Troubleshooting

Fix common Floodgate agent and API configuration problems.

401 Unauthorized

• Confirm the API key is active in your Floodgate dashboard.
• Make sure the key has no extra spaces or quotation marks.
• Use Authorization: Bearer YOUR_API_KEY.
• Open a new terminal if you configured environment variables in shell profile files (~/.bashrc, ~/.zshrc).

404 Not Found

Check how your client handles the base URL path:

• Codex, OpenCode, Pi, Grok Build, Aider, and OpenAI SDKs use https://api.floodgate.dev/v1.
• Claude Code uses https://api.floodgate.dev because it appends /v1/messages automatically.

Model not found

Use the exact model ID:

gpt-6-sol
claude-opus-5-5
deepseek-v4.1-flash
grok-4.7

Confirm which models are enabled for your key:

curl "https://api.floodgate.dev/v1/models" \
  -H "Authorization: Bearer $FLOODGATE_API_KEY"

The CLI still uses its original provider

• Codex: run /status and confirm model_provider is floodgate.
• Claude Code: run /status and confirm the Anthropic base URL points to https://api.floodgate.dev.
• OpenCode: run /models and select a model under Floodgate.
• Pi: run /model and select a model under floodgate.
• Grok Build: run grok inspect and check ~/.grok/config.toml.
• Aider: confirm both OPENAI_API_BASE and OPENAI_API_KEY are exported.

Tool calls fail

Try the recommended frontier model for your agent harness:

Harness	Recommended model
Codex	gpt-6-sol
Claude Code	claude-opus-5-5
Grok Build	grok-4.7
OpenCode, Pi, Aider	gpt-6-sol or claude-opus-5-5

Different models can vary in tool-calling schema strictness and reasoning behavior.