RetractorDB

RetractorDB is an Edge Signal Processing Engine for processing and recording regular time series close to their sources. It combines a declarative query language (RQL), continuous signal processing and binary storage for telemetry, monitoring and IoT applications.

⚠️ This is work in progress: Active development is ongoing.

Website | Documentation (PL) | Documentation (EN) | Releases

Overview

• Regular streams: exact rational sampling intervals, synchronization, interleaving, rate conversion and sliding windows.
• Continuous queries: RQL compiles stream transformations, field expressions and rules into an execution plan.
• Execution and storage: sequential plan evaluation, named engine instances, shared-memory access to live results, and binary artifacts with descriptors and metadata.
• Python and Jupyter: an embedded engine with NumPy output, DLPack windows and optional PyTorch integration.

Linux x86-64 and ARM64 are the deployment platforms. macOS on Apple silicon is a development and test platform built from source; see the build guide for requirements and tested configurations. The standalone engine supports clocked processing and optional Linux real-time scheduling, without hard real-time guarantees. The embedded Python engine runs on a logical timeline.

Components

Component	Role
xretractor	Compiles and executes RQL plans, standalone or as a service.
xqry	Reads live results, inspects instances and sends ad-hoc queries or plan replacements through IPC.
xtrdb	Inspects and edits offline binary storage, descriptors and metadata.
Python Engine, Storage, Descriptor	Executes plans inside Python and accesses offline storage without a running service.

The monitoring clients for Python and C++ use xqry to access a running engine. They are separate from embedded plan execution.

Getting started

The Polish manual is the canonical documentation; the English manual is its synchronized translation. Installation, configuration and runnable examples are maintained there:

Task	PL	EN
Install Linux binaries: portable archives, DEB, systemd, upgrades	Instalacja	Linux releases
Build from source: Linux and macOS, Conan 2, CMake, Ninja	Budowanie	Source build
Run a first plan or load a service plan	Uruchamianie	Running
Set up Python, Jupyter or Colab	Python i Jupyter	Python and Jupyter
Use the embedded Python API, NumPy and PyTorch	API Python	Python API
Configure the engine, storage and service	xretractor	xretractor

For local examples, see RQL plans and Python notebooks. The manuals also include signal-filter, ECG and OHLC walkthroughs: PL / EN.

Development

See CONTRIBUTING.md for contribution guidelines and CLAUDE.md for repository build, test and collaboration rules. Release packaging is documented in scripts/release_package/README.md.

License and author

MIT License. Created and maintained by Michal Widera (2003-2026).

Contact: [email protected]