Adapted community discussion: Paraphrased and condensed from an anonymized public OpenClaw Discord support conversation from August 2026. Technical guidance was reviewed for OpenClaw 2026.9.3.
Every agent request fails before the reply with an HTTP 401 and no response body. I am using an NVIDIA model on Windows.
A 401 usually means the provider rejected the active credential, but first confirm which provider and Gateway process handled the request:
openclaw --version
openclaw status --all
openclaw gateway status --deep
openclaw models status
openclaw logs --follow
Redact keys, tokens, headers, and private prompt content.
The process using the Gateway port is node.exe running OpenClaw. Task Scheduler says its copy is stopped, but another manually started Gateway owns the port.
Keep the one healthy Gateway and do not start a second copy. Once its connectivity probe is healthy, test only NVIDIA authentication:
openclaw models list --provider nvidia
openclaw models status --probe --probe-provider nvidia
A probe makes a real provider request. Auth: yes means a credential source exists; only the live probe establishes whether NVIDIA accepts it.
The live probe identifies the stored nvidia:default profile and returns an authentication failure.
Replace that exact stored profile through OpenClaw’s protected prompt, then probe it again:
openclaw models auth --agent main paste-api-key \
--provider nvidia \
--profile-id nvidia:default
openclaw models status --probe --probe-provider nvidia
Enter the complete key only at the terminal prompt. Never paste it into a forum, log, or command argument. If the new key is still rejected, create a fresh provider key and verify its account access.
The replacement profile now passes the live probe with HTTP 200.
That confirms the original 401 was the rejected NVIDIA credential. Retry the agent request. If the existing Gateway cached the old credential, restart that single Gateway once, then test again.