Troubleshooting

Start with the simplest test. If this works, your key and balance are fine, and the problem is in the tool’s config:

curl https://api.vikey.ai/v1/chat/completions \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "openai/gpt-6-luna", "max_tokens": 16, "messages": [{"role": "user", "content": "ping"}]}'

Common problems

Symptom	Most likely cause	Fix
`404` on every request	Wrong base URL. The OpenAI SDK needs `/v1`, the Anthropic SDK and Claude Code do not.	See Base URL
`401 Invalid API key`	Wrong key, a stray space, or a key from another account	Copy it again from API Keys
`401` although the key is right	The tool sends the key in the wrong header, or the environment variable is not read	Check your tool’s guide under Coding agents
`402 Insufficient balance`	No balance left, or `max_tokens` too large for what is left	Top up, or lower `max_tokens`
`403 API key is inactive`	The key is turned off	Turn it on again in API Keys
`Model '...' not found`	A typo in the model id, or the tool sends its built-in model name	Use an id from the pricing page. For Claude Code, add `modelOverrides`.
`ambiguous_model`	A short name matches several providers	Use the full id, such as `deepseek/deepseek-v4.1-flash`
Codex: `wire_api = "chat" is no longer supported`	An old Codex config	Change it to `wire_api = "responses"`, see Codex
opencode: `Model not found: vikey/...`	The unsupported `providers` format, or the model is missing from `models`	Use the format in opencode
Claude Code asks you to log in	The key is in project settings, not user settings or the shell	See Claude Code
Cost higher than expected	Long prompts resent without cache, or an expensive model used for light tasks	Check Usage per model and consider Vikey Auto
The answer stops halfway	The `max_tokens` limit was reached	Raise `max_tokens`
`502` or `503`	The model has a problem on the provider side	Retry shortly, or use another model

Still stuck?

The request_id from the error message, if any
The tool and its version, for example Claude Code 2.1.323
The model id you used
When it happened

Never send your API key to anyone, including support.