znyx-core
Engine, as a library
The detection engine as a Python library. It runs the checks in-process, inside your own app. No server, no network hop.
pip install znyx-core
For developers
Which package do I install?
ZNYX checks text, your prompts, model output, and tool calls, against your policies and returns a decision: allow, block, or redact. There are three open-source packages, and you almost always want just one or two of them. This page helps you pick in under a minute.
The packages
The three packages
Two of them do the actual checking, znyx-core and znyx-runtime. The third, znyx-sdk, only calls a runtime over the network.
znyx-runtime
Engine, as a service
znyx-core wrapped in a FastAPI web service. Run it once, then many apps in any language call it over HTTP. It already contains the engine.
docker run znyx/runtime # or pip install znyx-runtime znyx-runtime serve
znyx-sdk
Thin client
A thin client library with no detection logic of its own. Add it to your app to call a running runtime over HTTP. Also available in TypeScript, Java, C#, Ruby, and Rust.
pip install znyx-sdk
Decision guide
Pick your setup
Find the row that describes you, then install what it points to.
One Python app, simplest, no server
Install znyx-core only
The engine runs inside your process. Nothing else to deploy or operate. This is the fastest way to start checking text.
pip install znyx-core
A shared service for many apps, non-Python apps, or production
Run znyx-runtime and install znyx-sdk in each app
Stand up the runtime once (Docker or pip), then each application calls it with the SDK. One place to manage policy, any language can talk to it.
docker run znyx/runtime pip install znyx-sdk
The runtime is already running (your company or a teammate set it up)
Install znyx-sdk only
You do not need the engine locally. The SDK is all you need to send text to the existing runtime and read back the decision.
pip install znyx-sdk
At a glance
The same decision, as a matrix
Read across your row to see what to install.
| Your situation | znyx-core | znyx-runtime | znyx-sdk |
|---|---|---|---|
| One Python app, simplest, no server | znyx-core | No | No |
| Shared service, non-Python apps, or production | No | znyx-runtime | Yes, in each app |
| Runtime already running elsewhere | No | No | znyx-sdk only |
The one rule to remember
You need exactly one way to do the checking: znyx-core or znyx-runtime, not both. The SDK is only needed to call a runtime over the network. So you would not combine core with runtime, since the runtime already contains core, and you would not combine core with the SDK, since core needs no runtime to talk to.
znyx-runtime on PyPI and znyx/runtime on Docker Hub are the same product
They are two ways to run the one runtime. Docker is the primary production channel; pip is the friction-free path for local development. The runtime is rules-only out of the box. To add ML detection, point the runtime at an inference sidecar.
FAQ
Package questions
Which of the three packages you need, and which combinations make no sense.
No, and you almost certainly should not install all three. You need exactly one thing that does the checking - znyx-core in-process, or znyx-runtime as a service - plus znyx-sdk only if your app has to call a runtime over HTTP.
They are the same engine with different packaging. znyx-core is a Python library that runs the checks inside your own process, with no server and no network hop. znyx-runtime is that engine wrapped in a web service, so many applications in any language can call it over HTTP. The runtime already contains core.
No. The SDK is a thin client with no detection logic of its own; it exists to call a running runtime. If core is running in your process there is nothing for the SDK to call.
The engine is Python. The client SDK ships for Python, TypeScript, Java, C#, Ruby, and Rust, and anything that can make an HTTP call can use the runtime without an SDK at all.