Fluxium

A fast, modern HTTP client for Python.

Fluxium provides a clean, requests-like API with HTTP/2 multiplexing, automatic retries, connection pooling, streaming, SSE, middleware hooks, built-in caching, and full async support — with zero blocking calls on the asyncio event loop.

import fluxium

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

Verify installation:

python -c "import fluxium; print(fluxium.__version__)"

Installation

pip install fluxium

Optional extras:

pip install "fluxium[socks]"     # SOCKS proxy support
pip install "fluxium[uvloop]"    # 20-40% faster async
pip install "fluxium[cache]"     # Hishel RFC 7234 cache backend
pip install "fluxium[all]"       # Everything

Features

Feature	Description
HTTP/2	Multiplexed connections, header compression (default enabled)
Connection Pooling	Up to 200 connections, 100 keep-alive, with optional pre-warming
Automatic Retries	Exponential backoff for 5xx and timeouts
Streaming & SSE	iter_content(), iter_lines(), iter_sse()
Built-in Caching	In-memory, disk-based, or RFC 7234 (hishel) with TTL
Middleware	Hooks for logging, auth refresh, rate limiting
OAuth2 / Bearer	Automatic token management and refresh
Async	Full asyncio support via AsyncSession
uvloop	Auto-used when installed for 20-40% async throughput

Quick Start

Basic Requests

import fluxium

r = fluxium.get("https://api.example.com", timeout=10)
print(r.status_code, r.json())

r = fluxium.post("https://api.example.com/items", json={"name": "widget"})

Sessions (Connection Pooling)

with fluxium.Session() as s:
    r1 = s.get("https://api.example.com/users")
    r2 = s.post("https://api.example.com/items", json={"name": "x"})
    # Cookies from r1 are automatically sent with r2

Pre-warming Connections

with fluxium.Session() as s:
    s.prewarm("https://api.example.com")  # Opens connection now
    r = s.get("https://api.example.com")  # Uses pooled connection

Async

import asyncio, fluxium

async def main():
    async with fluxium.AsyncSession() as s:
        tasks = [s.get(f"https://api.example.com/item/{i}") for i in range(10)]
        results = await asyncio.gather(*tasks)

asyncio.run(main())

Retries & Caching

from fluxium import Session, MemoryCache

with Session(max_retries=3, cache=MemoryCache()) as s:
    r = s.get("https://api.example.com")  # retried on failure, cached on success

Streaming & SSE

with fluxium.Session() as s:
    r = s.get("https://api.example.com/stream", stream=True)
    for line in r.iter_lines():
        print(line)

    r = s.get("https://api.example.com/events", stream=True)
    for event in fluxium.iter_sse(r):
        print(event.event, event.json())

Timeouts

from fluxium import Timeout

# All components same timeout
fluxium.get("https://api.example.com", timeout=30.0)

# Structured timeout
fluxium.get("https://api.example.com", timeout=Timeout(connect=5.0, read=30.0))

# Or use tuple shorthand (connect, read)
fluxium.get("https://api.example.com", timeout=(5.0, 30.0))

Rate Limiting

from fluxium import Session, RateLimitMiddleware

s = Session()
s.add_middleware(RateLimitMiddleware(calls=100, period=60))  # 100 req/min

Per-Request Hooks

s = Session()
s.add_hook("response", lambda response, request: log(response))
s.add_hook("error", lambda error, request: notify(error))

Performance

Cached workloads: fluxium is 4.3x faster than httpx. On unique requests, performance is comparable.

Scenario	vs httpx
Repeated GET with MemoryCache	~4.3x faster
Session GET (pooled)	~1.6x slower (per-request overhead)
One-shot GET	~1.03x (negligible)
Async concurrent	~1.5x slower
Body encoding / CookieJar	~1.0x (identical)

Key takeaway: fluxium wins on repeated requests (caching) and is competitive on unique requests. Install uvloop for 20-40% async throughput improvement.

Documentation

• Getting Started — install, first request, compatibility
• Guides — step-by-step tutorials for every feature
• API Reference — complete class and method signatures
• Advanced — custom middleware, performance, SSE deep-dive
• Changelog — version history and migration guide

Contributing

See CONTRIBUTING.md.

License

MIT — © 2026 Siddhant Bayas