Billing a client's AI to them

Some clients want their AI usage on their own bill — either because they already have an Anthropic account, or because they'd rather pay the model provider directly than have it marked up.

You can give any single workspace its own Claude API key. Everything in that client's workspace then runs on their key and appears on their Anthropic invoice. Every other client carries on using yours.

How keys are chosen

When one of your products runs, Productised picks a key in this order:

Setting a client's key

Switch into that client's workspace, then go to Settings → Claude API Key. You'll see two cards: your account key, and one for the workspace you're in.

Paste the client's key into the workspace card and save. To undo it, choose Remove, use account key — the client goes back to running on yours.

Check you're in the right workspace before saving. A key saved here applies to that client only, and it's easy to set it against the wrong one if you've just switched.

Ask the client to create the key for one workspace in their Claude Console. When they create it (Settings → API keys → Create key), they should pick a Workspace there — their Default workspace is fine. A key created for their whole Anthropic organisation won't work and fails with an error saying it "is not scoped to a workspace". This is a workspace inside their Claude Console, which has nothing to do with the Productised workspace you're saving it in. See Create the key for one workspace.

It's also a privacy answer

Billing is the obvious reason to use a client's own key. The other one comes up when a client asks who else touches their data.

With their own key in place, the AI that runs their products runs on their Anthropic account, under their agreement with Anthropic. Anthropic becomes their provider rather than a link in your chain. For a client who is careful about which companies process their customers' data — common in Germany and in regulated work — that is a shorter chain and an easier conversation than any assurance you could give on our behalf.

Worth being straight about what it does not cover:

• Where leads are stored. The answers people give, and the AI output generated for them, are still stored by Productised. The key changes who runs the AI, not who holds the results.
• Voice input and the Brain. A few paths still run on OpenAI — speech-to-text when someone uses the hold-to-speak button, and parts of the Brain. A client's Claude key doesn't change those.
• Anthropic's training. Neither their key nor ours puts data into model training; Anthropic excludes API traffic from it either way. The key is about whose account and whose agreement, not whether training happens.

If a client asks for this in writing, it belongs in your contract with them, not just in this setting. We can supply our own data processing agreement to sit behind it — email hello@productised.ai.

How keys are stored

Keys are encrypted before they're stored and are never shown again in full. Once saved, the page shows only a masked version like sk-ant-…4f2a, so you can confirm which key is set without the key itself being sent back to your browser.

If you need the original value later, get it from the Anthropic Console — Productised can't show it to you.

A key that doesn't begin with sk-ant- is rejected when you save it. That's deliberate: a mistyped key would otherwise be stored happily and every run would quietly fall back to your credits while you believed the client was being billed.

A key is used from the first run, not the last

A key isn't held back until credits run out. The order above is checked on every run, so the moment a key exists it wins — your Productised credits stop being spent on that workspace entirely, rather than being spent first and handed over when they're gone.

Those credits aren't wasted. They sit unused as a safety net: an agency that removes or rotates a key isn't instantly dead in the water while it sorts the key out.

Because the balance stops moving, it can't tell you how much a client is actually using. Settings → Account & billing → Usage & cost breaks the real cost down per workspace, and the Claude connector will report runs and cost on your own Anthropic account if you ask it about usage.

What it doesn't change

Giving a client their own key changes who pays Anthropic for running the products. It doesn't change your Productised subscription, your plan's product limits, or your Brain allowance — those stay with your account.