Data Framework.
Zero Overhead.

Building blocks to create, control and own data pipelines on any device, assembled by developers and the AI agents that write their code.

View on GitHub

Deterministic
On-device
Auditable
Open source

Data in

Sensor streams
Telemetry
Documents
Email
Scans
and more to come

DF-0 pipeline · on your device

ingest| transform| AI model| validate| act

Then

Store
Alert
Trigger a process
Run inference
Feed the next pipeline

The zero in DF-0

Zero runtime dependencies

Native code that links into your binary. No interpreter, no virtual machine, no background service.

Zero data egress

Data is processed where it lives. Cloud models stay optional and under your control.

Zero telemetry

Nothing reports back to us. What runs on your infrastructure stays on your infrastructure.

What we’re building

A C++20 framework of composable building blocks that you embed directly in your application.

DF-0 is designed so that you ingest data from any source, compose a pipeline in which an AI model is one step among many, and shape the output for whatever comes next. Everything runs on the device, with no network and no hidden services.

Each block is small, tested and documented, so AI coding agents can assemble pipelines from proven parts instead of generating fragile glue code.

The core will be open source under AGPLv3, with a commercial licence for proprietary deployments.

Where DF-0 comes from

Behind DF-0 is a team with more than a decade of hands-on experience in data processing.

We learned where data processing really breaks by building and supporting software that commercial clients such as PwC Singapore, Quantios and ClearTrail run in production: corrupted inputs, undocumented formats and data that must never leave its environment.

DF-0 is the data processing framework we are building on that knowledge, designed to become the standard in its field.

Our vision

The standard for data processing, from autonomous drones to the enterprise edge.

Data processing is moving onto devices, for reasons of privacy, latency and cost. The hardware and the local models are ready.

But on a drone, a medical wearable or an industrial sensor, every megabyte of memory, every milliwatt of battery and every millisecond counts. In trading, a single garbage-collection pause at the wrong moment has a price. Most data tools were built for environments where none of this matters, and they bring interpreters, garbage collectors and heavy runtimes with them.

The lean, reliable software layer that feeds on-device models is what’s missing, and that’s the layer we’re building.

Design principles

Your data stays with you

Sensitive data should be processed where it lives, without leaving your perimeter.

Same result, every run

Deterministic context: the same input gives the same pipeline behaviour, and bad input degrades gracefully instead of crashing.

You own the stack

Transparent, open code that you control, with no black boxes and no lock-in.

Build with us early.

We are working with a small group of design partners and investors while the framework takes shape. If your systems process data that cannot leave its environment, we would like to hear from you.