How it works

From a familiar signal to a trusted outcome.

Git events, schedules, APIs, manual actions and external services all enter one request model. What happens next is the same governed path every time, whatever started it.

Book a demo Explore the platform

Operating model

One path from intent to evidence.

NopsAI connects control-plane decisions and data-plane execution without turning AI work into a separate, ungoverned system.

Signal

Normalize intent

Git, schedules, APIs, manual actions and external services enter one request model.

Govern

Resolve permission

AAA, scope, secrets, knowledge, model profiles and tool profiles are checked before dispatch.

Run

Execute in isolation

The dispatcher selects an eligible Docker or Kubernetes runner for a per-run agent.

Prove

Keep the evidence

Logs, task state, approvals, AI usage, monitoring and outputs remain attached to the run.

The governed run

Six steps, every time, whatever started the run.

The order matters. Deterministic work happens before reasoning, reasoning happens inside a boundary, and nothing high-risk executes before a named human agrees.

Normalize the trigger
A Git event, schedule, API call, external service or manual action becomes one consistent run request with a known effective subject.
Resolve the definition
Pipeline YAML, reusable steps, knowledge context, scope, variables, secrets and access decisions are loaded and snapshotted.
Authorize the caller
Default-deny checks cover the route and every protected resource the run references. Failures fail closed, before dispatch.
Dispatch to a runner
The dispatcher filters connected runners by scope, routing, capacity, affinity and runtime pool, then launches a per-run agent.
Advance the graph
Deterministic tasks and AI goals execute in dependency order. The run pauses at approvals without holding a runner.
Attach the evidence
Status, logs, task history, approvals, AI usage and declared outputs stay owned by the run and survive the conversation.

Architecture

Durable control plane. Isolated data plane.

The API, AAA, configuration, dispatcher and persistence layers decide and record what should happen. Docker and Kubernetes runners execute transient work and report state back through authenticated service boundaries.

Experience layer

Operator UINopsAI CLIREST APIHosted MCP

Control plane

AuthenticationAAA decisionsConfig syncRun lifecycle

Dispatch layer

Scope routingCapacityAffinityRuntime pools

Data plane

Docker agentsKubernetes podsStep runtimesWorkspaces

Evidence layer

Postgres stateRun logsMetricsFinal outputs

Enterprise assurance

Controls stay active across every capability.

Governance is part of route composition, configuration resolution, tool discovery, dispatch, execution and monitoring rather than a separate review step.