Quickstart

Install warpllm, export one API key, and make your first routed request.

1. Install the package

   Python

   pip install warpllm
   

   Requirements: Python 3.10+, Node 20+, or Rust 1.85+. See Installation for platform details.

2. Export a provider key

   warpllm reads keys from the environment at request time. Export the one belonging to the provider you are about to route to.

   export OPENAI_API_KEY=sk-...
   

   Every provider and its variable is listed in Authentication.

3. Send a chat completion

   The model string is provider/model. The provider/ prefix is required.

   Python

   from warpllm import WarpLLM
   
   client = WarpLLM()
   
   completion = client.chat_completions({
       "model": "openai/gpt-5-nano",
       "messages": [{"role": "user", "content": "Hello!"}],
   })
   
   print(completion["choices"][0]["message"]["content"])
   

   warpllm ships no async runtime, so a Rust caller brings one: cargo add tokio --features macros,rt-multi-thread.

4. Switch providers by changing the string

   Nothing else in the call moves. Export the new provider’s key and change the model:

   - "model": "openai/gpt-5-nano",
   + "model": "deepseek/deepseek-v4-flash",
   

   export DEEPSEEK_API_KEY=sk-...
   

Runnable examples

The repository carries a commented, runnable version of each snippet above in examples/, along with an .env.example template for the provider keys.

Next steps

Stream a reply

Set stream: true and iterate chunks in any of the three languages.

Handle failures

One error taxonomy across every provider, in OpenAI’s vocabulary.

Browse the model roster

Every routable model, its context window, and its output ceiling.

Run the gateway

Keep your existing OpenAI client and change only its base URL.