One controller,
for every edge node.
Acceleration, security, monitoring and operations in one controller, all running on your own servers with visitor traffic never passing through a third party.
Tags show how each feature is made available; see the pricing page
Faster first paint,
lighter origin.
Edge hits answer directly; a miss asks the regional L2 first and shares origin fetches within the region, so your origin handles only the requests that need it.
Set cache rules per site; purge and prefetch so updates take effect promptly.
Origin fetches are shared within a region; in a test with 2 edge nodes and 2,400 requests, origin requests fell from 544 to 300 (−45%)
Both HTTP/2 and HTTP/3 (QUIC) are supported and switched on per site; clients upgrade through Alt-Svc negotiation.
103 Early Hints for preloading; when the origin is unreachable, stale cached content can still be served (Always Online).
Spread origin requests across several origins and fail over automatically when one goes down.
Forward TCP and UDP traffic by port, with sessions, peak concurrency and traffic reported per port and protocol.
Measured: 2 edge nodes, 2,400 requests; origin fetches 544 → 300 (−45%)
Certificates renew
before they expire.
Certificates are issued and renewed automatically and swapped by hot update; TLS 1.3 and 0-RTT save returning visitors a round trip.
HTTPS certificates are issued and renewed automatically; nothing to do before expiry.
Certificates are swapped by hot update, with no node restart. No failures were seen during swaps in our tests.
TLS 1.3 with 0-RTT: returning visitors send the request in the first round trip (GET/HEAD only, to prevent replay). Session-ticket keys are shared across nodes and rotated daily.
Six layers of protection,
configured in one panel.
Lists, CC, bots, WAF, upload scanning and auto-ban filter in turn, all at the edge; each layer can be enabled globally or per site.
Rate limiting by IP and URL, with a challenge or block when exceeded, to protect your origin and bandwidth.
A built-in WAF blocks common web attacks, including SQL injection, XSS, command injection, path traversal and Log4j; switch it on per site and choose the action.
Standard/strict protection levels, security posture screen, semantic detection, deep bot analysis, anti-tamper, leak prevention, and upload scanning with trojan signatures, antivirus and a sha256 blocklist
Virtual patches for exploited-in-the-wild vulnerabilities, tracked against the CISA KEV catalog, plus emergency rules, bot lists and a high-risk IP library, pushed hourly to every edge node with no version upgrade
Kernel-level XDP fast filtering: packets from blocklisted sources are dropped at the NIC driver layer, before the kernel network stack; in a test with a 4 vCPU node and 1.17 million pps of blocked-source traffic, network interrupt CPU fell from 65% to 15%
Allow or block by IP or range, globally or per site.
The client TLS fingerprint (JA4) serves as a CC counting key and an access-control condition against automation that rotates IPs; JA4 is also recorded in access logs.
20 checks across TLS configuration, security headers, origin exposure and known-vulnerability probing; protection baselines by site type, with bulk remediation.
Allow or block by visitor region. A free geolocation database is built in; the commercial edition has more accurate province and carrier data, updated faster.
Attack trends and service health,
on one screen.
Traffic, hit ratio and attack posture in one place; when something goes wrong, alerts reach the tools your team already uses.
Bandwidth, requests and hit ratio in real time, with downloadable access logs.
Attack trends, blocked traffic and attack records in one place; the protection score and to-do list reflect checkup results.
Automatic alerts when something goes wrong, delivered through DingTalk, Feishu, WeCom, Telegram or generic Webhook channels; several channels can be configured and test-sent.
Nodes, users and plans,
managed in one place.
Every change rolls out in stages; once an offsite directory is configured, the controller database's increments are pushed offsite every minute.
Manage regions, line groups and nodes together; config is pushed to every node automatically.
Run one command on a supported Linux server to turn it into a CDN node that registers with the controller.
Config rolls out 1 node → 10% → all, advancing only when a stage raises no new alerts; nodes roll back automatically on a regression after a change (5xx ratio, worker crashes). Upgrade packages are signature-verified; node upgrades can be staged and hot-swapped (from 2.1.3).
Sites are isolated per user. Plans bind to regions and line groups and meter usage against quotas; online top-up and real-name verification are supported, and admins manage users, plans and quotas in one place.
The controller database supports offsite incremental backup (binlog pushed every minute) and point-in-time recovery (PITR) to any chosen second, losing at most about a minute of data if the controller host is lost.
Measured performance,
conditions listed per item.
All figures come from real-server load tests; test conditions are listed under each item
Zero-interruption config hot updates
About 2.9 million requests during config pushes and certificate swaps
Cache-hit time to first byte
p95 0.99 ms
Single-node throughput
4-core node, 1 KB HTTPS cache hits; median of three runs about 8.1k–13.8k
Live on a single node about 0.55 s after commit
nginx validation 144 ms
Real-server load tests; nodes with 4 vCPU


