nairobi-os 0.4.2

Nairobi OS: Heavy Iron Data Science Infrastructure with Lagos Vision (Hardware-Accelerated Zero-Copy Rendering)

English | 简体中文 | Español | Deutsch

Nairobi OS

Overview

Nairobi OS is a high-performance, distributed AI and data science infrastructure designed for extreme resource efficiency. It enables processing of massive datasets in constrained environments (Edge, IoT, Serverless) by leveraging a specialized Rust-based refinery daemon, and provides Computer Use without pixels via its MCP-compliant accessibility bridge.

By utilizing kernel-level features such as io_uring, memfd, and Huge Pages, Nairobi OS achieves sub-millisecond IPC overhead and zero-copy data pipelines.

Key Features

• Computer Use Without Pixels: Interacts directly with the Linux desktop through AT-SPI2 and the TOON (Token-Oriented Object Notation) compression algorithm, bypassing OCR or visual processing for AI agents.
• Zero-Copy Ingestion: Hardware-accelerated data loading using io_uring and 1GB Huge Pages.
• Hardware-Accelerated Visualization: Interactive Jupyter plotting via the Lagos Vision engine (wgpu and egui).
• Fused Analytics Pipeline: Ingest, crunch, and correlate data in a single D-Bus round trip.
• Kernel-Bypass Performance: Vectorized analytics leveraging Polars and Rayon for maximum hardware saturation.
• Sovereign Interface: A fluent Python API that hides the complexity of low-level IPC and memory management.

Architecture

Nairobi OS is built on a triad of specialized components connected via D-Bus and shared memory:

1. Nairobi Axum Refinery: The high-performance Rust core. Manages raw data ingestion and parallelized analytics.
2. Nairobi Hub: The IPC orchestrator. Coordinates file descriptors and signals between the refinery and clients.
3. Lagos Vision: The visual cortex. A headless rendering engine that maps memfd handles directly into the GPU pipeline.
4. Nairobi Connector: The semantic bridge. An MCP server exposing the Linux desktop's accessibility tree to LLMs.
5. Nairobi Python: The high-level bridge. Provides a Pythonic interface to the entire Rust ecosystem.

Installation

From PyPI

pip install nairobi-os

Build from Source

1. Clone the Repository:

   git clone https://github.com/KevinKenya/nairobi-connector-open-source
   cd nairobi-connector-open-source
   

2. Setup Virtual Environment:

   python3 -m venv .venv
   source .venv/bin/activate
   pip install maturin pyo3-build-config zbus anywidget traitlets
   

3. Build the Entire Stack:

   ./build_wheel.sh
   

Usage

Data Analytics

import nairobi_os

# Ignite the refinery
nairobi_os.connect()

# Ingest data into a SovereignFrame
df = nairobi_os.read_csv("dataset.csv")

# Perform vectorized analytics
print(f"Mean: {df.column_name.mean()}")

# Spawn interactive visualization
df.plot()

Computer Use (MCP Server)

AI agents using the Nairobi Connector should follow this fundamental loop via MCP tools:

1. Target a window using nairobi_find_window.
2. Observe the current state via nairobi_get_ui_map.
3. Read the TOON [ID: N] of the desired interactive element.
4. Execute an action on that element via nairobi_interact or nairobi_type_text.

System Configuration (Contributor Guide)

Huge Pages

The refinery engine prioritizes 1GB Huge Pages for zero-copy buffers. To enable these on your host:

echo 1 | sudo tee /sys/kernel/mm/hugepages/hugepages-1048576kB/nr_hugepages

Note: If 1GB pages are unavailable, the engine will automatically fall back to Transparent Huge Pages (THP).

io_uring and SQPOLL

The DiracEngine uses io_uring with SQPOLL for maximum I/O throughput.

Support

If you find Nairobi OS useful, consider supporting the project:

License

This project is licensed under the Apache License 2.0. (Note: Portions of the TOON format and bridge implementation are credited to The TOON Authors.)

---

© 2026 Kevin Chege. All Rights Reserved.