How it works
From a raw signal to an approved fix.
Six layers, each with one job and a record of what it handed to the next. Read this to know what the agents see, what they can touch, and where a person stands.
The layers
Six layers, one direction.
Only the ingest layer writes telemetry. Only a gated, verified action can change anything, and nothing changes without an approval.
- 1
Sources
What you already run. NetPulse listens to it and does not replace it.
- SNMP and traps
- gNMI
- gRPC dial-out telemetry
- NETCONF
- OpenTelemetry (metrics, logs, traces)
- NetFlow and IPFIX
- Syslog
- Kafka and message buses
- REST and API polling
- Prometheus
- TWAMP and path probes
- Cloud flow logs and metrics
- Inventory and CMDB
- Log shippers
- 2
Ingest
Each source is normalized into the same fact tables. Records that fail validation are set aside for inspection rather than dropped. Credentials are held by reference, not stored in a config file.
- 3
Facts and the entity graph
Sites, devices, links, tunnels, providers and applications as entities, with typed edges between them. A metric, an alert, an incident and a change all resolve to the same entity. Root-cause analysis walks this graph.
- 4
NetPulse agents
Detection, Noise, Anomaly, Correlation, Root-Cause Analysis, the Incident Agent, the Remediation Agent and Forecast. Rule-based, measured and reproducible.
The eight agents - 5
CEEZ Crew
The incident is handed to a team. Specialists query the facts, a commander adjudicates, a planner maps the cause to a playbook, an executor carries it out under your approval gates, and a verifier checks recovery. Results come back onto the incident.
The NOC team - 6
Outcomes
One ticket per incident, a page for what is critical, a recommended fix behind an approval, a written report, and boards and a wall for the people watching.
Autonomy
Earned on evidence, lost on a regression.
Autonomy is not authority, and neither says how much an action matters. Each fix has its own tier, set per playbook and per kind of incident.
- 1
Suggest only
The agent proposes and explains. People do the work.
Recommend only - 2
Approve every action
A named person approves each step before it runs.
Approval first - 3
Approve the exceptions
Routine work inside the limits runs on its own, and tells you.
Automatic with notice, once earned - 4
Autonomous within limits
Runs inside a bounded policy envelope, on the record, reverting on a regression.
Within a set envelope
- Recommend only is where every playbook starts, and where one stays until it has a proven track record and a rollback plan.
- Approval first needs a good record for that kind of incident. It runs with an approver and is held without one.
- Automatic with notice needs a strong record, repeated verified recoveries and a proven rollback.
- A regression reverts it. Autonomy moves back down automatically, and the demotion is recorded.
Guardrails
What stops an agent doing too much.
Each of these is enforced at run time, in order, before an action is allowed.
Kill switch
One control stops every action. It is checked first.
Change freeze
An active maintenance window blocks a fix on the entities it covers.
Rollback first
A fix with no rollback plan is recommend-only. It cannot be promoted.
Blast radius
An action over its cap runs as a canary first. Over three times the cap it is refused outright. Autonomous work outside its envelope drops back to approval.
Verify, then close
Every action is checked against the condition that fired. A fix that does not clear is rolled back.
One ticket per incident
A recurring problem updates one record. Hand-overs and suppressions are logged too.
Security is exempt from noise rules
The Noise Agent never hides a security alert by default.
Evidence only
Investigating workers read live evidence through read-only queries and must cite each one.
Control
You stay in control.
Four things that are true however much autonomy has been earned.
- Nothing changes without a gate. Every fix passes the kill switch, change freeze, rollback and blast-radius checks first.
- Your change process stays yours. Any recommendation can be taken into your own tooling instead of being executed here.
- Observing agents report before they act. Noise, Anomaly and Forecast build a record first, and you decide when they page or suppress.
- Every call is explained. Each root cause, suppression and fix carries its evidence, so you can audit it later.
See it on your network.
Connect a source and follow a use case on live screens.