FAQ
Only cross-app, general questions live here. App-specific issues belong in each app integration page.
Keys and authentication
My key leaked — what now
Immediately disable or reset it in the console. Resetting immediately invalidates the old secret. Then inspect the request log for suspicious traffic and update every service that used the key. API keys cannot be deleted.
How many keys can I create
There is no per-account cap. Use a separate key per environment (app, stage, team member) so you can attribute usage and revoke surgically.
What’s the difference between several keys in the same group
The key itself is just a credential. Usage stats are tracked per key; access scope and pricing multipliers come from the group it belongs to.
Models and protocols
Should I call Claude through OpenAI or Anthropic protocol
Use the Anthropic protocol (https://api.tokenbay.com/v1/messages) whenever your client natively supports it — Claude Code, Cline, Roo Code, official Anthropic SDKs. That path preserves prompt caching, tool-result blocks, and other Anthropic-specific fields.
If your client only speaks OpenAI, /v1/chat/completions also routes to Claude, but some advanced features will be dropped.
The console lists a model but my call returns 404
The public return scope for /v1/models is still being finalized. For now, use the live Models page and console-visible models as the source of truth. If a tool cannot auto-fetch models, manually enter the model ID and check your API key, balance, and quota rules.
Upstream raised its prices — does TokenBay follow
Yes. TokenBay’s multipliers track upstream cost, with prior notice. Existing in-flight requests bill at the rate when the request started; new requests use the new rate.
Network and proxies
Slow connection from mainland China
See status for current node coverage. A local reverse proxy is a reasonable short-term fix. Do not route DNS through untrusted nodes — they can intercept your API key.
My MITM proxy (e.g. mitmproxy) is failing TLS
Some SDKs refuse self-signed certs. Add the proxy’s root CA to your OS trust store, or configure the SDK to trust it explicitly.
Billing
Do failed requests get charged
See Monitoring and Troubleshooting → How failed requests are billed. Gateway-level rejections and upstream errors are not charged. A stream that disconnects midway is billed for whatever tokens were already emitted.
My actual spend doesn’t match what I expected
Use the request log for per-call detail. Usual culprits:
- The client doesn’t read
usagefrom the streamed response and undercounts - An autocomplete-style extension (Continue autocomplete, Copilot-like UX) fires on every keystroke — call volume dwarfs chat sessions
- Prompt caching didn’t kick in (window expired, model unsupported), so the cache multiplier degrades to the input rate
Docs and feedback
My tool isn’t in your docs
If it accepts a custom OpenAI- or Anthropic-compatible endpoint, it’ll work. Point it at https://api.tokenbay.com/v1 (OpenAI) or https://api.tokenbay.com (Anthropic), and fill in your sk-... key. Open a request if you’d like an integration page added.
How do I report a bug or suggest something
File a ticket from the console, or use the feedback link on any docs page. Security issues should go through the dedicated channel — do not paste API keys or raw logs in public issues.