Hosted in the EU, by designWalma AI Hub puts a governance layer in front of Claude, GPT, Cursor and Codex. Set budgets, enforce model policy, and see every euro spent. Your prompts and keys never leave the EU.
Developers reach for whatever model works. Finance sees the bill a month later. Security finds out when the data has already left the building.
Personal API keys, unapproved tools, no attribution. You can't govern what you can't see, and right now you can't see any of it.
Usage-based pricing with no budgets means the cost only becomes real when it's already too late to do anything about it.
Every prompt through a US endpoint is a GDPR question you don't want to answer in an audit, and most teams can't even tell where it went.
Not another model. The governance layer that sits in front of all of them.
Personal keys, personal tools, zero visibility. Cost and data out of control.
One gateway in front of every model. One policy, one invoice, full visibility.
Walma runs in your own EU cloud region. Prompts, keys and usage data never leave the EU. Governance and legal say yes instead of blocking the rollout.
Per-user and per-team budgets, hard caps, automatic model downgrades, and a live month-end forecast per model. Turn an unpredictable bill into a line item you control.
Claude, GPT, Cursor, Codex and your own integrations behind one key and one policy engine. Self-service install gets a developer productive in minutes, with no vendor lock-in.
Point your tools at the gateway. Developers run one signed installer, or paste a BASE_URL and key into their settings. SCIM syncs the rest.
Set budgets, model allowlists and quota tiers per team. Policies apply at the gateway, so there's nothing for developers to configure or bypass.
Watch cost, adoption and policy events in real time, filtered to any month, quarter or custom range, in the currency you report in.
Walma projects each model's spend forward using its real daily history (weekday rhythm, adoption trend and all), so the forecast bends with reality instead of drawing a straight line to a wrong number.
Your team's flow shouldn't stop because one provider had a bad afternoon, and every request should cost as little as it can.
When Anthropic wobbles, Walma fails over across providers and regions in milliseconds. Your developers keep shipping while the status page is still loading.
Our repo-context, caching and retrieval plugins trim the tokens every task burns, so the same work costs measurably less. Bring your own MCP servers too.
429s, timeouts and rate limits are retried and rerouted at the gateway. A blip at the provider never becomes a failed request in someone's terminal.
Every prompt, file and tool response is an attack surface: poisoned packages, injected instructions, quiet exfiltration attempts. Walma inspects every request and stops the bad ones before they reach a model or your codebase.
When an agent pulls event-stream@4.0.1, Walma checks it against deps.dev & Socket and blocks the known-bad version before it touches your repo.
Hidden instructions in files, web pages and tool output that try to hijack the agent get flagged and stopped, not executed.
IP allowlists, business hours and model allowlists apply to the request itself. There's no client-side switch for a developer to turn off.
Every flagged request (who, what, when, which repo) is logged and exportable. Audits become a query, not a fire drill.
They want the best AI, but the rules say stop. We build it so legal, security and IT can say yes.
Claude, GPT, Mistral and the other leading models run in Azure regions inside the EU. Never outside.
The gateway lives in your own Azure tenant, in the EU region you choose. Your agreement, your keys.
Your tools reach the nearest region. Prompts, responses and logs never leave the EU.
Claude, Codex and the other leading models on European servers. One contract, one invoice, one place to set identity, policy and budget.
AI Act, GDPR and other regulation. The objection that stalls every European deal is the one we answer first.
Every use recorded and costed per team, per person, per project. Value shown, not assumed.
By working with AWS and Microsoft we've been able to deploy quickly and win big customers, and to help them along on the AI journey. Our customers choose us as inference provider and strategic AI partner. And they stay, because we make sure AI turns into ROI.
Built for European requirements from day one. Details, subprocessors and policies live in our Trust Center.
Open the Trust CenterBeyond the AI Hub, we build RAG databases and agents tailored to your data, your processes and your organisation, on our backbone Noda and Ocle.
Discover more →
Drag the sliders for a back-of-the-envelope estimate, then see what governance claws back.
Rough estimate on typical token usage. Book a demo for your real numbers.
A 20-minute walkthrough with an engineer, no slideware. We'll map it to your tools, your region and your budget model.
An engineer will reach out within one business day to set up your walkthrough.
Walma is deployed in an EU region of your Azure tenant. Prompts, responses, keys and usage rollups stay in the EU. Requests to model providers use your own contracts; nothing is routed through a US-hosted Walma service.
Claude (Opus, Sonnet, Haiku), GPT via your OpenAI key, Cursor, Codex, Claude Code, and the Microsoft 365 add-ins. Any OpenAI-compatible client can point at the gateway, and you can add your own integrations.
Most teams are governing real traffic the same afternoon. Developers run one signed installer or paste a base URL and key; SSO/SCIM handles provisioning for the rest.
No. You bring your own model contracts and pay providers directly. Walma is a flat per-seat fee for the governance layer, so the more you save on spend, the better the deal gets.
Policies are enforced at the gateway, not in the client. Budgets, model allowlists and business-hours rules apply to the request itself. There's no client-side setting to turn off.
One gateway for every tool, hosted where your data belongs.