A /40 we announce ourselves.
2a10:fa81:1700::/40 from our US gate in Ashburn and 2a10:fa81:1600::/40 from our NL gate, both announced on our own AS218876. The gate binds a source address out of its prefix and connects straight to your destination, 2⁸⁹ addresses across the two to rotate through, no thread limit, and full UDP so QUIC and HTTP/3 just work.
2a10:fa81:17a3:9f2c::e41b2a10:fa81:16b7:4d81::a92fRead this first: the pool is IPv6-only
Every address here is IPv6, so your destination must publish an AAAA record. An IPv4-only target simply cannot be reached on this product - there is no v4 fallback. Most large sites are dual-stacked and will be reached over v6, but check yours before you buy: if your targets are IPv4-only, Datacenter Per GB or Residential is the plan you want. One request against https://ipv6.icanhazip.com tells you where you stand.
Our prefix, our ASN.
One /40 per gate on AS218876, announced by us and served from Ashburn and the Netherlands.
6.19 × 10²⁶ addresses
A /40 is 2⁸⁸ addresses and we announce one per gate, so 2⁸⁹ across the two. Rotation is effectively infinite, an address does not realistically repeat, however long the job runs.
Announced on AS218876
2a10:fa81:1700::/40 in Ashburn and 2a10:fa81:1600::/40 in the Netherlands are our own BGP-announced prefixes. The gate binds a source address out of its own prefix and connects straight to your destination.
No thread limit
Open as many concurrent connections as your job needs, on either plan. There is no cap on threads or sockets to work around.
Full UDP
QUIC and HTTP/3 work. UDP egresses from the same pool address as your TCP connections, so a sticky session stays coherent across both.
Two ways to buy the same pool.
Same prefixes, same gates, same rotation. The only difference is how the meter runs.
Per-GB bandwidth sits on your account until you use it, there is no expiry clock. Unlimited bills by the day, from one day up.
One endpoint, both protocols.
Standard user:pass auth against one host per region. Rotation is the default; session behaviour is spelled out in the username, so nothing about your client has to change.
gate-us.vaultproxies.com:30 · :31 · US gategate-eu.vaultproxies.com:30 · :31 · NL gate-sess-<id>-life-<seconds> · up to 86400-end-die · stop instead of rotating
import requests
# One endpoint, both protocols. HTTP on :30, SOCKS5 on :31.
# The gate binds a source address out of a /40 we announce ourselves on
# AS218876 and connects straight to your target.
HTTP = "http://USER:[email protected]:30"
SOCKS5 = "socks5h://USER:[email protected]:31"
# Two gates, one /40 each. The hostname picks the region - same credential,
# same ports, no country token:
# gate-us.vaultproxies.com United States 2a10:fa81:1700::/40
# gate-eu.vaultproxies.com Netherlands 2a10:fa81:1600::/40
NL_HTTP = "http://USER:[email protected]:30"
# Rotating by default: a different source address on every request.
# Note the target - it must publish an AAAA record.
r = requests.get("https://ipv6.icanhazip.com",
proxies={"http": HTTP, "https": HTTP})
print(r.text.strip()) # -> 2a10:fa81:17xx:... (16xx out of the NL gate)
# Sticky: hold one address for -life- seconds, up to 86400 (24h).
# When the timer runs out the session rotates to a fresh address and
# keeps working. That is the default: end-rotate.
STICKY = "http://USER-sess-a1b2c3d4-life-3600:[email protected]:30"
# -end-die stops the session at expiry instead of rotating. The
# credential is refused once the timer is up, so nothing silently
# continues from a different address mid-sequence.
ONESHOT = "http://USER-sess-a1b2c3d4-life-600-end-die:[email protected]:30"
# The same tokens work on SOCKS5 on :31, and UDP is carried end to end -
# so QUIC and HTTP/3 leave from the same address as your TCP connections.When the target speaks v6.
Dual-stacked CDN targets
Sites behind Cloudflare, Fastly, Akamai or Google commonly publish AAAA records. Where they do, the endpoint is reachable over v6 and the whole pool is in play.
All-day scraping
A scraper that runs continuously is exactly what the flat daily rate is for, past roughly 20 GB a day it costs less than metering the same traffic.
High-frequency polling
No thread or connection limit, so an inventory watcher or price monitor can fan out as wide as your machine allows.
Sequenced jobs
Hold one address with -sess- for as long as the sequence takes, and use -end-die so it stops rather than continuing from a different IP.
Very wide rotation
2⁸⁸ addresses in each /40, one per gate. Per-request rotation across a pool that size means repeats are not a practical concern.
QUIC and HTTP/3 work
UDP is carried end to end, so v6 services that only speak QUIC, and HTTP/3 endpoints generally, are reachable from the same session.
IPv6 Datacenter vs. the rest.
The deciding question is not price, it is your target list. If your destinations publish AAAA records and do not score you on ASN reputation, this is the lowest per-gigabyte plan we sell and the widest pool we have. If either of those is false, another plan fits better, so here is the honest map.
Estimate your bill
Drag the slider. Bandwidth never expires.
Running this every day? Past ~20 GB a day, the flat $3.00/day plan costs less.
Final amount is confirmed at checkout.
IPv6 Datacenter, straight up.
How a request leaves, how sessions behave when the timer runs out, and the one target type this pool cannot reach.