FAQ

General

What is Pounce?

Pounce is a free-threading-native ASGI server for Python 3.14t. It serves ASGI applications using real OS threads sharing a single interpreter, rather than the traditional fork-based model.

Does Pounce work on standard (GIL) Python?

Yes. On GIL builds, Pounce automatically uses processes instead of threads. The API and configuration are identical — the supervisor detects the runtime viasys._is_gil_enabled()and adapts.

What Python version do I need?

Python 3.14 or later. For thread-based workers (the primary use case), you need a free-threading build (Python 3.14t).

What's the current status?

Pounce is in beta. HTTP/1.1 is core; HTTP/2, HTTP/3, and WebSocket are optional extras with their own install and proof boundaries. TLS, multi-worker serving, compression, static files, middleware, rate limiting, OpenTelemetry, Prometheus, and streaming are available as optional server features or integrations. It's a young project, so expect rough edges and check the feature-specific docs before treating a path as production-critical.

Compatibility

Does Pounce work with FastAPI / Starlette / Django?

Yes. Pounce serves any standard ASGI application. If your framework produces a valid ASGI callable, Pounce can serve it.

Can I use Pounce as a drop-in replacement for Uvicorn?

For most use cases, yes. The CLI syntax is similar:

# Uvicorn
uvicorn myapp:app --host 0.0.0.0 --port 8000 --workers 4

# Pounce
pounce myapp:app --host 0.0.0.0 --port 8000 --workers 4

See Migrate from Uvicorn for details.

Does Pounce support ASGI lifespan?

Yes. Pounce sendslifespan.startup and lifespan.shutdownevents per the ASGI specification.

Technical

Why threads instead of processes?

On Python 3.14t, the GIL is removed. Threads can execute Python code in true parallel. Threads sharing one interpreter means:

• One copy of the app — not N copies
• Shared frozen configuration — oneServerConfigobject, with per-request mutable state kept separate
• No IPC overhead — workers communicate through memory, not pipes

What's the overhead of Pounce vs raw asyncio?

Minimal. The request pipeline is: parse → scope → app → serialize → write. The ASGI bridge is per-request with no shared mutable state. The dominant cost is your application, not the server.

How does compression work?

Pounce negotiates content-encoding viaAccept-Encoding. Priority: zstd > gzip > identity. Zstd uses Python 3.14's stdlib compression.zstd(PEP 784) — zero external dependencies.

What is Server-Timing?

A W3C standard that surfaces server-side latency in browser DevTools. When enabled (--server-timing), Pounce injects parse/app/encode timings into every response header.

Part of Bengal

Do I need other Bengal packages to use Pounce?

No. Pounce works with any ASGI framework. It's part of the Bengal ecosystem but fully standalone. See Ecosystem for the full picture.