Troubleshooting
Fixes for the most common setup problems.
Authentication failures (401)
- Make sure the whole key was copied. Keys are long and easy to truncate.
- Export the key in the same shell session that runs your tool, or add it to your shell profile. Check with
echo $EVOLVED_API_KEY. - In Claude Code, an older
ANTHROPIC_API_KEYfrom another provider can take precedence. Unset it, then run/statusto confirm which base URL and credentials are active. - If the key was revoked or replaced, use the newest key we sent you, or message support for a new one.
Model not found (404)
Model IDs must match exactly: use claude-opus-4.8, not claude-opus-4-8 or Claude Opus 4.8. List the IDs your key can use:
curl https://api.evolved.to/v1/models \
-H "Authorization: Bearer $EVOLVED_API_KEY"Also check that the endpoint serves that family: GPT models are not available on /v1/messages. See protocol support.
Wrong base URL
A 404 on every request usually means the base URL has a missing or duplicated /v1.
| Client | Correct base URL |
|---|---|
| Claude Code, Anthropic SDKs | https://api.evolved.to |
| Codex, Cursor, OpenAI SDKs, Cline (OpenAI Compatible), Roo Code, n8n | https://api.evolved.to/v1 |
Streaming stalls
- Corporate proxies and some load balancers buffer server-sent events. Disable response buffering for the gateway host.
- Frontier models can think for a while before the first token. Raise client timeouts rather than retrying early.
- Prefer streaming for long outputs. It keeps the connection alive and shows progress.
Nothing here helped? Contact support with your request ID.