Keys and spend
Where model keys live, how account and project keys resolve, the projected cost by engine, and how to control what a month of tracking costs.
Updated October 20, 2026
Two bills, and keeping them apart explains almost every question about cost. We invoice you a flat fee for the software. Your model provider invoices you for the prompt runs, at list price, on your own key.
That is why tracked prompts are unlimited on every plan: we are not reselling the model calls, so there is nothing for us to meter.
Account keys and project keys
An account key applies to every project you own. A project key overrides it for one project.
The resolution order is exactly this, with no other steps:
- The project's own key, if it has one.
- Otherwise the account key.
- Otherwise nothing runs, and the project says so.
Both live as an OpenAI field and an OpenRouter field. Keys are encrypted at rest, used only to run your own scheduled prompts, never logged, and shown as their last four characters once saved.
The agency pattern
One key per client project, created in the client's own provider account. The client's tracking then lands on the client's provider bill, and your cost stays the flat fee however many prompts you run for them.
Per-project keys are an Agency feature, and they are most of the reason that plan exists.
Projected cost by engine
Settings shows the projected monthly cost per engine for the current prompt set and cadence. Two things it answers that an invoice cannot:
- Which engine is expensive. It varies more than people expect and changes when providers change prices.
- What a cadence change will do, before you make it rather than after.
The track-a-prompt dialog shows the same estimate for a single prompt across all platforms as you set its cadence and sample count.
What actually drives the bill
Three numbers multiply out to your whole model spend:
Prompts x assistants x cadence x samples.
Twenty prompts, four assistants, weekly, one sample, is a couple of dollars a month. The same twenty on daily is seven times that. Three samples per run triples it again. Nothing else you can change comes close to the effect of those.
Reading the spend breakdown
Cost is recorded per probe: input tokens, output tokens and web searches. Rolled up, that gives you cost per prompt, per assistant and per run, which is where to look before deciding a prompt set is too expensive. Usually two or three daily prompts are most of the bill and could be weekly.
We never see your provider balance. A key with no credit fails every run, and the only places that shows are the run failure and your provider's own dashboard. Keep a little headroom on it, and check the provider side first when runs stop.
Questions about keys and spend
- Could a run ever be charged to Omnipresence instead of me?
- No. With no key the project runs nothing. There is no hidden fallback, which is deliberate: a run that silently cost somebody else money would be worse than one that did not happen.
- Which key covers which assistant?
- OpenAI covers ChatGPT. OpenRouter covers Claude, Gemini and Perplexity through one key.
- Can I see what a prompt has cost?
- Yes. Cost is recorded per probe, including tokens and web searches, so the spend breakdown goes down to the individual run.