Description

When using deepseek-ai/deepseek-v4-pro as the inference provider via NVIDIA Endpoints, responses through Discord and WeChat messaging channels consistently take 2–4 minutes to arrive. This makes the messaging channel experience unusable for interactive conversation. The latency appears to be end-to-end (from user message sent to bot reply received in the channel).

Environment

Device:        MacBook Pro (Apple Silicon)
OS:            macOS 26.0.1 (Darwin, arm64)
Architecture:  arm64
Node.js:       v22.22.1
npm:           10.9.4
Docker:        29.2.1
OpenShell CLI: 0.0.39
NemoClaw:      v0.0.49
OpenClaw:      2026.4.24 (cbcfdf6)
Model:         deepseek-ai/deepseek-v4-pro
Provider:      NVIDIA Endpoints

Steps to Reproduce

1. nemoclaw onboard with NVIDIA Endpoints provider, model deepseek-ai/deepseek-v4-pro
2. nemoclaw <name> channels add discord (or wechat) — configure channel credentials
3. nemoclaw <name> policy-add discord (or wechat)
4. Send a simple message (e.g. "hello" or "what is 1+1") via Discord DM or WeChat to the bot
5. Wait for response

Expected Result

Bot responds within a reasonable time (under 30 seconds for simple queries).

Actual Result

Bot response takes 2–4 minutes on average for both Discord and WeChat channels. Observed consistently across multiple messages and both channel types, not a one-off spike. Simple prompts like "hello" exhibit the same latency.

Logs

Not captured — latency observed from user-facing Discord and WeChat clients.

NVB#6205544