Frequently asked questions
Eight answers before you need to ask.
What is metaproxy?
A model aggregation gateway: one API key and one OpenAI-compatible interface for available models from multiple providers. To switch models, change only the model field in your request.
How does it relate to official provider APIs?
Your request is routed through the gateway to the corresponding provider. We unify integration, billing, and invoices; the model capability and output still come from that provider. metaproxy is an independent third-party gateway and is not affiliated with, authorized by, or endorsed by OpenAI, Anthropic, Google, or other providers.
Why can prices be lower than official prices?
Aggregated purchasing and comparison across multiple upstreams. We claim a discount only when a model has a public USD provider price and our price is genuinely lower. Every row on the pricing page links to the provider source so you can verify it yourself.
How does billing work?
Each ¥1 adds $1 of account balance. Models continue to consume that dollar-denominated balance by token, with separate input and output rates per million tokens. All keys share one balance, and every charge can be reconciled against the usage log.
Which models are supported?
Current models and providers are listed on the live pricing page. The pricing page has the complete list and live status, all sourced from the same product catalog.
Which SDKs are compatible?
Any OpenAI-compatible client: the official Python and Node SDKs, LangChain, curl, and applications that let you set a custom base_url. The integration page shows the exact change for each client.
Do you store my conversations?
Requests are forwarded to the selected provider. Billing logs retain only metadata such as model, token counts, and cost; your prompt and response bodies are not stored.
What should I do if a key leaks?
Disable and replace it immediately in the console. The full key is shown only once when created; everywhere else it is displayed in masked form.
How do I switch back to my own subscription afterwards?
Run curl -fsSL https://metaproxy.agency/uninstall.sh | bash. The installer writes a marked block appended after your existing settings, and never deletes or edits your own lines — removing that block puts them back in effect. Your current terminal still holds the exported values afterwards, so open a new shell or run the unset line the uninstaller prints.
Can it run alongside my own subscription?
Yes — pass --isolated when installing. It writes nothing to your shell config and instead installs a claudemp command (the name is configurable): claude keeps using your own subscription, claudemp goes through metaproxy. Claude Code only for now; codex and opencode read config files rather than environment variables, so isolating them needs a separate config tree that does not exist yet.
Can I use other providers’ models inside Claude Code?
Yes. The gateway also exposes an Anthropic-compatible interface and routes from it to other providers, so claude --model accepts any model id in the catalog — Gemini, Kimi, GPT included. Note that the model is per session: one session uses one model at a time, so running several at once means running several processes. Also, reasoning models such as Kimi K3 spend tokens thinking first; too small a max_tokens can hit the limit before any visible output and return empty content. That is the budget, not the model.