punkreq

punkreq is an async HTTP client for Python, built on top of httpunk.

Warning: punkreq is in an early stage and still work in progress.

Note: punkreq was built with substantial help from LLMs, under human supervision.

In a nutshell

• Async only, with multiple runtime backends: import the client from the module matching your runtime – punkreq.asyncio or punkreq.tonio
• HTTP/1.1 and HTTP/2, negotiated via TLS ALPN by default; HTTP/2 prior knowledge with http1=False
• Connection pooling with keep-alive (h1) and multiplexing (h2)
• Transparent response decompression: gzip and deflate everywhere, zstd on Python 3.14+, brotli with the punkreq[brotli] extra
• Conservative automatic retries — only requests the server provably never processed (HTTP/2 graceful GOAWAY, refused streams, keep-alive races)
• Opt-in cookie jar, basic/bearer auth, multipart uploads
• HTTP proxies: HTTP_PROXY/HTTPS_PROXY/NO_PROXY environment support, CONNECT tunneling for https destinations.
• Zero runtime dependencies beyond httpunk

Installation

pip install punkreq

Quickstart

import asyncio
from punkreq.asyncio import Client

async def main():
    async with Client(base_url="https://api.example.com") as client:
        response = await client.post("/items", json={"name": "widget"})
        response.raise_for_status()
        print(await response.json())

asyncio.run(main())

One-off requests without managing a client:

from punkreq import asyncio as punkreq

response = await punkreq.get("https://www.example.com")
print(response.status_code, response.headers["content-type"])

Responses resolve as soon as the head arrives; the body is read on demand — await response.read() / text() / json(), or streamed:

async with Client() as client:
    response = await client.get("https://example.com/large.bin")
    async for chunk in response.iter_bytes(chunk_size=65536):
        ...

Reading to the end releases the connection automatically. When you might stop early, use the request as a context manager so the response is always closed:

async with client.get("https://example.com/huge.ndjson") as response:
    async for line in response.iter_lines():
        if found(line):
            break  # the block exit closes the response and frees the connection

Configuration

import punkreq
from punkreq.asyncio import Client

client = Client(
    base_url="https://api.example.com",
    headers={"x-api-key": "..."},
    auth=("user", "pass"),                       # or punkreq.BearerAuth("token")
    cookies={},                                  # enables the cookie jar (off by default)
    timeout=punkreq.Timeout(5.0, total=30.0),    # no timeouts by default
    limits=punkreq.Limits(max_connections=100),
    proxy="http://proxy.internal:3128",          # env vars honored by default
    verify=True,                                 # bool | CA bundle path | ssl.SSLContext
)

By default: redirects on (max 10), no timeouts, no cookie jar unless requested, HTTP/2 enabled, 90s keep-alive expiry, unlimited connections.

Timeout has four independent fields — total (a deadline for the whole request, redirects included), connect, read and pool. There is no write timeout: request bodies are written full-duplex from background tasks, so total is the bound for slow uploads.

Backends

punkreq never asks for a backend at call sites — pick it by import:

from punkreq.asyncio import Client   # asyncio, available everywhere
from punkreq.tonio import Client     # tonio, free-threaded CPython >= 3.14 on Unix

License

punkreq is released under the BSD License.