Welcome to the OpenVoiceOS Blog

Discover the latest updates, guides, and insights from the open-source voice assistant ecosystem. Join our journey of innovation and collaboration.

OVOS Is a Voice Operating System, and It Has a Written Protocol Spec
Featured Article

OVOS Is a Voice Operating System, and It Has a Written Protocol Spec

The OVOS architecture repository is a set of formal, implementation-agnostic specifications for how a voice OS's components talk to each other: specs covering the intent stack, the bus and sessions, the pipeline, audio, GUI, and media. Written in RFC-2119 language, so what counts as correct is written down instead of argued over.

JarbasAl
JarbasAlDecember 2, 2026
Read Article

Latest Articles

By JarbasAl

AGENTS.md and SKILL.md: Making Every OVOS Repo Legible to Your AI

We're rolling out AGENTS.md and SKILL.md across the OpenVoiceOS repositories, not to flood ourselves with AI-generated pull requests, but to put a capable coding assistant in the hands of every user who wants to customize their own assistant.

By JarbasAl

Most OVOS Bus Messages, Typed Models You Can Validate

ovos-pydantic-models is a machine-readable specification of the OVOS MessageBus: hundreds of message types, each a Pydantic model with validated fields. Build a message and get it right, or receive one and check it, before a malformed payload becomes a mystery bug three services away.

By JarbasAl

Teaching OVOS to Match Meaning: Embeddings-Based Intent Parsing

Intent matching has always leaned on keyword and template engines. This R&D work explores semantic-embedding pipelines that generalize to unseen phrasings, alongside strong minimal-dependency baselines for constrained devices.

By JarbasAl

Train Your Own Wake Word for OVOS: WakeForge and a Notebook Anyone Can Run

Custom wake words have always been gated by data, especially recordings of your chosen phrase. WakeForge, a research-grade training suite, tackles this with frozen self-supervised featurizers and a public Colab/Kaggle notebook, so any user can train a personal wake word for their own name, language, or secret phrase.

By JarbasAl

OVOS Speaks Wyoming: Use OpenVoiceOS Engines from Home Assistant

Three bridges expose any OVOS STT, TTS, or wake-word plugin as a Wyoming service, so Home Assistant's Assist pipelines can run the OVOS engine catalog. Deploy them with one docker-compose file.

By JarbasAl

One Agentic Plugin Became Eight: ovos-persona Learns to Reason and Use Tools

The MoU asked for one agentic plugin. We shipped eight reasoning-loop architectures, six built-in toolboxes, and MCP/UTCP tool discovery, turning personas from single-shot LLM calls into local, private agents.

Want to submit a blog?

Share your knowledge with the OpenVoiceOS community! We welcome contributions on topics related to OpenVoiceOS, voice assistants, open-source technology, and more.

Learn How to Contribute