Pricing

Test mode Purchasing is switched off while we hold the quality bar. Runs on your own model key work normally.

Hosted — we run the models

$ — (price not yet published) run fee, per run

plus metered inference at the rate card below, markup included. A run never starts a stage it can't afford.

  • Cost estimate before you start, live spend while it runs
  • Set a hard cap per run
  • Approve the spend and the release — two gates, in the browser

Bring your own key

$ — (price not yet published) flat, per run

Anthropic, OpenAI or Azure AI Foundry. Your provider bills your inference — we never meter or mark it up.

  • Name any deployment on your own resource
  • Keys are write-only and never logged
  • Connection check before anything is spent

A real run, priced

Illustrative rates

A small offline habit tracker: twelve work items, two retries, no image stages. Every amount below is a placeholder until the rate card goes live.

One run, stage by stage
Line Tier Amount
Run Credit — $ — (price not yet published)
intake + research Fast $ — (price not yet published)
storyboard + decompose Balanced $ — (price not yet published)
implement + audit Deep $ — (price not yet published)
Total as configured $ — (price not yet published)
Same idea, all economy $ — (price not yet published)

Shorter research, thinner work items, more retries.

As priced above $ — (price not yet published)

Deep only where the code is written.

All frontier $ — (price not yet published)

Every stage at the top tier, including the ones that don't need it.

The spread is the point. Tier names are stable; which deployment sits behind a tier is ours to keep current, so a saved run profile never breaks.

Rate card

Hosted inference is billed per million tokens, markup included. You pick a capability tier per stage, not a vendor: which deployment sits behind a tier is ours to manage, so your saved configuration never breaks when a vendor changes something.

Per million tokens, markup included
Tier Good for Input Output
Fast summarization-class stages: intake, research, store copy, audits $ — (price not yet published) $ — (price not yet published)
Balanced general reasoning: storyboarding, decomposition $ — (price not yet published) $ — (price not yet published)
Deep code and structural planning: implementation, publishing $ — (price not yet published) $ — (price not yet published)
Image design and marketing imagery — billed per image, not per token $ — (price not yet published)

Prices are not yet published — every amount above reads "not yet published" until the rate card goes live. The tier names are stable; saved run profiles reference tiers, never the deployments behind them.