Bug type

Regression (worked before, now fails)

Beta release blocker

No

Summary

Setting http_proxy environment variables to bypass a local Fake IP SSRF block causes OpenClaw's CLI and WebUI to silently hang and fail without returning any errors when querying Gemini models.

Steps to reproduce

1. Set up a Linux environment (e.g., Ubuntu) running a transparent proxy (like v2rayA in tproxy mode) with "Fake IP" enabled for DNS resolution.

2.Configure OpenClaw (v2026.4.21) with a valid Gemini API key.

3.Attempt to call the Gemini model without explicitly setting proxy variables:
openclaw infer "Test" --model google/gemini-3-flash-preview

4.Check the Gateway logs and observe the internal SSRF security block:
blocked URL fetch... reason=Blocked: resolves to private/internal/special-use IP address

5.To bypass the Fake IP resolution, explicitly inject the local HTTP proxy variables into the terminal:
export http_proxy="http://127.0.0.1:20171"
export https_proxy="http://127.0.0.1:20171"

6.Run the inference command again in the same terminal:
openclaw infer "Test" --model google/gemini-3-flash-preview

7.Observe the bug: The CLI (and similarly the WebUI) hangs, generates no response, and silently fails without throwing any HTTP errors or timeout warnings to the user interface.

Expected behavior

1.When http_proxy and https_proxy environment variables are provided, the OpenClaw Gateway should respect them and route the outgoing API request through the specified proxy. This should bypass the local Fake IP DNS resolution and successfully complete the inference call to the Gemini model.

2.If the request fails for any reason (e.g., proxy connection refused, timeout, or an upstream API error), OpenClaw should not fail silently. It must surface an explicit, descriptive error message to the CLI and WebUI to inform the user of the exact failure.

Actual behavior

After setting the http_proxy and https_proxy environment variables and initiating the request, the CLI (openclaw infer or openclaw tui) and the WebUI fail silently.

There is absolutely no response generated from the Gemini model, and the interface simply hangs or terminates without surfacing any explicit error messages (such as network timeouts, HTTP 400/500 codes, or connection refused warnings). In the background logs, the agent appears to quit quietly (sometimes leaving a brief warn agent/embedded embedded run agent end trace), but the user is left completely unaware of the underlying failure.

OpenClaw version

2026.4.21

Operating system

Ubuntu 24.0.4 / Windows11 25H2

Install method

No response

Model

google/gemini-3.1-flash-lite-preview

Provider / routing chain

Provider: Google Model: google/gemini-3-flash-preview (also tested with google/gemini-3.1-flash-lite-preview) Routing Chain: OpenClaw Gateway -> Local HTTP/HTTPS Proxy (http://127.0.0.1:20171 provided by v2rayA) -> Official Google API endpoint (generativelanguage.googleapis.com)

Additional provider/model setup details

• Auth Profile: API key is managed via auth-profiles.json. openclaw models status correctly reflects the Google provider as effective with api_key=1.
• Environment Context: Environment variables (http_proxy, https_proxy) are injected via terminal export and verified within the shell before running openclaw infer.
• Config Context: Using default agent settings under ~/.openclaw/agents/main/agent. No custom per-agent routing overrides are currently applied.
• Network Stack: The system uses v2rayA for transparent proxying, but OpenClaw is specifically directed to the local HTTP port (127.0.0.1:20171) to avoid the SSRF block triggered by Fake IP resolution on the Gateway.

Logs, screenshots, and evidence

Impact and severity

• Affected users/systems/channels: Users running OpenClaw on Linux environments with transparent proxies (specifically those utilizing Fake IP/tproxy setups). Affects both CLI and WebUI channels.
• Severity: Blocks workflow. The core functionality of the agent is completely inaccessible because the request cycle cannot be completed.
• Frequency: Always. The issue occurs consistently every time the Gemini API is called while the HTTP proxy environment variables are active.
• Consequence: Total service failure for the affected provider. Users cannot send or receive messages from the AI, rendering the OpenClaw deployment non-functional for its primary purpose.

Additional information

No response