Make your first request¶

Make a model call, read its response, and display text as it arrives from a second request. You need Python 3.11 or later, an OpenAI API key, and access to the model you choose.

Install Republic¶

In a virtual environment, install Republic:

python -m pip install republic

Supply your key¶

In your shell, set:

export REPUBLIC_OPENAI_API_KEY="your-api-key"

Republic reads this variable when it creates the provider. You can also pass api_key= to get_model() if your application already loads credentials.

Send a message and stream a response¶

Save this as hello.py. Replace gpt-6-sol if your account uses a different model.

import asyncio

import republic
from republic.events import TextDelta


async def main():
    model = republic.get_model("openai:gpt-6-sol")

    response = await model.chat("Explain a tool call in one sentence.")
    print(response.text)
    print("Tokens:", response.token_usage.total_tokens)

    async with model.stream("Explain how an agent uses a tool result.") as stream:
        async for event in stream:
            if isinstance(event, TextDelta):
                print(event.chunk, end="", flush=True)
        print()
        print("Tokens:", stream.response.token_usage.total_tokens)


asyncio.run(main())

Run it:

python hello.py

The first answer appears after chat() finishes. The second arrives in text chunks, followed by its token count. The wording and counts will vary. These are two independent requests; a model does not retain earlier messages unless you supply conversation history.

The async with block closes the stream. Consume the iterator before reading stream.response; reading it early raises errors.StreamNotFinishedError. The final response has the same shape as the response from chat().

Try another provider¶

With a Google API key in REPUBLIC_GOOGLE_API_KEY, replace the model construction line in hello.py:

model = republic.get_model("google:gemini-flash-latest")

Run the script again. The request methods, text events, and usage fields are the same. Each service still determines which models and features your account can use; see the provider directory.

The provider directory offers API-key setup and account login paths. The authentication guide explains explicit authorization, credential storage, and renewal.

To build on a model call, write a minimal agent that reads local project files and returns tool results. For a typed response, see structured output.