Today I received the official notification: my paper, "DoSync: Governing AI-Agent Intents over Heterogeneous IoT Devices", was peer-reviewed and accepted at the Workshop on IoT Interoperability and the Web of Things (IIWOT) at IEEE CCNC 2027[cite: 1].
For an independent developer building open-source tooling outside of an academic lab or corporate R&D budget, getting an architecture through rigorous technical peer review is a huge milestone. Three specialists in IoT and distributed systems audited the design, checked the benchmarks, and validated the core premise: we cannot let non-deterministic LLMs directly trigger physical actuation in the real world.
Why I Built DoSync
As agentic frameworks and protocols like MCP make it trivial to connect LLMs to local networks and smart devices, a dangerous gap has opened up. Language models are probabilistic they hallucinate, misinterpret context, or can be steered via prompt injection. Physical hardware, on the other hand, has irreversible consequences.
An AI agent should be able to express an intent (e.g., "prepare the living room for the night"), but a deterministic layer must govern what actually happens on the wire.
What the Reviewers Validated
The paper introduces DoSync, an open protocol and reference hub designed to sit between high level AI intents and heterogeneous IoT devices. During the review process, three core pillars of the project stood out:
- Nine Deterministic Guarantees (G1–G9): Formal rules enforced at runtime covering strict capability matching, hierarchical place scoping, oscillation dampening, tamper-evident hash-chained audit logging, and human-in-the-loop holds for critical actions.
- Standards-Native WoT Integration: Rather than inventing a closed ecosystem, DoSync maps directly to W3C Web of Things (WoT) Thing Descriptions (TD 1.1) via an extension vocabulary (
dosync:), validated against hundreds of public third-party TDs. - Sub-10 ms Edge Execution: The entire resolution and governance pipeline runs locally on constrained hardware (tested on a Raspberry Pi 5) with under 10 ms latency—zero cloud dependency required.
Keeping It 100% Open Source
Academic publishing often locks validated research behind expensive paywalls and conference fees. Since DoSync was built from day one as a free, community-driven project for everyone, I am keeping the full architecture, specification, and implementation completely open:
- 📄 Read the Full Paper: Grab the pre-print PDF directly from the DoSync Protocol Repository.
- 📱 Try DoSync App (Beta): Want to govern your smarthome devices locally using LLMs from your phone—with no accounts, no extra commercial hubs, and zero cost? Check out the DoSync App page or download the latest release from GitHub.
- 🌐 Protocol Overview & Docs: dosync.dev
Getting peer-reviewed confirmation that an independent, open-source architecture solves a real frontier problem is the best fuel to keep building. If you're working on local AI agents, home automation, or IoT security, feel free to inspect the repo, break things, or open an issue!
