* fix(proxy): keep native runtime loopback requests direct In 9.4,8169aacddaadded the native Codex loopback inference relay while managed proxy routing still cleared NO_PROXY. Use Proxyline's shared bypass policy and supply loopback exclusions to native children. Explicit proxy/block modes remain unchanged. Doctor reports failed loopback probes with recovery commands. Fixes #153930. Reported by @eefreenyc. * fix(proxy): align loopback diagnostics and client regression coverage Keep default loopback routing direct after canary readiness and assert that external requests still reach the proxy. Attribute readiness failures to the proxy only in explicit proxy mode. Keep the TUI client and proxy lifecycle in one module generation after mock resets, assert the full block-mode recovery message exactly once, and verify the inherited policy at Gateway WebSocket construction. Include the existing test-only outbound custody repair fromc159663b12(#154123), preserving nested maintenance, physical exclusion, native writability, failure causes, and migration receipts without longer budgets. Related: #153930 * test(proxy): remove userinfo from Doctor fixture
21 KiB
summary, title, read_when
| summary | title | read_when | ||
|---|---|---|---|---|
| How to route OpenClaw runtime HTTP and WebSocket traffic through an operator-managed filtering proxy | Network proxy |
|
OpenClaw can route runtime HTTP and WebSocket traffic through an operator-managed forward proxy. This is optional defense in depth: central egress control, stronger SSRF protection, and destination auditability at the network boundary. Because the proxy evaluates the destination at connect time, after DNS resolution and immediately before it opens the upstream connection, it also narrows the gap a DNS-rebinding attack relies on between an earlier application-level DNS check and the actual outbound connection. A single proxy policy also gives operators one place to enforce destination rules, network segmentation, rate limits, or outbound allowlists without rebuilding OpenClaw.
OpenClaw does not ship, download, start, configure, or certify a proxy. You run the proxy technology that fits your environment; OpenClaw routes its own HTTP and WebSocket clients through it.
Configuration
proxy:
proxyUrl: http://127.0.0.1:3128
You can also set the URL through the environment:
OPENCLAW_PROXY_URL=http://127.0.0.1:3128 openclaw gateway run
proxy.proxyUrl takes precedence over OPENCLAW_PROXY_URL. A configured URL activates managed proxy routing; removing both URLs disables it.
| Key | Type | Default | Notes |
|---|---|---|---|
proxy.proxyUrl |
string | unset | http:// or https:// forward proxy URL. Credentials embedded in the URL are treated as sensitive and redacted from snapshots/logs. |
proxy.tls.caFile |
string | unset | CA bundle for verifying an https:// proxy endpoint signed by a private CA. |
proxy.loopbackMode |
gateway-only | proxy | block |
gateway-only |
Controls loopback bypass behavior; see below. |
For managed gateway services, store the URL in config so it survives reinstall, rather than relying on foreground env:
openclaw config set proxy.proxyUrl http://127.0.0.1:3128
openclaw gateway install --force
openclaw gateway start
The OPENCLAW_PROXY_URL env fallback is best for foreground runs. To use it with an installed service, put it in the service's durable environment ($OPENCLAW_STATE_DIR/.env, default ~/.openclaw/.env), then reinstall so launchd/systemd/Scheduled Tasks picks it up. This variable is copied into the generated service environment rather than tracked as a managed dotenv key, so systemd's restart-only managed dotenv refresh does not apply.
HTTPS proxy endpoint with a private CA
proxy:
proxyUrl: https://proxy.corp.example:8443
tls:
caFile: /etc/openclaw/proxy-ca.pem
proxy.tls.caFile verifies the proxy endpoint's own TLS certificate. It is not a destination MITM trust setting, a client certificate, or a substitute for the proxy's destination policy. Use NODE_EXTRA_CA_CERTS instead only when the entire Node process must trust an additional CA from startup (for example, an enterprise TLS-inspection system re-signing every HTTPS destination certificate) — that variable is process-global and must be set before Node starts, so OpenClaw cannot apply it mid-run the way it applies proxy.tls.caFile. Prefer proxy.tls.caFile for HTTPS proxy endpoint trust: it is scoped to managed proxy routing instead of the whole process.
openclaw config set proxy.proxyUrl https://proxy.corp.example:8443
openclaw config set proxy.tls.caFile /etc/openclaw/proxy-ca.pem
openclaw gateway run
How routing works
With a valid proxy URL, protected runtime processes (openclaw gateway run, openclaw node run, openclaw agent --local) route normal HTTP and WebSocket egress through the proxy:
OpenClaw process
fetch, node:http, node:https, WebSocket clients -> operator proxy -> destination
Internally, OpenClaw installs Proxyline as the process-level routing runtime. It covers fetch, undici-backed clients, node:http/node:https, common WebSocket clients, and helper-created CONNECT tunnels, and it replaces caller-provided Node HTTP agents so explicit agents (including axios, got, node-fetch, and similar Node-agent-based clients) cannot silently bypass the proxy.
