A Python HTTP client framework with sync and async clients for building resilient service clients. httpware is a thin opinionated wrapper around httpx2 — it re-exports httpx2.Request/httpx2.Response as the public request/response surface, adds a middleware chain (with a built-in resilience suite: AsyncRetry/Retry + RetryBudget, AsyncBulkhead/Bulkhead, AsyncCircuitBreaker/CircuitBreaker, and AsyncTimeout), opt-in typed response decoding, and a status-keyed exception tree raised automatically on 4xx/5xx.

Why httpware

Typed exceptions per HTTP status, typed response bodies, and composable resilience (retry, bulkhead, circuit breaker, timeout) — a thin wrapper over httpx2, not a new HTTP abstraction. See the project README for the full pitch.

Status: Pre-1.0. Public API is subject to change between minor releases until v1.0.

Install

pip install httpware

Optional extras:

pip install httpware[pydantic]   # PydanticDecoder — handles BaseModel + dataclasses + primitives + generics
pip install httpware[msgspec]    # MsgspecDecoder — handles Struct + dataclasses + primitives + generics
pip install httpware[pydantic,msgspec]   # both extras — both decoders register; BaseModel routes to pydantic, Struct to msgspec

First request

Async usage:

import asyncio

from httpware import AsyncClient

async def main() -> None:
    async with AsyncClient(base_url="https://jsonplaceholder.typicode.com") as client:
        response = await client.get("/users/1")
        print(response.json())

asyncio.run(main())

Sync usage:

from httpware import Client

with Client(base_url="https://jsonplaceholder.typicode.com") as client:
    response = client.get("/users/1")
    print(response.json())

Typed decoding via response_model= works the same way in both worlds:

from httpware import AsyncClient
from pydantic import BaseModel


class User(BaseModel):
    id: int
    name: str


async def main() -> None:
    async with AsyncClient(base_url="https://api.example.com") as client:
        user = await client.get("/users/1", response_model=User)
        print(user.name)

Need the raw response and a decoded body from the same call (e.g., for header-based pagination)? See Link header pagination — it uses send_with_response.

Decoder dispatch

When response_model= is set, the client walks decoders in order and picks the first decoder whose can_decode returns True; ordering encodes your preference for shapes more than one decoder could claim. If none claims your response_model, the call raises MissingDecoderError before the HTTP request. See Decoders for the resolution rules and pydantic/msgspec routing.

With resilience middleware

Compose resilience middleware at construction; AsyncBulkhead goes outside AsyncRetry so one slot covers all retry attempts.

from httpware import AsyncClient, AsyncBulkhead, AsyncRetry


async def main() -> None:
    async with AsyncClient(
        base_url="https://api.example.com",
        middleware=[
            AsyncBulkhead(max_concurrent=10),  # cap total in-flight
            AsyncRetry(),                       # default: 3 attempts, full-jitter backoff
        ],
    ) as client:
        user = await client.get("/users/1", response_model=User)

Streaming responses

For large responses or server-sent events, stream the body chunk-by-chunk. stream() is an async context manager:

from httpware import AsyncClient


async def main() -> None:
    async with AsyncClient(base_url="https://api.example.com") as client:
        async with client.stream("GET", "/big-file") as response:
            async for chunk in response.aiter_bytes():
                process(chunk)

stream() auto-raises StatusError subclasses on 4xx/5xx with the response body pre-read, so exc.response.content is accessible from the caught exception.

It does NOT pass through the middleware chain: AsyncRetry, AsyncBulkhead, and any custom middleware are bypassed. (AsyncRetry separately refuses to retry any request — stream or non-stream — whose body was an async-iterable, since streams can't replay across attempts.)

Capping response body size

Both clients accept an opt-in max_response_body_bytes: int | None = None. When set, a response body that exceeds the cap raises ResponseTooLargeError instead of being returned; the default None is unbounded. See Errors for the full trip conditions.

Errors

All errors inherit httpware.ClientError: 4xx/5xx responses raise a typed StatusError subclass automatically, and response_model= decode failures raise DecodeError. See Errors for the full tree and catching strategies.

Observability

Every resilience middleware emits stdlib-logging records (always) and OTel span events (when opentelemetry-api is installed), under stable logger and event names. See Observability for the full contract.

Where to go next

• Resilience reference — every parameter on AsyncRetry, RetryBudget, and AsyncBulkhead; the retry-rule matrix; Retry-After parsing; budget sharing.
• Middleware guide — write your own middleware. Covers the AsyncMiddleware Protocol, the phase decorators, a worked Request-ID propagation example, and OpenTelemetry wiring.
• Errors reference — the full exception tree, catching strategies, exc.response.* access pattern.
• Observability — the stdlib-logging and OTel span-event contract emitted by the resilience middleware.
• Testing guide — mock-transport injection pattern for testing code that uses httpware.
• Recipes — wiring AsyncClient into a modern-di container.
• Architecture Notes — per-capability design notes — invariants, the three protocol seams, exception contract, module layout, testing patterns — under architecture/. Lives in the repo under architecture/.
• Contributing — setup, conventions, workflow.
• Release notes — per-version changelogs.

Part of modern-python

httpware ships under the modern-python org. See the org profile for the categorized index of related templates and libraries.