Open frontier models. Run on your terms.
Qwen, DeepSeek, Kimi, GLM and more through one API, or the same open-weight models in your own jurisdiction. Gateway sign-up is coming soon.
- 0:00 Sign up
- 1:00 Top up by card
- 2:00 Create one key
- 3:00 Change base URL
- 5:00 First call
Example only — Gateway sign-up is coming soon; you cannot buy or get a key yet. Example blended rates from a real coding-agent workload (18 Sep–26 Oct 2026, mostly cache reads) versus the same tokens at Claude Sonnet 5 list prices. Not Meruva retail prices.
One key to Qwen, DeepSeek, Kimi and GLM.
One OpenAI-compatible endpoint. Switch models by changing one string. Pay by card, no contracts. Change your base URL and leave whenever you like.
See the GatewayYour AI platform. Your jurisdiction. Your keys.
The same open-weight models, on your servers or in a data centre in your country. We start in the UK, then grow one market at a time.
See SovereignFastest sale first.
Developers start on the Gateway today. Sovereign then follows the same ladder, in any country.
1 Developers
Startups and teams who want one key to open-weight frontier models, paid in pounds.
2 Enterprise
Banks, insurers, law firms and pharma that cannot send data abroad. Shortest Sovereign cycles.
3 National verticals
Health, universities and research, defence, energy and utilities, and financial regulators.
4 Government
Ministries and national agencies, with a local partner and Meruva as prime contractor. The longer play.
Open models are good now. Using them is still a chore.
Your agent bill keeps climbing
Long coding-agent sessions burn premium tokens on routine work.
One account per lab
Separate sign-ups, consoles, invoices and API formats for every model family.
Setup eats an afternoon
Base URLs, request formats, cache settings and agent config, all different.
Your data can't leave the country
Regulated and government work needs AI that stays in your jurisdiction.
A gated multi-model loop, now being open-sourced.
WireApps, a software studio, uses a gated multi-model delivery loop to take work from idea to merged code. Meruva is open-sourcing that loop as the Meruva Agent Orchestrator. It is Coming soon — being open-sourced, and not available as a public product yet.
Problem
Premium models for every agent role cost too much for routine work, and each open-model lab meant another key, console and bill.
Setup
Each role — planner, challenger, builder, reviewer, fact checker, designer and QA — can run on a different model. Hard human approval gates sit at plan, design and merge. Several roles already run on open-weight models, including Qwen, DeepSeek, GLM and Kimi. An all-open-weight setup is still to be proven.
What Meruva is doing
The same orchestrator will be part of every Meruva package, designed to run on in-country open-weight models via Meruva Gateway or a Sovereign stack. The open core will be free; we will sell set-up and support.
The Orchestrator is being open-sourced and is not available yet. Read more.
First call in five minutes.
Coming soon You cannot sign up, top up or get a key yet.
Sign up
Create an account and top up by card through Stripe.
Get a key
Create a key in the dashboard. One key for every model.
Connect
Point your app or coding agent at Meruva and pick a model.
# OpenAI-compatible: change the base URL and the model from openai import OpenAI client = OpenAI(base_url="https://api.meruva.ai/v1", api_key="MERUVA_API_KEY") r = client.chat.completions.create( model="qwen-max", # or deepseek-chat, kimi-k2, glm-4 messages=[{ "role": "user", "content": "Hello"}]) # Example endpoint and model names; final values in the docs.
No logo wall. Straight answers instead.
We say where it runs
Every request's route is shown up front, next to sign-up, not buried in the terms.
Open models, easy exit
OpenAI-compatible API and open-weight models. Change your base URL and go.
People you can reach
Two founders who answer setup questions themselves. Meet us.
What we can't claim yet
No ISO 27001 or SOC 2 yet. Our progress is on the Trust page.
Pay for what you use. No subscription.
Launch pricing, coming soon. You cannot buy or top up yet. The figures below are indicative only, not an offer or a checkout.
| Model example prices | Input / 1M tokens | Output / 1M tokens | Cache read / 1M |
|---|---|---|---|
| Qwen (example) | £1.60 | £4.80 | £0.32 |
| DeepSeek (example) | £0.25 | £0.95 | £0.03 |
| Kimi (example) | £2.40 | £11.90 | £0.24 |
Launch pricing, coming soon. Indicative only, not an offer. Sovereign is quoted per project after discovery.
Questions we get.
Which models can I use?
Qwen, DeepSeek, Kimi, GLM and more. Exact versions change as labs release new models; the current list will be published on the models page.
Where are Gateway requests processed, and what happens to my data?
Gateway requests are processed by our upstream provider in mainland China, under PRC law. Meruva stores account details, billing records and usage metadata (model, token counts, cost, time), not your prompts or outputs. Our upstream says it keeps no request logs, but until we can publish written terms, assume prompts may be retained. The Gateway suits experiments, side projects and code you're happy to share; please don't send personal, regulated or confidential data. For that, use Meruva Sovereign, which runs in your jurisdiction. Full details are on our Trust page.
Who is Meruva for?
Developers and startups start on the Gateway. Sovereign is then sold fastest first: enterprises (banks, insurers, law and pharma), then national verticals such as health, universities, defence, energy and regulators, then government. We start in the UK, then grow one market at a time; the same stack works in any country.
What is the Meruva Agent Orchestrator?
A gated multi-model agent delivery loop, included in every package and designed to run on in-country open-weight models via Meruva Gateway or a Sovereign stack. The open core will be free; we sell set-up and support. It is being open-sourced and is not available yet.
Are you partnered with the model labs?
We work with open-weight frontier models, and we are talking to leading labs. Those conversations are contacts today, not signed partnerships.
Why no US models?
We focus on open frontier models you can also run yourself. That keeps the Gateway and the sovereign stack consistent, and US labs' terms do not allow us to resell their APIs. If you need US models, use them directly alongside Meruva.
Does Meruva train on my data?
No. Meruva never trains on your data.
How is pricing set?
You cannot buy, top up or get a key yet. When the Gateway opens: pay as you go by card, per model usage, shown in the dashboard before you commit. The figures on the Pricing page are launch pricing, coming soon, and are not an offer. Sovereign: a scoped project plus a support agreement, quoted after discovery. Agent Orchestrator: the open core will be free; we will sell set-up and support.
Can I start on the Gateway and move to Sovereign later?
Yes, once the Gateway is open. It is the same stack, so your integrations will carry over, and your data moves into your own jurisdiction. You cannot get a Gateway key today.
How do I leave?
The API is OpenAI-compatible and the models are open-weight. Change your base URL and go.
What certifications do you hold?
None yet. Meruva doesn't hold ISO 27001, SOC 2 or Cyber Essentials certification, and isn't listed on G-Cloud. If your procurement needs one of these, tell us early and we'll be straight with you about timing.
Pick your path.
Gateway sign-up is coming soon. Enterprises, sectors and governments can start with a conversation now.