hidden reasoning or fabricated progress. Example-brand timelines remain concept previews.

Give the studio boundaries

The current worker is limited to text documents and its allocated operating balance. New work waits when funds or model support are insufficient. A reservation covers the model’s full listed context allowance plus a maximum of 2,048 output tokens; it is deliberately larger than typical usage. The actual provider charge settles the run and releases the unused portion.

People remain responsible for sample quality, product claims, rights to artwork, pricing, and customer commitments. The worker saves public text documents for review. It does not generate images, execute code, browse the web, publish to external sites, or fulfil orders. Its suggestions cannot change treasury policy or authorise financial transactions.