NIK Agent defaults to our own private model, running open-source weights in infrastructure we host domestically. A frontier model is one setting away if you want one, with your key and your rates, and it never becomes the default behind your back.
Open-source weights have closed most of the practical gap for the work an assistant actually does: reading, summarizing, drafting, extracting and reasoning over your own material. Running them ourselves is what makes sovereignty and the price both possible.
If you want Claude, GPT, Gemini or Grok in the loop, an administrator adds the organization's API key and chooses who may use it. The requests bill to your account with that vendor, and the rest of the workspace stays on the private model.
| Data location | Those requests go to the vendor's cloud |
| Training | Governed by that vendor's policy |
| Cost | Billed by the vendor at their rates |
| Everything else | Unchanged: identity, guardrails, traces, admin control |
A per-seat frontier plan bundles inference into a flat price that has to cover its heaviest users. Metering the tokens and paying $5 for the seat keeps the predictable part predictable and the variable part honest.
| Component | What you pay | Who sees it |
|---|---|---|
| Seat | $5 per user per month on Team, $0 on Free | Finance, once a month |
| Private model tokens | $0.25 per million tokens | Administrators, by person and by team |
| Frontier model tokens | The vendor's rates, on your own key | You, in the vendor's console |
| Free plan | 20 million tokens a week, refreshed weekly | Nobody, it is free |
For the wider Netarx platform, including the identity layer that NIK Agent builds on, see netarx.com pricing.
Most teams find the default is enough. Send a document the agent has never seen and judge the answer, not the benchmark.