EdgePilotAI CDN Orchestrator
LinUCB reinforcement learning · sub-ms edge decisions

The CDN that picks itself.
One request at a time.

EdgePilot is an AI agent that sits between your users and every major CDN. For every request, it decides which provider — Cloudflare, Akamai, Lumen, Fastly, CloudFront — will deliver the lowest latency at the lowest cost. It learns from every request.

5
CDN providers
6D
feature context
< 5ms
decision latency
LinUCB
algorithm
The decision loop

Four signals, one decision, per request

Live telemetry

Continuous latency, error-rate, saturation and cost samples ingested from each provider's health API and RUM beacons.

Contextual bandit

LinUCB estimates the expected reward of each CDN as a linear function of the request context — region, content type, size, time of day.

Edge execution

Deploy the agent as a Cloudflare Worker or Lambda@Edge so the decision runs in single-digit milliseconds next to the user.

Continuous learning

Every observed latency updates the posterior. The agent explores enough to discover regressions, and exploits the winner otherwise.

Speed

Route Asia-Pacific video traffic through Akamai during peak hours, EU static assets through Cloudflare, LATAM through Lumen — automatically.

Cost

The reward function blends latency with $/GB. When two providers tie on speed, the agent picks the cheaper one — silently trimming your egress bill.

Resilience

Abnormal traffic patterns can be routed to a scrubbing provider. When a CDN degrades, the bandit shifts traffic away within minutes, not incidents.

See the agent decide, live

Open the control room to fire simulated requests, watch the LinUCB scores per provider, and see the bandit shift traffic as it learns.

Enter control room