Pre-submission checklist | 提交前检查
Problem Statement | 问题陈述
When running MemOS as a plugin within OpenClaw deployed in WSL2 (Windows 11 + Ubuntu), two related issues are observed:
-
VPN Required for API Model Connections: The MemOS plugin cannot reach any upstream LLM API endpoint (e.g., OpenAI, DeepSeek, Anthropic) without a VPN or proxy connection active on the Windows host. This suggests the plugin either does not respect WSL2's proxy environment variables (HTTP_PROXY/HTTPS_PROXY) or does not handle WSL2's network bridge correctly.
-
Extremely Slow Streaming Response Under VPN: When the VPN is active and connections succeed, streaming log/token responses are noticeably laggy — there is a significant delay between each streamed chunk, making real-time interaction nearly unusable. This does not occur with the same model on the same network when used outside of WSL2 (e.g., native browser or a Windows-hosted client).
Combined, these symptoms suggest poor WSL2 network stack compatibility — likely related to how MemOS (or its plugin layer) handles:
- System proxy / environment variable discovery in WSL2
- WebSocket/SSE streaming over WSL2's virtualized network adapter
- DNS resolution or HTTP keep-alive under proxied connections
Expected Behavior | 预期行为
- MemOS should work out-of-the-box in WSL2 without requiring manual VPN configuration
- API model connections should respect system proxy settings (HTTP_PROXY, HTTPS_PROXY, NO_PROXY) or offer explicit proxy configuration
- Streaming responses under a proxy/VPN should not have excessive latency compared to native environments
Additional Context | 其他信息
Environment:
- Host OS: Windows 11
- WSL2 Distro: Ubuntu (WSL2)
- MemOS: Running as OpenClaw plugin (memos-local-plugin)
- Network Requirement: VPN/proxy mandatory for model API access (region-dependent)
- Symptom: Without VPN -> connection refused / timeout; With VPN -> streaming lag
The existing issues around network stability in proxied environments and WSL2 compatibility would also benefit from proper documentation of proxy configuration in the MemOS docs.
Willingness to Implement | 实现意愿
Pre-submission checklist | 提交前检查
Problem Statement | 问题陈述
When running MemOS as a plugin within OpenClaw deployed in WSL2 (Windows 11 + Ubuntu), two related issues are observed:
VPN Required for API Model Connections: The MemOS plugin cannot reach any upstream LLM API endpoint (e.g., OpenAI, DeepSeek, Anthropic) without a VPN or proxy connection active on the Windows host. This suggests the plugin either does not respect WSL2's proxy environment variables (HTTP_PROXY/HTTPS_PROXY) or does not handle WSL2's network bridge correctly.
Extremely Slow Streaming Response Under VPN: When the VPN is active and connections succeed, streaming log/token responses are noticeably laggy — there is a significant delay between each streamed chunk, making real-time interaction nearly unusable. This does not occur with the same model on the same network when used outside of WSL2 (e.g., native browser or a Windows-hosted client).
Combined, these symptoms suggest poor WSL2 network stack compatibility — likely related to how MemOS (or its plugin layer) handles:
Expected Behavior | 预期行为
Additional Context | 其他信息
Environment:
The existing issues around network stability in proxied environments and WSL2 compatibility would also benefit from proper documentation of proxy configuration in the MemOS docs.
Willingness to Implement | 实现意愿