RequestX

Drop-in replacement for httpx, powered by Rust. Up to 4x faster, scales linearly with concurrency.

pip install requestx

import requestx as httpx

r = httpx.get("https://api.example.com/data")
print(r.json())

Every httpx API works. No code changes needed.

---

Performance

Benchmarked on Python 3.12, Apple Silicon, local HTTP server, 30s per run.

Requests Per Second (higher is better)

Sync clients:

Concurrency	requestx	httpx	requests	urllib3
1	1,630	1,034	773	1,459
4	5,602	3,208	3,139	3,164
10	6,635	2,391	3,390	1,762

Async clients:

Concurrency	requestx	httpx	aiohttp
1	875	424	1,119
4	5,164	2,633	5,599
10	7,163	1,637	7,167

Speedup vs httpx

Concurrency	Sync	Async
1	1.58x	2.06x
4	1.75x	1.96x
6	2.23x	2.88x
8	2.63x	3.82x
10	2.78x	4.38x

httpx performance degrades under concurrent load (1,576 → 1,322 RPS from c=1 to c=10). RequestX scales linearly (875 → 7,163 RPS).

---

Usage

Basic:

import requestx as httpx

# GET request
r = httpx.get("https://api.example.com/users")
print(r.json())

# POST request
r = httpx.post("https://api.example.com/users", json={"name": "Alice"})

# With a client (connection pooling)
with httpx.Client(base_url="https://api.example.com") as client:
    r = client.get("/users")
    print(r.status_code)

Async:

import requestx as httpx
import asyncio

async def main():
    async with httpx.AsyncClient() as client:
        r = await client.get("https://api.example.com/users")
        print(r.json())

asyncio.run(main())

AI SDKs (OpenAI, Anthropic):

RequestX is a drop-in performance upgrade for AI SDKs that use httpx internally:

import requestx
from openai import OpenAI

# Sync client - up to 4x faster
client = OpenAI(http_client=requestx.Client())
response = client.chat.completions.create(
    model="gpt-4",
    messages=[{"role": "user", "content": "Hello"}]
)

# Async client - scales linearly with concurrency
from openai import AsyncOpenAI
import asyncio

async def main():
    async_client = AsyncOpenAI(http_client=requestx.AsyncClient())
    response = await async_client.chat.completions.create(
        model="gpt-4",
        messages=[{"role": "user", "content": "Hello"}]
    )

asyncio.run(main())

import requestx
from anthropic import Anthropic

client = Anthropic(http_client=requestx.Client())
message = client.messages.create(
    model="claude-3-5-sonnet-20241022",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Hello"}]
)

---

Why It's Fast

RequestX replaces httpx's Python internals with Rust (reqwest + Tokio), compiled via PyO3.

	httpx	requestx
HTTP engine	Python (httpcore)	Rust (reqwest)
Async runtime	Python asyncio	Tokio (GIL-free)
JSON parsing	Python json	sonic-rs (SIMD)
Connection pool	Python-managed	Rust hyper
GIL during I/O	Held	Released
Concurrency scaling	Degrades	Linear

All network I/O runs outside Python's GIL, enabling true parallelism that httpx cannot achieve.

---

Features

• 100% httpx API compatible — 1,406 tests passing, mirrored from httpx test suite
• Sync + Async — Client and AsyncClient with full feature parity
• HTTP/2 — native support via rustls
• Compression — gzip, brotli, deflate, zstd
• Auth — Basic, Digest (RFC 7616), NetRC, custom auth flows
• Proxies — HTTP/SOCKS proxy support, environment variable detection
• Streaming — byte, text, line, and raw iterators (sync and async)
• Cookie persistence — domain/path-aware jar
• Transports — Mock, WSGI, ASGI for testing
• Event hooks — request and response hooks, including on redirects

---

Integration Tests

RequestX includes integration tests that verify compatibility with real AI SDK APIs (OpenAI and Anthropic). These tests make actual API calls and require API keys.

Setup

1. Install integration dependencies:

pip install -e ".[integration]"

2. Set environment variables:

export OPENAI_API_KEY="sk-..."
export ANTHROPIC_API_KEY="sk-ant-..."

Running Integration Tests

# Run all integration tests
pytest tests_integration/ -v

# Run only OpenAI tests
pytest tests_integration/test_openai_integration.py -v

# Run only Anthropic tests
pytest tests_integration/test_anthropic_integration.py -v

Important Notes

• Cost: Tests make real API calls and incur costs (~$0.01 per full run)
• API Keys: Tests skip gracefully if API keys are not set
• CI/CD: These tests should NOT run in regular CI (require secrets, cost money)
• Tests use minimal tokens (max_tokens=10) to minimize costs

---

Compatibility

RequestX passes the full httpx test suite (1,406 tests). API coverage is 98.5% — the only excluded symbol is main (httpx's CLI entry point).

# These all work identically to httpx
import requestx as httpx

httpx.get(...)
httpx.Client(...)
httpx.AsyncClient(...)
httpx.URL(...)
httpx.Headers(...)
httpx.Response(...)
httpx.stream(...)
httpx.Timeout(...)
httpx.HTTPStatusError
httpx.TimeoutException

---

License

MIT