Run a bigger network with the team you already have.
Rapax is AI-native service assurance for telecom operators. Six AI agents handle fault correlation, alert triage, Tier 1 support, and integration work continuously — so your operations scale with the network instead of with headcount.
What happens when one line card fails?
Forty-nine alarms. One root cause. Forty-eight subscribers who don’t know yet.
An OLT line card fails at Kings Garden. Every ONT behind it loses GPON signal and reports independently. Under a manual model that is an engineer reading forty-nine rows to work out they describe one event — thirty to sixty minutes before anyone knows which customers are affected.
Rapax collapses it to a single service record in thirty-seven seconds: root cause, affected segment, subscriber count, service tier. One thing to act on instead of forty-nine things to read.
The pattern repeats across every layer of the operation. Alarm volume grows with the network. Support tickets grow with the subscriber base. Integration work grows with every vendor added. Headcount is the only lever most operators have, and each NOC engineer costs $150,000 to $200,000 fully loaded and spends most of the first year getting up to speed.
Could you have seen it coming?
Yes. That line card ran at 91°C against a 55°C daily average, and its optical power fell from −18 dBm to −30 dBm over the three hours before it failed.
The signal was there the whole time. Nobody was watching that particular metric on that particular card, because nobody can watch every metric on every card. That is what continuous monitoring is for — not dashboards for people to stare at, but agents that notice.
What does Rapax actually do?
Six AI agents, each owning a layer of network operations.
| Frank | Intercepts incoming support tickets. Deflects what your knowledge base can answer, auto-closes routine questions, validates intake, and escalates the rest with context attached. |
| Oscar | Continuous operational briefing. Removes the time managers spend joining bridges and pulling status. |
| Wade | Institutional memory. Retrieves runbooks, procedures, and vendor documentation, with a confidence score on every answer. |
| Grace | Alert notifications and ticket updates, drafted and sent automatically, with subscriptions and quiet hours. |
| Nora | Conversational network inventory. One source of truth for device topology. |
| Bruce | Builds and deploys integrations — Syslog, SNMP, webhooks, REST — and commits the code. |
Ask Oscar what is happening and you get a briefing, not a dashboard — affected service, downtime, customer tier, and the physical location of the fault down to the PON port.
The agents are the architecture, not a feature bolted onto a monitoring tool. And Rapax runs on any LLM provider, including fully local deployment. If your data cannot leave your environment, it does not have to.
What’s new in v1.2
Frank — Tier 1 support that runs itself
Around 60 to 70% of support tickets do not require human judgment. They are knowledge base questions, incomplete reports, duplicates, and false alarms. Frank picks up every incoming ticket within seconds and makes five decisions on each one.
Deflect what your knowledge base can already answer — no ticket is created at all. Classify it correctly, even when the customer files a bug that is really a configuration question. Validate intake, so missing logs and repro steps get requested automatically and your engineers never open an incomplete ticket. Check reported outages against live telemetry, downgrading with an explanation when monitoring is green — and never silently upgrading, because outage priority carries billing and paging consequences that stay with the customer. Escalate what needs a person, with context attached, and page on-call immediately when a customer’s tone shifts.
Frank is grounded, not generative. Every decision is anchored to Wade, which returns a confidence label on every query — high confidence deflects or auto-closes, medium drafts a reply for review, low escalates to a person. You set the thresholds. If the answer is not in your documentation, Frank does not invent one.
Frank deflects roughly 40% of incoming load and auto-closes about 15% of tickets, deferring four to nine support hires over twelve months.
Native incident management
v1.2 adds incident management to Rapax itself. Tickets, queues, SLAs, and escalation paths now live in the same platform as your fault correlation and your network inventory — so an alert, the service it affects, the customer on that service, and the ticket about it are one object instead of four systems and a synchronization job.
Running ServiceNow, Jira, or Zendesk today? Rapax migrates your history and you retire the license. Prefer to keep it? Rapax integrates over REST and Frank works against it in place.
How do you know it works?
Frank runs our own support desk. Rapax has operated Frank in production on its own support queue since May 2026. The deflection and auto-close rates are production numbers from a live queue, not a lab benchmark.
A mid-market fiber ISP deployed it in under two hours. 600,000+ subscribers across 19 markets in 7 states, GPON and XGS-PON, multi-vendor. They had solid subscriber telemetry but no way to answer the only question that matters during an outage — how many customers are down right now. Rapax consumed their existing Kafka streams, built the topology graph from backbone to subscriber, and published correlated service-impact events back into their stack. No agents. No schema changes. No downtime. Reference available on request.
The platform is tested and audited. 369 automated tests across 12 suites. 179.9 ms average API latency. 10,000 events per second write throughput. Validated at 200 concurrent users. An OWASP ZAP audit across 97 URLs returned zero high-severity findings, with three medium risks accepted and disclosed. (v1.1 validated.)
The team has done this before. Rapax is built by the founder of Assure1, a service assurance platform acquired by Oracle in 2021 — 25 years in telecom operations and 12 patents in OSS/BSS.
Multi-vendor proof at scale. Demonstrated at TM Forum DTW Ignite 2026 as part of a multi-vendor Catalyst project: three sites, 113 devices, six equipment vendors, sub-60-second provision-to-revenue.
How do you find out if it works on your network?
You run it on your network.
Rapax deploys in five days. You evaluate for thirty against success criteria you define in writing before anything is signed, for $25,000 — credited against the license if you move forward. If it does not meet them, the evaluation ends and you owe nothing further.
The ask is fifteen minutes, not a demo.