Access & Credits

Prompt refinement, extracting a prompt from an image, Atelier reasoning sessions and image generation consume credits. Copying the finished prompt to the clipboard also costs one credit, because the copy logs it to your history and project spine. Choosing engines, presets, Logic Gates, styles, materials and every control on the parameter deck costs nothing.

New accounts start with 50 credits, granted once at signup. There is no daily refill — when the balance runs out you either top up or move to a paid plan.

Do all your structural thinking with the free controls, and spend credits only when you need the model to reason or copy the final prompt. Most people burn credits on refinements they could have made by hand.

MCP access lets an outside AI agent or tool talk to your Witness Labs account — checking who you are, listing and reading your prompt history, deleting entries, and reviewing recent usage. Everything runs as you, under the same row-level security as the Vault.

Programmatic access to your own data. Not a public API for other people's prompts.

Use it to pull finished prompts into whatever you already work in, rather than copying by hand. It reads your Vault, so anything you've saved is reachable.

Lower tiers reset credits each cycle. Architect and Studio roll unused credits forward, so a light month isn't wasted and a heavy month can draw on what you banked.

Rollover is capped by tier — Architect one month, Studio extended. Explorer and Operator do not roll over.

If your usage is spiky, the rollover on Architect/Studio is often worth more than the raw monthly number.

Every credit change is written as a signed entry on a ledger: grants and top-ups are positive, spends are negative, and each entry records the exact balance_after the change. Nothing is a silent backfill. A separate reconciliations audit table records any corrective adjustment, with the reason attached.

The Usage tab shows the ledger newest first, with running balance visible on every row. If something looks wrong, the row that broke it is on that page.

Check the ledger, not the headline balance, when a number surprises you. The ledger is the source of truth; the badge in the header is only a summary of it.

Admins can flip a per-account test_role that renders the workstation, gates and pricing surfaces as if they were on a chosen tier. Real billing, real credits and real Vault content are untouched — only the visible tier and the gates that read from it change.

The simulation is visible: the header shows the simulated tier badge, and gated features either open or lock accordingly. Turning simulation off restores your real tier immediately.

Use it to verify a feature's gate behaves correctly across tiers before shipping a change. Never test payments through it — payments run against your real account regardless.

A dedicated branded signup page where you can register interest in features that aren't public yet. Requests land in an access queue with a proposed plan grant and credit grant attached, which an admin reviews and applies to your account when the cohort opens.

Submitting an email doesn't change your current plan or balance. If your request is approved, the grant lands on your account and you'll see the tier or credit change announced the next time you sign in.

Use the note field to say what you actually want to build. Requests that describe a real job get reviewed first, because they tell us which feature to open next.

The connect page publishes an MCP server URL and step-by-step instructions for pointing an external AI agent — Claude, an IDE assistant, or any MCP-capable client — at your account. The agent authenticates as you and can read your prompt history and usage rather than guessing at context.

Agent access is read-scoped and tier-gated to Studio. It doesn't bypass credits: anything the agent triggers that costs credits is charged to your ledger like a click in the app.

Connect one agent at a time and check the Usage ledger after the first session, so you can see exactly what the agent touched before you rely on it.