Skip to main content
Credits measure AI work in Mythos. Each action has a fixed price chosen by the server; writing a longer prompt does not change that price. Mythos is currently in private, owner-only access. Public registration and paid credit top-ups are not available during this phase. This page describes how the balance on an authorized account is used.

​
What actions cost

If a Build or edit also produces a downloadable file, there is no separate artifact fee. It remains part of that same action.

​
When credits are kept

Mythos reserves the required balance when an action starts so the same credits cannot fund two runs at once.
  • A Plan turn is kept only after Mythos saves a durable response, set of questions, or plan draft.
  • A Build or edit is kept only after a result is delivered.
  • A failed or interrupted action releases its reservation automatically.
You do not need to request a routine failure refund. The balance may take a moment to reflect the terminal result, but the failed action is not treated as a delivery.

​
Actions that do not use credits

These actions do not start an AI run:
  • Editing and saving the Plan draft yourself.
  • Approving or cancelling a Plan session. The Build started after approval has its own normal price.
  • Editing a project file directly in the workspace editor.
  • Viewing an earlier version or reverting to it.
  • Publishing or republishing an already delivered project.
External services you connect may have their own billing. Mythos credits do not pay a Supabase, domain registrar, or other provider bill.

​
If the balance is too low

The action does not start when the available balance cannot cover its fixed price. Because public top-ups are not active during owner-only access, there is no checkout path to follow in these docs. If the displayed balance or a refund looks wrong, contact Mythos support with the project and action that produced the mismatch.

​
Keep usage predictable

  • Use Plan mode when a first request still has important open decisions.
  • Ask for one coherent outcome per edit, then inspect the result before sending another.
  • Use the workspace editor for a small change you already know how to make directly.
  • Revert a poor direction before asking for more work on top of it.
Continue with Building and editing to see how an AI action moves from a prompt to a delivered result.