An assistant that knows your network, not just the internet

Local, private, offline

The language model runs next to the platform, on your server. Configs, IPs and logs are never sent to a cloud service. You can also plug in any OpenAI-compatible provider if you prefer.

Answers grounded in real data

The assistant has read-only tools into the platform and reads the data before it answers.

• Device state and open alerts
• SLA summaries and alert history
• NetFlow top talkers
• Expiring certificates and WAN path quality

Incident analysis with evidence

An “Analyze” button on every alert opens the chat with full context. Root-cause answers must cite evidence: timestamps, config versions, log lines and topology neighbours.

Config security audit

A device's stored configuration is checked against a security checklist; findings come back with severity, explanation and remediation commands — and any finding can become a four-eyes change proposal.

Shift reports and postmortems

A handover briefing goes to the team's channels at the end of every shift, and a draft postmortem is ready when an incident resolves. If the model is unavailable, a fixed Persian/English template is used instead.

Plain language instead of forms

Ask “which targets flapped this week?” or say “alert me if the Mashhad link goes above 200 ms”. Creating an alert rule is the assistant's only write action, and it always asks for confirmation first.