| Symptom | Likely cause | Fix |
|---|---|---|
An error with a [code · req_…] tail in the text | the code in the tail and error.action name the cause and the next step | Do what action says; if that does not help, send the tail to support. Full list: Error codes. |
| Claude Code is "silent" or unreliable | outdated CLI, unapproved API key, or incorrect variables | Update Claude Code, set ANTHROPIC_BASE_URL and ANTHROPIC_AUTH_TOKEN, run regular claude, and approve the key. |
| Asks to log in to Anthropic | the CLI cannot see the API key or lacks one-time approval | Check both variables, restart regular claude, and approve the API key. Use --bare only for minimal diagnostics. |
export doesn't work in PowerShell | that's a Linux command | In PowerShell: $env:ANTHROPIC_AUTH_TOKEN = "gd-…" |
401 on every request | key invalid/revoked | Copy the key from the bot again. |
402 / no funds | the key's credit is exhausted | Check your Guard account and top up the active key; issue a new one only if it was revoked. |
404 model not found | typo in the model id | See Models. |
429 with *_spend_limit | the key reached its hourly/daily spend cap | Change the limit under Manage API key or wait for the rolling window to free up. |
bash
curl -i https://api.guardrelay.ai/v1/models \
-H "Authorization: Bearer $GUARD_API_KEY"If it returns 200 and a list of models — the key and network are fine, look for the problem in the body of the specific request. Still stuck — message support with the [code · req_…] tail from the error text @Guard_help.
Start with the model catalog for your key. It separates authentication problems from generation-body errors.
bash
curl -i https://api.guardrelay.ai/glm/v1/models \
-H "Authorization: Bearer $GUARD_API_KEY"- If the catalog does not return 200, fix the key or check the status page.
- If the catalog works, retry a minimal JSON request with
glm-5.3and a 4096 budget. - For media, switch to
glm-5.3-flashand use only a confirmed payload shape. - For Responses, check
storeand the origin ofprevious_response_id. - For 429, honor
Retry-After; the GLM queue waits up to 10 seconds for a slot.
For support, keep the UTC time, endpoint, model, HTTP status, error.code, and an ID if it appears in the response or your Guard account.