Limits and quotas

Default values for every limit: concurrency, connection rate, timeouts, account counts, whitelists, rate limits and risk-control thresholds.

One checklist to consult while designing concurrency and retry strategy. Every number is the current production default and may be adjusted per node.

Connections and concurrency

Item	Default
Total gateway connections	100,000
Pre-auth concurrency per source IP	5,000
Pending handshakes	10,000
New connection rate per source IP	1,000/s, burst 5,000
New connection rate per account	2,000/s, burst 5,000
Per-account concurrency baseline	5,000

Per-account concurrency is dynamic

5,000 is not a hard ceiling under light load. The gateway adjusts with overall pressure:

Gateway load	Per-account concurrency ceiling
≤ 60%	Can relax up to the total gateway connection count
60% – 85%	Tightens linearly between the two
≥ 85%	Tightens to 5,000 (or a lower value from the account snapshot)

The per-source-IP 5,000 only protects the pre-auth stage

Once authentication succeeds, the source IP's concurrency slot is released. So that limit exists to prevent floods of unauthenticated connections, not to cap your business concurrency.

Authenticated account concurrency or connection-rate limits return HTTP 429 connection_limit; exhausted traffic or account quota returns 402. Protection before protocol detection may close TCP directly, and an established tunnel cannot receive a new HTTP error status. See the error reference.

Timeouts

Item	Value	Meaning
First byte	5s	How long after connecting the first byte must arrive
Full handshake	10s	Total budget for the whole proxy handshake
Authentication	3s	The gateway's auth call to the control plane
Target policy / DNS	8s	Target address check and hostname resolution
Exit TCP connect	8s	Default; connection conditions may result in an earlier timeout
Exit handshake	30s	Default; connection conditions may result in an earlier timeout
Idle	5 minutes	Global ceiling; connection conditions may result in an earlier timeout
Maximum connection lifetime	24 hours	For established relays
Half-close drain	15s	Time to drain reverse data after a TCP half-close

Accounts and resources

Item	Limit
Proxy sub-accounts per user	10
Proxy sub-accounts per user (first 24 hours)	2
IP whitelist entries per sub-account	10
API keys per sub-account	1
Sub-account traffic cap	1 MB – 1 TB
Open support tickets per account	20
Notification channels per traffic alert	20

What exceeding them returns:

Case	Response
Sub-account limit exceeded	`409 proxy_account_limit_reached`
Whitelist limit exceeded	`409 proxy_api_limit_reached`

API rate limits

Endpoint	Limit
Extract API	120/minute, counted separately per source IP and per API key
CDK redemption	20/minute
Registration code (per email)	3/hour
Registration code (per IP)	10/hour, 30/day
Code resend interval	60s
Successful registrations per IP per day	3 accounts

Exceeding returns 429 with Retry-After: 60.

Item	Threshold
Failure counting window	10 minutes
Failures per IP	30
Failures per account	10
Failures per IP + account pair	5
Cooldown once triggered	15 minutes
Verification code lockout	5 attempts
Verification code validity	600s
MFA ticket validity	5 minutes

Hitting any threshold writes a cooldown, during which sign-in returns 429 directly.

Token lifetimes

Item	Default
Access token	15 minutes
Refresh / session	12 hours
Refresh (with "remember me")	30 days
MFA ticket	5 minutes

See Authentication and conventions.

Orders and quotes

Item	Value
Dynamic residential quote validity	5 minutes (shorter if a promotion or delisting comes sooner)
Unpaid order expiry	30 minutes
Stripe checkout link retention	30 days
Static residential per-cart	100 IPs, 100 per country
Static residential lease	Fixed 30 days
Datacenter lease tiers	7 / 30 / 90 days
CDK face value	100 MB – 1 TB
CDK validity	1 – 30 days

Data retention

Data	Retention
Daily usage aggregates	Long-term
Hourly usage detail	60 days
Hourly per-domain detail	60 days
Domain list	Most recent 500
Daily endpoint per-call range	1 – 90 days
Whitelist auto-cleanup when unused	10 days

See Usage statistics.

Target restrictions

Item	Rule
Private / loopback / link-local / multicast addresses	Blocked
Target port 25	Blocked
Hostname resolution	Resolved at the gateway and pinned to the first IP
UDP / QUIC / WebRTC	Unsupported (SOCKS5 has no UDP ASSOCIATE)

See Target restrictions.

Option value ranges

Option	Range
`country`	Two uppercase letters
`state` / `region` / `city`	1–100 bytes (UTF-8)
`asn`	1 – 4294967295
`session`	1–64 bytes, `[A-Za-z0-9.-]`
`time`	1 – 120 minutes
Full username	1–255 bytes
Base username	1–64 bytes
Password	1–512 bytes (255 in practice for SOCKS5)

See Username option reference.

Limits that don't exist

Listed so you don't design around mechanisms that aren't there:

What you might assume exists	Reality
A general QPS limit on the client API	No separate configuration
A hard daily traffic ceiling	None
A global session count ceiling	None
ASN extraction quotas	The Extract API doesn't support that dimension at all
Monetary balance alert thresholds	Traffic balance alerts only, no monetary alerts
Automatic account suspension rules	None

Designing concurrency around the limits

A workable starting point:

python

# Per-account concurrency has a 5,000 baseline, but the real bottleneck is
# usually your own machine and the target site, not the gateway. Start small.
MAX_CONCURRENCY = 64

# The Extract API allows 120/min, so don't call it per request.
# Extract a batch at startup and cache it.
EXTRACT_INTERVAL_SECONDS = 60

# The gateway idles out at 5 minutes; keep the client pool shorter so you
# never hand out a connection the gateway has already closed.
KEEPALIVE_TIMEOUT_SECONDS = 90

# Client timeouts slightly wider than the gateway's, so you get the gateway's
# status code rather than your own timeout exception.
CONNECT_TIMEOUT_SECONDS = 15   # gateway: 8s
TOTAL_TIMEOUT_SECONDS = 120

More in this section