Governance & lifecycle

Author → Check → Promote → Govern → Observe.

One contract from a developer's laptop to a production run. The manifest a team writes is the resource the cluster runs and the policy engine checks — no translation, no drift.

Stage I

Author

Describe an agent's model, tools, memory, guardrails, approvals and budgets in one declarative manifest. Custom logic is standard Python against a typed SDK.

Stage II

Check

Two-stage validation, policy gates and evals run in CI. Behavioural regressions fail the build before anything ships.

Stage III

Promote

GitOps promotes the agent through dev, staging and production. A Kubernetes operator reconciles it into a running workload — no tickets.

Stage IV

Govern

Every request passes through a policy-as-code engine. Agents inherit the invoking user's permissions. High-stakes actions require dual-control approval.

Stage V

Observe

A tamper-evident, hash-chained audit log records every decision, tool call and model interaction. Traces flow to Splunk, Datadog or ELK.

Author as code

One manifest is the contract.

The same declarative resource a developer writes is what the cluster runs and what the policy engine checks. Custom logic is ordinary Python against a small, well-typed SDK.

apiVersion: udal.io/v1alpha1
kind: Agent
metadata:
  name: treasury-analyst
spec:
  identity: scoped-analyst-role
  budget:
    daily: $50.00
    kill_switch: hard
  guardrails:
    - redact-pii
    - approval-required: over-$10k
  tools:
    - name: oracle-erp
      scope: read-only

Provable, on the tools you already run.

OpenTelemetry traces, hash-chained audit logs, and per-team cost attribution — exportable as compliance evidence.

See what each leader gets

Run this lifecycle on your own ground.

Technical briefing, not a sales deck. A human replies within one working day.

Request access