Your agents.
Your ground.

Udal is a self-hosted platform for building, running, and governing AI agents inside your own environment. It provides a declarative manifest format, a typed Python SDK, an OPA policy-as-code engine, hard per-team budgets with kill-switches, and a tamper-evident audit log. The vendor never sees customer data.

Copilot Studio and other SaaS agent platforms are hosted services: your prompts, tool calls, and often your data traverse and sit inside the vendor's boundary. Udal is installed software that runs inside your Kubernetes cluster. Data never leaves your environment, agents inherit the invoking user's real permissions, and you get the same governance, audit, and cost controls without renting them from an outside landlord.

Yes. Udal ships as signed container images with an SBOM and SLSA provenance, and there is an air-gapped install bundle. Telemetry is off by default. The vendor has no runtime dependency on your environment.

Every agent request passes through a central control plane before any action happens. That plane enforces policy-as-code (OPA), applies RBAC and PII redaction, requires dual-control approval for high-stakes actions, tracks token spend against per-team budgets with automatic kill-switches, and writes every decision to a tamper-evident, hash-chained audit log.

Udal runs on Amazon EKS, Azure AKS, Google GKE, Red Hat OpenShift, and bare-metal Kubernetes. The same manifest and control plane runs on all of them, including air-gapped installations.

Request access from the contact page. A human replies within one working day and books a technical briefing for your platform, security, and risk teams — or sends the security pack (signed images, SBOM, SLSA provenance, threat model, and control-framework mappings).