openclaw - 💡(How to fix) Fix [Bug]: Agent ignores all proxy settings, cannot connect to Telegram behind DPI [1 comments, 2 participants]

Bug type

Regression (worked before, now fails)

Beta release blocker

No

Summary

OpenClaw completely ignores all proxy settings (HTTP_PROXY, HTTPS_PROXY, network.proxy). Traffic to Telegram API and cloud providers fails with timeout behind DPI, despite a working Tailscale SOCKS5 proxy confirmed with curl.

Steps to reproduce

1. Run OpenClaw 2026.4.26 in Docker on a network where direct access to api.telegram.org is blocked by DPI.
2. Set env vars: HTTP_PROXY=http://172.20.0.2:1080, HTTPS_PROXY=http://172.20.0.2:1080, ALL_PROXY=http://172.20.0.2:1080.
3. Start the gateway with a Telegram bot token configured.
4. Check logs: Telegram channel reports UND_ERR_CONNECT_TIMEOUT and deleteWebhook failed.
5. From the same container, run curl -x socks5h://172.20.0.2:1080 ifconfig.me – it returns the VPS IP, proving the proxy works.

Expected behavior

OpenClaw should route its outbound HTTP/HTTPS requests through the configured proxy, as standard Node.js applications do when HTTP_PROXY/HTTPS_PROXY are set. The Telegram channel should connect successfully.

Actual behavior

Telegram channel fails to connect. Logs show UND_ERR_CONNECT_TIMEOUT and Network request for 'deleteWebhook' failed!. The gateway never uses the proxy, even though it is reachable.

OpenClaw version

2026.4.26 (also tested on 2026.4.14-beta.1, 2026.4.14)

Operating system

Synology DSM 7.3.2 (Linux 5.10.55+), Docker

Install method

Docker (official image ghcr.io/openclaw/openclaw:latest and custom Debian-based image)

Model

Used for testing: ollama/qwen2.5-coder:7b (local), but the issue is network-level, not model-specific.

Provider / routing chain

openclaw -> (should go through proxy) -> Tailscale SOCKS5 proxy (172.20.0.2:1080) -> VPS Exit Node -> internet. In reality, openclaw attempts direct connection, bypassing proxy.

Additional provider/model setup details

Ollama base URL: http://172.20.0.4:11434 (local network, excluded from proxy via NO_PROXY). Gemini API key is set but cannot be reached due to proxy bypass.

Logs, screenshots, and evidence

[telegram] fetch fallback: enabling sticky IPv4-only dispatcher (codes=UND_ERR_CONNECT_TIMEOUT)
[telegram] deleteWebhook failed: Network request for 'deleteWebhook' failed!
[telegram] webhook cleanup failed: Network request for 'deleteWebhook' failed!
# Proof that proxy works:
$ docker exec -it magura-oc curl -x socks5h://172.20.0.2:1080 ifconfig.me
92.246.139.79  # VPS IP

Impact and severity

Affected: all Telegram users of the bot behind DPI. Severity: High (complete inability to send/receive messages). Frequency: Always when DPI is active. Consequence: Bot is unusable unless network restrictions are lifted.

Additional information

The same Docker image works perfectly on a VPS without DPI, confirming the issue is solely the missing proxy support. Temporary workaround: none found; transparent proxying via redsocks/iptables is blocked by Docker network isolation.