solution model high macos linux windows

Moonshot (Kimi): API works in browser but OpenClaw keeps failing (wrong endpoint)

Fix Moonshot/Kimi model failures by using the correct baseUrl for your region (api.moonshot.ai vs api.moonshot.cn), and ensuring MOONSHOT_API_KEY is set on the gateway host.

By CoClaw Team • • Updated February 13, 2026

Symptoms

  • Moonshot/Kimi models time out, fail intermittently, or never return (while direct API calls seem fine).
  • openclaw models status --probe fails for moonshot/....

Cause

Most often:

  • The gateway is using the wrong Moonshot endpoint for your network/region, or
  • MOONSHOT_API_KEY is set on your laptop but not on the gateway host (VPS/Docker).

Moonshot supports multiple endpoints:

  • International: https://api.moonshot.ai/v1
  • China: https://api.moonshot.cn/v1

If your gateway host can only reach one of them reliably, the other will fail.

Fix

1) Ensure the key exists on the gateway host

Run on the gateway host:

openclaw models status

If Moonshot shows as unauthenticated/missing, set MOONSHOT_API_KEY on the gateway host and restart the gateway.

2) Set the correct baseUrl for your region

Use openclaw configure (Model/auth → Moonshot) or set it in config:

{
  models: {
    providers: {
      moonshot: {
        baseUrl: "https://api.moonshot.cn/v1",
      },
    },
  },
}

Then restart the gateway.

3) Probe again

openclaw models status --probe

Verify

  • Probes succeed for moonshot/....
  • A test chat returns normally and logs show successful Moonshot requests.

Verification & references

  • Reviewed by:CoClaw Code Team
  • Last reviewed:March 14, 2026
  • Verified on: macOS · Linux · Windows
Want to explore more? Browse all solutions or ask in the Community Forum .
Report a problem

Related Resources

Venice AI: models unavailable or requests make no API calls
Fix
Fix Venice provider issues by checking VENICE_API_KEY, network reachability to api.venice.ai, model refs, and credits/billing.
Web search: missing_brave_api_key even though it's configured
Fix
If `web_search` fails with `missing_brave_api_key` but you set the key via `openclaw configure --section web`, make sure the running gateway service can see the secret (often via `~/.openclaw/.env`) and restart the gateway.
API works in curl, but OpenClaw still fails
Fix
Fix custom or local AI API integrations where direct curl requests succeed, but OpenClaw still errors, returns blank output, or fails during real agent runs.
Telegram: 'setMyCommands failed' / Bot API requests fail
Fix
Fix Telegram Bot API failures caused by outbound HTTPS/DNS/IPv6 issues to api.telegram.org (common on restricted VPS or proxies).
OpenClaw Relay & API Proxy Troubleshooting (NewAPI/OneAPI/AnyRouter): Fix 403s, 404s, and Empty Replies
Guide
A practical integration guide for using OpenClaw with OpenAI/Anthropic-compatible relays and API proxies (NewAPI, OneAPI, AnyRouter, LiteLLM, vLLM): choose the right API mode, set baseUrl correctly, avoid config precedence traps, and debug 403/404/blank-output failures fast.
OpenClaw Not Responding: Fix 'no output', Incorrect API, Rate Limits, and Silent Failures
Guide
A high-signal checklist for when OpenClaw stops replying (TUI shows '(no output)', channels go quiet, or logs show 401/403/429). Covers config precedence, provider auth, model allowlists, relay API-mode mismatch, and rate-limit/billing traps.