Introduction

Built for developers who need total control, Orkes is a low-abstraction, lightweight, Python library for orchestrating complex workflows. While it is well-suited for agentic systems, its core focus is on providing a general-purpose graph-based framework that combines deep customization with native observation tools. Orkes provides the transparency required to move complex systems from experimental hacks to reliable, observable environments.

Background

The High Abstraction Gap

Today’s higher-level libraries are often abstractions on top of abstractions, hidden under layers of dependencies. While this simplifies the “happy path,” it turns niche production requirements (like properly closing an underlying HTTP connection on a client disconnect) into a complete clusterfuck.

In a self-hosted LLM environment, these hanging ghost connections are detrimental. They act as resource parasites, holding onto VRAM and compute cycles. When your framework buries the low-level hooks needed to manage these lifecycles, a simple networking fix becomes an archeological dig through source code.

The Observability Gap

Because some populare frameworks gatekeep it’s Visualization platform, observability for developer often becomes a second priority. Without out-of-the-box I/O and state observation, we faced a black box effect.

The result? The we cannot understand where on when agent began to hallucinate. We fell into the trap of “spray and pray” prompt engineering, only to realize the “intelligence” wasn’t the problem, turned out (for example) the failure lay in the framewrok inherent tool calling prompt.

Mission

Orkes was built to close these gaps. By keeping abstractions low and making I/O and execution observation a first-class citizen, Orkes ensures that your application’s logic isn’t undermined by the complexity of the underlying systems.

Installation

From PyPI

You can install the latest stable version of the Orkes using pip:

pip install orkes

Verifying the Installation

You can verify the installation by running a simple import:

import orkes
print(orkes.__version__)

If no errors occur and a version number is printed, the installation was successful.