Overview
Novesis separates deciding from doing. A control plane holds the policies you have written and a live model of your fleet. A data plane in each region carries your traffic and reports what it sees. The control plane sends decisions down; the regions send telemetry up.
Because the loop runs continuously and automatically, routine decisions about where to run work and where to send requests do not wait for a person.
Control plane
The control plane has three cooperating parts. Each reads the same model of the fleet and each is bound by the same policies.
- Capacity decides how much of each workload runs, and where.
- Routing decides where requests are sent.
- Rollback watches releases and reverses them when service levels slip.
Policies are declarative. You state the outcome you want, for example a latency budget or a rule about where data may live, and the control plane works out how to satisfy it.
# checkout-api latency.p99 <= 40ms regions allow [eu, us, apac] data.residency eu-customers -> eu rollout canary 5% · hold 90s · rollback on slo-breach
Data plane
Each region runs a data plane that serves traffic and a lightweight local agent that reports health and saturation. If the control plane is unreachable, the data plane keeps serving with the last known good configuration. Losing the decision-maker never stops traffic.
Capacity
The capacity component combines current load, recent trends and your policies to decide how much headroom each region needs. It scales up ahead of demand where it can, and scales down where doing so is safe, so you carry less idle capacity without risking your service levels.
Routing
Requests are steered at the edge using live measurements of latency and health per region. When a region degrades, traffic shifts away from it gradually rather than all at once, which avoids overloading the regions that absorb it.
Rollback
Every release goes out in stages. A small share of traffic sees the new version first, and the control plane compares its behaviour with the previous version against your objectives. If the new version falls outside them, the rollout stops and reverses without waiting for a person to notice.
- Rollouts advance only while service level objectives hold.
- Reversal is automatic, and the previous version stays ready until the rollout completes.
- Every step is written to the decision log.
Security model
The control plane can change how your production systems run, so access to it is tightly controlled. Actions are authenticated, authorised against the policy that governs them, and recorded. Automation acts within the limits you define and cannot widen those limits by itself.
For our wider approach to protecting environments, see Cloud Security.
Observability
Automation you cannot inspect is automation you cannot trust. For each decision, Novesis records the signals it saw, the policy it applied and the action it took, so an engineer can read the reasoning after the fact and challenge it if needed.