The proxy URL scheme describes the hop from OpenClaw to the proxy, not to the final destination:
http://proxy.example:3128— plain TCP to the proxy; OpenClaw sends HTTP proxy requests, includingCONNECTfor HTTPS destinations.https://proxy.example:8443— OpenClaw opens TLS to the proxy itself (verifying the proxy's certificate), then sends HTTP proxy requests inside that session.
Destination TLS is independent of proxy-endpoint TLS: for an HTTPS destination, OpenClaw always asks the proxy for a CONNECT tunnel and starts destination TLS through that tunnel.
In the default gateway-only mode, loopback HTTP and WebSocket requests connect directly. This includes localhost, literal IPv4/IPv6 loopback addresses, and ephemeral listeners used by the native Codex inference relay. OpenClaw installs this rule in the shared routing owner and replaces no_proxy/NO_PROXY with loopback-only entries for native child processes. External destinations still use the proxy; inherited bypasses for other destinations do not apply. On shutdown, OpenClaw restores the prior proxy environment and resets cached routing state. The explicit proxy and block modes continue to clear no_proxy/NO_PROXY.
Some plugins own a custom transport that needs its own proxy wiring even with process-level routing active. Telegram's Bot API client uses its own HTTP/1 undici dispatcher and separately honors process proxy env plus the OPENCLAW_PROXY_URL fallback.
Gateway loopback mode
Local Gateway control-plane clients normally connect to a loopback WebSocket such as ws://127.0.0.1:18789. proxy.loopbackMode controls whether that traffic bypasses the managed proxy:
proxy:
proxyUrl: http://127.0.0.1:3128
loopbackMode: gateway-only # gateway-only, proxy, or block
A configured proxyUrl or OPENCLAW_PROXY_URL enables managed routing. Set
proxy.enabled: false only as an advanced opt-out that keeps the URL stored
without activating it.
| Mode | Behavior |
|---|---|
gateway-only (default) |
Loopback HTTP and WebSocket traffic connects directly, including Gateway, browser CDP, update readiness, and native Codex inference relay listeners. No endpoint registration is required. |
proxy |
Loopback traffic goes through the proxy. A remote proxy resolves 127.0.0.1/localhost against itself, so this mode can break local runtime listeners. |
block |
OpenClaw denies Gateway control-plane, browser CDP, and guarded local-provider loopback connections before opening a socket. Other requests use the proxy. |
The implicit bypass covers localhost and literal loopback IP URLs. LAN, tailnet, private-network, and public hosts continue through the managed proxy. Application-level SSRF checks still apply to untrusted destinations.
Update canary /startupz and /readyz probes use the same loopback routing policy. If a running canary never answers within the validation budget, the update records the observed failure as a warning, including the next troubleshooting step, and continues best effort.
Environment-only HTTP proxy routing honors no_proxy/NO_PROXY (lowercase takes precedence). These environment bypass lists do not override managed proxy policy.
WebChat or Codex fails after upgrading
The native Codex inference relay added in 2026.9.4 uses an ephemeral local listener. Older managed proxy routing could forward that hop to the external proxy, producing 502 Bad Gateway, Handshake not finished, or stream disconnected before completion for a 127.0.0.1 Responses URL.
Current default routing keeps these hops direct without changing config. Run openclaw doctor to test loopback connectivity under the active proxy. If the diagnostic fails, inspect openclaw config get proxy. For an explicit restrictive loopback mode, restore the default with:
openclaw config set proxy.loopbackMode gateway-only
openclaw gateway restart
For an older installation, the temporary workaround is to set proxy.enabled to false, keep HTTP_PROXY/HTTPS_PROXY pointing to the operator proxy, set NO_PROXY=127.0.0.1,localhost,::1, and restart the Gateway. Persist those environment variables in the Gateway service environment if needed.
