LiteLLM on one prepaid key

• Python 3.10 or newer with LiteLLM installed. The SDK and the proxy are separate extras:
  bashCopy

  1pip install litellm # the SDK

  2pip install 'litellm[proxy]' # the proxy server
• A RelaySwitch key. Sign in with an email code at the console and create one. Keys start with sk-relay-. Export it as RELAYSWITCH_API_KEY; both code blocks below read it from there.
• A balance. Top-ups start at $10 by Card or in USDT, USDC, BTC, ETH, SOL. Pay as you go, no subscription, no KYC. The console chat gives you 12 free turns a day before the first top-up; LiteLLM requests bill the balance.

The SDK needs three things per call: the model id behind openai/, the base URL and the key. LiteLLM's docs say the openai/ prefix routes to /chat/completions on whatever api_base you give it, and that api_base should end in /v1.

Claude ids answer the OpenAI format here, so openai/claude-sonnet-5-5 is correct even though the model is Anthropic's. There is no anthropic/ prefix to set and no second key. To change models, change the id after the prefix; the key and base stay.

The proxy maps a client-facing model_name to a litellm_params block with the real target. The os.environ/RELAYSWITCH_API_KEY value tells LiteLLM to read the key from the environment at start, so the file holds no secret. drop_params under litellm_settings makes the proxy drop a parameter a model rejects instead of failing the request (drop_params docs, read 2026-10-28).

Export the key and start the proxy:

Then any OpenAI-compatible client talks to the proxy. The quick start (read 2026-10-28) uses a placeholder api_key on the client; the proxy adds the real key from the config.

If the proxy is reachable by anyone but you, set general_settings.master_key (it must start with sk-) and clients send that as their key. See LiteLLM's virtual keys page (read 2026-10-28); the full config reference is on the proxy config page.

Any id on the Models page works after openai/. The id is the slug in the model page URL, not the display name. The 6 in the config above cover the usual spread:

GET https://api.relayswitch.io/v1/models with your key returns every id it can call, which is the list to generate a longer model_list from.

LiteLLM's own cost tracking multiplies tokens by the developer list prices in its model map. The charge here is the RelaySwitch rate, which the console's Usage page shows per request. These are the live rates for the models in the config:

One chat turn of 2,000 input tokens and 500 output tokens costs $0.0012 on GLM 5.3 and $0.0106 on Claude Opus 5.5 at those rates. How the balance and the cached rate work is on Pricing.

Two curl calls separate a key or base URL problem from a LiteLLM one. The first lists the ids your key can call:

The second sends one request the way LiteLLM does. -i prints the headers, including the request id and, on a non-streamed answer, the charge in USD.

If both work, what is left is in the LiteLLM call or config. Send one request through LiteLLM and it shows on the console's Usage page within seconds, with its tokens and its charge.

• LLM Provider NOT provided: the openai/ prefix is missing from the model string.
• NotFoundError with no model code: api_base is missing /v1 or carries an extra path. LiteLLM adds /chat/completions itself.
• 401: the key is wrong, revoked or cut short, or RELAYSWITCH_API_KEY was not exported in the shell that started the proxy. Paste it again and run the curl check.
• 402: the balance is too low for the request. Top up in the console; the key works again at once.
• 404 with a model code: the id after openai/ is not in the catalog. Check for a typo or a display name in place of the id.
• 429: the key hit its rate limit. Wait and resend, or create a key with a higher Requests / min in the console.
• A parameter error on one model only: set drop_params: true as in the config above, or pass drop_params=True to that completion() call.