Plain answers about Udal.

Udal is an enterprise AI control plane. It lets organisations build, deploy, run and govern AI agents inside their own environment, so they keep control of their infrastructure, data, models, security, governance and costs. The brand line is "Your agents. Your ground."

A control plane is the layer that decides what AI agents are allowed to do. It checks who is asking, applies your policy, chooses an approved model, mediates access to tools and data, pauses for human approval where required, enforces budgets and records everything in a tamper-evident audit trail.

Regulated and accreditation-sensitive organisations in the UK and Europe, including the public sector, financial services, healthcare, legal, energy and industry. It is designed for the whole leadership team — CEO, CFO, CRO or General Counsel, CISO and CIO — as well as the technical teams who run it.

Boards want AI, but risk, compliance and security teams block it because data, cost and accountability are unclear. Pilots stall. Udal removes those blockers by letting organisations run governed AI agents on their own infrastructure, so they can say yes to AI without taking on new risk.

No. Udal is not a model and not a chat product. It is the platform that runs and governs agents, and it works with the open-weight and frontier models you choose.

Udal is an old Norse and Scottish form of land tenure meaning outright ownership, free of any feudal superior. It reflects the idea of owning your AI rather than renting it from a provider.

Udal is built by Exception (exceptionuk.com), following 20 years of helping regulated businesses adopt technology safely.

Inside your own environment: AWS, Azure, Google Cloud or your own servers. It can also run on premises, at the edge or fully air-gapped. Udal is installed in your infrastructure rather than being a service you send data to.

No. AWS is the documented reference deployment (Helm on Amazon EKS in your own account), but Udal also runs on Azure, Google Cloud and your own hardware or private cloud.

Yes. Udal can be deployed in air-gapped environments using self-hosted open-weight models, with no outbound internet access.

Not unless you allow it. By default work stays inside your boundary. External frontier models are opt-in and allow-listed by policy, so you decide which data, if any, can be sent to them.

No. Udal is licensed software that you run in your own environment. There is no shared multi-tenant service holding your data.

Both self-hosted open-weight models and commercial frontier models. Open-weight examples include Qwen, Llama, Mistral, DeepSeek, Gemma and Phi, plus models published on Hugging Face. Frontier options include OpenAI, Anthropic and Google models, Azure OpenAI and Amazon Bedrock.

Through policy-based routing. Agents ask for a class of model, and the gateway maps that to an approved backend. Sensitive and internal work can be kept on open-weight models inside your boundary, with frontier models used only where they are allowed and earn their cost.

Yes. Because agents request a model class rather than a specific provider, you can change the underlying model without rewriting agents. This reduces dependence on any single provider.

Zero Data Retention is a useful contractual promise, but it is a policy, not a boundary — data still travels to a third party's infrastructure. Udal lets you keep sensitive work inside your own environment so it does not need to leave at all.

Every request is checked against your policy before it runs. New installations grant no permissions until you define them, agents reach tools and data only through controlled gateways, and the system fails closed if a required control is unavailable.

Yes. Workflows can pause for human approval, with separation of duties, and notifications through Teams, Slack, email or webhooks.

Pluggable PII and content guardrails can detect and block personal data and unsafe content before it reaches a model, and audit logging can record metadata only.

Udal keeps a tamper-evident audit chain of what each agent did, who authorised it, which model and tools were used and what it cost. Sealed evidence reports and exports support your assurance process.

Yes. Each project runs in its own sandboxed worker with its own security and isolation, so one project's work does not share an execution environment with another.

Cyber Essentials Plus certified and listed on G-Cloud 14 and 15. ISO/IEC 27001 certification is in progress, planned for early 2027.

Udal supports UK GDPR and data residency by keeping data in your environment. An optional EU AI Act module supports risk classification and an AI system register; it supports compliance work but is not a guarantee of compliance.

A Pilot is £6,000 per year for your first agent. The full Platform is £30,000 per year for one environment, up to five projects and unlimited agents, with additional projects at £4,000 each. Every level includes every capability.

A project is a distinct area of work that needs its own security and isolation — for example a separate department, data set or regulatory scope.

Model usage is metered per agent and project, and hard budget stops halt work that exceeds its budget, so a usage spike cannot turn into an unexpected bill.

Support runs 9am to 5pm, five days a week, with a named contact.

Start with a 30-day governed pilot or book a technical briefing through the contact page. A person replies within one working day.

Your agents keep running. You can route work to another approved model, including self-hosted open-weight models you control.

Udal runs in your environment, so it keeps working. Releases are signed and come with a software bill of materials (SBOM), and your agents, policies, data and audit records stay in your infrastructure.

Provider platforms run your agents on their infrastructure under their terms. Udal runs them on yours, with your policies, your choice of models and your own audit evidence — owning rather than renting.