// About
An operating system born from hitting the same wall, repeatedly.
SynapticOS started as a researcher's frustration: modern microcontrollers ship with genuinely capable neural accelerators, but the operating systems running on them were designed for a world of control loops and sensor polling. The accelerator is the most expensive silicon on the die — and the software treats it as a peripheral.
What we believe
On an AI-era microcontroller, inference is not a task among tasks — it is the workload the chip was built for. That conviction drives every design decision in SynapticOS: memory is organized by tensor lifetime rather than by heap, accelerator state is explicit rather than implied, models are managed resources with versions and integrity checks, and the scheduler's unit of work is the inference job itself.
We also believe engineering claims belong next to their evidence. Every number on this site is measured, the measurement setup is documented, and limitations are stated plainly — stub-baseline results are labeled as stub-baseline results. The research page carries the full argument.
Who builds it
SynapticOS is created and maintained by Dimitrios Kafetzis, a researcher working on systems for edge AI. The project spans the full stack — from the memory allocator and NPU hardware abstraction to the developer shell and the accompanying research papers.
Where it is going
SynapticOS is an open-source project today, and it is engineered like a product: a six-phase roadmap to v1.0, hardware verification at every milestone, a test suite that runs without hardware, and documentation that tracks the code. The ambition is bigger than a repository — as the runtime matures toward production hardening and an ecosystem SDK, SynapticOS is built to grow into a platform, and potentially into a company. If the problem it solves matters to your team or your product, we want to hear from you.
Get in touch
GHGitHub — Dimitrios-Kafetzis/SynapticOS: star the repo, open an issue, watch releasesRSResearch — executive summaries and preprints@Contact — dimitrioskafetzis@gmail.com
Community contributions open in an upcoming phase with a contributor guide and issue templates. Until then: stars, issues, and hard questions are very welcome.