Containers
For openclaw --container ... commands, OpenClaw forwards OPENCLAW_PROXY_URL into the container-targeted child CLI when it is set. The URL must be reachable from inside the container — 127.0.0.1 there refers to the container itself, not the host. OpenClaw rejects loopback proxy URLs for container-targeted commands unless you set OPENCLAW_CONTAINER_ALLOW_LOOPBACK_PROXY_URL=1 to explicitly override that check.
Related proxy terms
proxy.enabled/proxy.proxyUrl— outbound forward-proxy routing for runtime egress. This page.gateway.auth.mode: "trusted-proxy"— inbound identity-aware reverse-proxy authentication for Gateway access. See Trusted proxy auth.openclaw proxy— local debug proxy and capture inspector for development and support. See openclaw proxy.tools.web.fetch.useTrustedEnvProxy— opt-in forweb_fetchto let an operator-controlled HTTP(S) env proxy resolve DNS while keeping strict DNS pinning and hostname policy by default. See Web fetch.- Channel- or provider-specific proxy settings — owner-specific overrides for one transport. Prefer the managed network proxy for central egress control across the runtime.
Validating the proxy
The proxy's destination policy is the actual security boundary; OpenClaw cannot verify that your proxy blocks the right targets. Configure it to:
- Bind only to loopback or a private trusted interface, reachable only by the OpenClaw process/host/container/service account.
- Resolve destinations itself and block by IP after DNS resolution, at connect time, for both plain HTTP and HTTPS
CONNECTtunnels. - Reject destination-based bypasses for loopback, private, link-local, metadata, multicast, reserved, and documentation ranges.
- Avoid hostname allowlists unless you fully trust the DNS resolution path.
- Log destination, decision, status, and reason — never request bodies, authorization headers, cookies, or other secrets.
- Keep the policy under version control and review changes as security-sensitive.
Validate from the same host/container/service account that runs OpenClaw:
openclaw proxy validate --proxy-url http://127.0.0.1:3128
With a private-CA HTTPS proxy endpoint:
openclaw proxy validate --proxy-url https://proxy.corp.example:8443 --proxy-ca-file /etc/openclaw/proxy-ca.pem
| Flag | Purpose |
|---|---|
--proxy-url <url> |
Validate this URL instead of resolving config/env. |
--proxy-ca-file <path> |
CA bundle for an HTTPS proxy endpoint. |
--allowed-url <url> |
Destination expected to succeed (repeatable). |
--denied-url <url> |
Destination expected to be blocked (repeatable). |
--apns-reachable |
Also verify the proxy can tunnel a direct sandbox APNs HTTP/2 probe. |
--apns-authority <url> |
Override the APNs authority probed with --apns-reachable. |
--timeout-ms <ms> |
Per-request timeout. |
--json |
Machine-readable output. |
If no config, environment, or --proxy-url value is available, the command reports a config problem; pass --proxy-url for a one-off preflight before changing config.
With no --allowed-url/--denied-url, the default checks are: https://example.com/ must succeed, and a temporary loopback canary server the proxy must not reach must be blocked. The loopback check passes on a transport failure, or on a non-2xx response that lacks the canary's per-run token; it fails on a 2xx response missing the token (an unexpected success from something other than the canary) and, especially, on any response carrying the matching token, since that proves the proxy actually forwarded a loopback destination it should have denied. Custom --denied-url targets have no such canary token, so they are fail-closed: any HTTP response counts as reachable, and a transport error fails the check too, because OpenClaw cannot confirm your proxy denied a reachable origin versus something else going wrong. Only the built-in loopback canary treats a transport error as proof of blocking. See openclaw proxy for the CLI-side statement of the same rule. --apns-reachable sends an intentionally invalid provider token, so a 403 InvalidProviderToken response counts as proof the tunnel reached Apple. The command exits 1 on any validation failure; proxy URL credentials are redacted from both text and JSON output.
