Open frontier models. Run on your terms.
Qwen, DeepSeek and Kimi through one API today, or the same open-weight models in your own jurisdiction.
Gateway requests are processed by our upstream provider in mainland China. Details on Trust.
Two ways up the same mountain.
Start with one API key. When your data needs to stay at home, run the same open-weight models on infrastructure you control.
One key to Qwen, DeepSeek and Kimi.
One OpenAI-compatible endpoint. Switch models by changing one string. Pay by card, no contracts.
Explore the Gateway 02 · SOVEREIGNYour AI platform. Your jurisdiction. Your keys.
Open-weight models on your servers or in a data centre in your region. Meruva designs, installs and supports it.
Explore SovereignFrom sign-up to first call in minutes.
One OpenAI-compatible endpoint for Qwen, DeepSeek, Kimi and more. No US models, no contracts, and nothing stopping you leaving: change your base URL and go.
- Pay. Top up by card through Stripe. No contracts.
- Get your key. Create a key in the dashboard.
- Connect your tools. Point your app or coding agent at Meruva.
- See every cost. Usage per key and per model, in the dashboard.
Control, layer by layer.
Sovereignty depends on who controls the hardware, the keys, the logs and the updates, not on who wrote the model.
- Discovery. Your data rules, workloads and budget.
- Design. Hardware or in-region hosting, models and governance.
- Deploy. We install and test the stack in your environment.
- Run. Support, updates and new models, under an agreement you set.
Open models, tested before they're trusted. Because the weights are open, you can inspect and test every model. Independent evaluations, including the US NIST CAISI review of DeepSeek, have found weaknesses such as jailbreak susceptibility. So before go-live we evaluate against your own standards, add guardrails, and route each task to the model that passes.
Who controls what
Sovereign| Hardware and location | You. Your servers, or a data centre in your country or chosen region. |
|---|---|
| Model weights | You hold a copy. Open-weight Qwen, DeepSeek, Kimi and more, from official sources, checked against published hashes, with licences listed per model. |
| Prompts and outputs | Stay inside your deployment. The platform doesn't call the model labs or the Meruva Gateway. |
| Encryption keys and identity | Yours, integrated with your directory. |
| Logs and audit | In your environment. Every call traceable to a user, a project and a cost. |
| Updates and new models | You approve them. Connected, or offline with updates delivered by a route you choose. |
| Meruva's access | None by default. Support access is granted by you, time-limited and logged. |
Start at base camp.
Try the Gateway when sign-up opens, or talk to us about your own platform.