How it works
Model ids, request formats, and how usage is paid.
The few ideas behind AivoKey, in plain words.
Model id
Every model has one id, shown on the Models page — for example deepseek-v4-pro or glm-5.3. Put that id in the model field of your request or in your tool's model setting. Case does not matter.
Three request formats
Different tools speak different "dialects". AivoKey understands all three, so you never have to convert anything:
| Format | Endpoint | Used by |
|---|---|---|
| OpenAI Chat Completions | POST /v1/chat/completions | Most tools and SDKs, OpenCode, Cline, Roo, Kilo |
| OpenAI Responses | POST /v1/responses | Codex CLI, newer OpenAI SDK code |
| Anthropic Messages | POST /v1/messages | Claude Code, Anthropic SDKs |
Any model works with any format. You can use a GPT model from Claude Code, or a Claude model from Codex.
How your usage is paid
Each request is paid from the first source that can cover it, in this order:
- Bonus tokens — from promotions, for a specific model.
- Free tokens — claimed on the faucet, for a specific model.
- Plan allowance — if you have a plan that includes the model.
- Balance — your prepaid credit. With a plan, the balance is only used by keys that have balance fallback turned on.
Prices are per 1 million tokens and listed on the Models page.
Prompt caching
Coding agents resend the same long context on every turn. AivoKey keeps a conversation on the same route so that repeated context can be served from cache, which is cheaper and faster. Claude Code gets this automatically; for other tools see Prompt caching.
Reliability
If a model is temporarily unavailable on one route, AivoKey retries on another one for you. You only see an error when every option has failed.