{
"ok": true,
"config": {
"enabled": true,
"proxyUrl": "http://127.0.0.1:3128/",
"source": "override",
"errors": []
},
"checks": [
{ "kind": "allowed", "url": "https://example.com/", "ok": true, "status": 200 },
{ "kind": "apns", "url": "https://api.sandbox.push.apple.com", "ok": true, "status": 403 }
]
}
Manual curl check (the public request should succeed; the loopback and metadata requests should be blocked by the proxy itself — curl alone cannot distinguish a proxy denial from an unreachable origin the way openclaw proxy validate's built-in canary can):
curl -x http://127.0.0.1:3128 https://example.com/
curl -x http://127.0.0.1:3128 http://127.0.0.1/
curl -x http://127.0.0.1:3128 http://169.254.169.254/
Recommended blocked destinations
Starting denylist for any forward proxy, firewall, or egress policy. OpenClaw's own SSRF classifier lives in src/infra/net/ssrf.ts and packages/net-policy/src/ip.ts (BLOCKED_HOSTNAMES, BLOCKED_IPV4_SPECIAL_USE_RANGES, BLOCKED_IPV6_SPECIAL_USE_RANGES, the RFC 2544 benchmark prefix, and embedded-IPv4 handling for NAT64/6to4/Teredo/ISATAP/IPv4-mapped forms) — useful references, but OpenClaw does not export or enforce these rules in your external proxy.
| Range or host | Why to block |
|---|---|
127.0.0.0/8, localhost, localhost.localdomain |
IPv4 loopback |
::1/128 |
IPv6 loopback |
0.0.0.0/8, ::/128 |
Unspecified / this-network addresses |
10.0.0.0/8, 172.16.0.0/12, 192.168.0.0/16 |
RFC 1918 private networks |
169.254.0.0/16, fe80::/10 |
Link-local, including common cloud metadata paths |
169.254.169.254, metadata.google.internal |
Cloud metadata services |
100.64.0.0/10 |
Carrier-grade NAT shared address space |
198.18.0.0/15, 2001:2::/48 |
Benchmarking ranges |
192.0.0.0/24, 192.0.2.0/24, 198.51.100.0/24, 203.0.113.0/24, 2001:db8::/32 |
Special-use and documentation ranges |
224.0.0.0/4, ff00::/8 |
Multicast |
240.0.0.0/4 |
Reserved IPv4 |
fc00::/7, fec0::/10 |
IPv6 local/private ranges |
100::/64, 2001:20::/28 |
IPv6 discard and ORCHIDv2 ranges |
64:ff9b::/96, 64:ff9b:1::/48 |
NAT64 prefixes with embedded IPv4 |
2002::/16, 2001::/32 |
6to4 and Teredo with embedded IPv4 |
::/96, ::ffff:0:0/96 |
IPv4-compatible and IPv4-mapped IPv6 |
Add any additional metadata hosts or reserved ranges your cloud provider or network platform documents.
Limits
| Surface | Managed proxy status |
|---|---|
fetch, node:http, node:https, common WebSocket clients |
Routed through managed proxy hooks when configured. |
| APNs direct HTTP/2 | Routed through the APNs managed CONNECT helper. |
| Gateway control-plane loopback | Direct for loopback HTTP/WebSocket destinations in the default mode. |
| Debug proxy upstream forwarding | Disabled while managed proxy mode is active unless explicitly enabled for local diagnostics. |
| IRC | Raw TCP/TLS; not proxied by managed HTTP proxy mode. Set channels.irc.enabled: false if your deployment requires all egress through the forward proxy. |
Other raw net, tls, or http2 client calls |
Must be classified by the raw socket guard before landing. |
- This is process-level coverage for JavaScript HTTP/WebSocket clients, not an OS-level network sandbox.
- Raw
net,tls,http2sockets, native addons, and non-OpenClaw child processes may bypass Node-level routing unless they inherit and respect proxy environment variables. Forked OpenClaw child CLIs inherit the managed proxy URL andproxy.loopbackModestate. - Loopback routing is direct in the default mode; this is not a general local-network bypass. LAN, tailnet, private-network, and public model hosts still use the managed proxy. Guarded local providers retain their configured-origin and DNS checks.
- The local debug proxy's direct upstream forwarding (for proxy requests and
CONNECTtunnels) is disabled by default while managed proxy mode is active; enable it only for approved local diagnostics. - OpenClaw does not inspect, test, or certify your proxy policy. Treat proxy policy changes as security-sensitive operational changes.
Related
- Threat model — adversarial threats to the OpenClaw platform and ClawHub, mapped to MITRE ATLAS
- Security — the trust model, safe defaults, and hardening guidance for running OpenClaw
- Proxy —
openclaw proxy, which validates operator-managed proxy routing and runs the local debug capture proxy