Skip to content

Glossary of Terms

This glossary defines common terms, acronyms, and concepts used throughout the OpenVoiceOS (OVOS) ecosystem. New to OVOS? Skim this first. Most pages assume these words.

Entries that cite a spec

An entry that links to an OVOS-* spec names part of the formal vocabulary of the OVOS architecture: a concept with a normative, implementation-agnostic definition, and the link goes to the spec that defines it. For the full set and how they fit together, see the spec index. You can safely skip those links on a first read. They matter once you are checking a component for spec conformance, not for everyday use.

Day-one terms

Brand new and just following Your First Skill? The entries you actually need first are: terminal / shell, pip, virtual environment (venv), skill, intent, utterance, wake word, STT, TTS, messagebus, and skill_id. Everything else here is reference material you can meet when a page links to it.

A

Adapt
A keyword-based intent parser used for simple, high-confidence commands. Its .voc keywords use the shared template grammar (OVOS-INTENT-1).
Agent Engine
(Formerly Solver) A plugin that provides answers to questions or interacts with LLMs (e.g., ChatEngine, RetrievalEngine).
ASR (Automatic Speech Recognition)
Another name for STT: it turns spoken audio into text.
Audio Service
The OVOS component (ovos-audio) responsible for TTS synthesis and audio playback.

B

Bus / messagebus
The WebSocket-based communication backbone of OVOS. Every service talks by sending JSON messages over it.

C

Common Play (OCP)
OpenVoiceOS Common Play: a framework and intent handler specialized in finding and playing media (music, video, podcasts). The per-session virtual media player it arbitrates is specified by OVOS-OCP-1.
Constraints file
A version "filter" (a pip constraints file) that pins which package versions a release channel allows.
Converse
A mechanism that allows skills to intercept utterances during an active session (continuous conversation). Specified by OVOS-CONVERSE-1.

D

Dialog
A .dialog file holding sentences OVOS speaks (often with {variables}). The output counterpart of an intent. Its format is specified by OVOS-INTENT-2, its grammar by OVOS-INTENT-1.
Dinkum Listener
ovos-dinkum-listener, the default voice loop: wake word → record → VAD → STT.
Dispatch-shaped topic
A bus topic containing :, assembled from identifiers to address one specific registered handler: canonically <skill_id>:<intent_name>. The : is the marker. Only a formal specification may define a colon-bearing topic shape. OVOS-MSG-1 §2.1.1.
Dotted addressed topic
An ordinary :-free dotted topic (<x>.<y>.<verb>) that names a specific recipient in one of its segments, for example the legacy {skill_id}.converse.request. It is an addressed message, not a dispatch. The spec treats these as legacy: per OVOS-MSG-1 §2.1.1 a spec-conformant dotted topic is a static string, and addressing travels in the payload instead (as in ovos.common_query.request naming its target in data.skill_id).

E

Embeddings
Numeric vectors that capture the meaning of text (or audio/images), so similar things sit close together. Used for semantic search and RAG memory.
Enclosure
The physical hardware housing the assistant (e.g., Mycroft Mark 1).
Entity
A specific piece of data extracted from an utterance (e.g., "London" in "What is the weather in London?").
Entry point
A line in a package's pyproject.toml / setup.py that advertises a plugin or skill class to OVOS, so OPM can discover it (e.g. under opm.stt or opm.skill).
Extras
Optional install bundles in brackets, e.g. ovos-core[mycroft], that pull in predefined groups of components. See Installation.

F

Fallback
A stage in the intent pipeline where skills can attempt to handle utterances that weren't matched by high-priority parsers. Specified by OVOS-FALLBACK-1.
First-match-wins
The pipeline's arbitration model: the orchestrator walks the pipeline plugins in session.pipeline order and dispatches the first one that returns a Match. Ordering is policy: an earlier plugin gets to answer before any later plugin is asked. This (not a confidence ranking) is what lets a stage such as Converse intercept an utterance ahead of normal intent matching. OVOS-PIPELINE-1.

G

G2P (Grapheme-to-Phoneme)
The process of converting written text into phonetic representations for pronunciation. See G2P plugins.
GUI Service
The component (ovos-gui) that manages visual displays and QML/HTML interfaces. Warning: the current ("legacy") GUI is deprecated. There is no usable OVOS GUI. A replacement is in progress.

H

Headless
A device with no monitor or keyboard (e.g. a Raspberry Pi you control over SSH). OVOS runs happily headless.
home.mycroft.ai
The old Mycroft AI cloud account/backend portal used for device pairing, remote skill settings, and STT. OVOS has no equivalent: it is backendless by design, with no account or pairing step required. See Deprecated & Archived Repositories and Migrating from Mycroft.
HiveMind
A protocol and ecosystem for connecting "satellites" (limited hardware) to a central OVOS server.
hivemind-core
In HiveMind, the central server that satellites and clients connect to for the heavy work (skills, STT/TTS, LLM).

I

Intent
The identified goal or request of a user's utterance (e.g., "WeatherIntent").
Intent file (.intent)
A list of example sentences a user might say to trigger a skill: a template (example-based) intent matched by Padatious.
Intent Pipeline
The ordered sequence of parsers and matchers used to resolve an utterance into an intent. Specified by OVOS-PIPELINE-1.
IPA (International Phonetic Alphabet)
A standard notation for the individual sounds of speech, used by G2P / phonemizers.

K

Kirigami
A UI framework from KDE used for building responsive OVOS GUI interfaces.

L

Listener
The service that captures microphone audio, detects the wake word, and records speech for transcription. See Speech Service.

M

Mark 1 / Mark 2
Mycroft's reference hardware devices. The Mark 1 is a faceplate-only speaker, fully supported; the Mark 2 is a Raspberry Pi 4 device with a touchscreen, discontinued and best-effort. See Mark 1 / Mark 2 for current status.
Match Contract
The single method every pipeline plugin exposes: match(utterances, lang, session) → Match \| None. It is the "system-call ABI" of the voice OS: the orchestrator knows nothing about a plugin except this signature. Returning a Match claims the utterance. Returning None declines it. OVOS-PIPELINE-1.
Message
A JSON object sent over the messagebus, containing a type, data, and context.
MiniCroft
A lightweight, in-process version of ovos-core used for end-to-end testing with ovoscope.
mycroft.conf
OVOS's main configuration file (JSON). The same filename is layered across system → distribution → user locations, with the user's copy winning.

O

OCP Stream Extractor
A plugin that resolves abstract URIs (like YouTube links) into playable media streams.
ONNX
An open, portable model format OVOS uses to run STT / TTS / wake-word neural models efficiently across platforms (CPU-friendly).
OPM (ovos-plugin-manager)
The library responsible for discovering and loading OVOS plugins.
Orchestrator
The logical role that drives the intent pipeline: it iterates the pipeline plugins, dispatches the winning Match to the owning handler on <skill_id>:<intent_name>, and emits the handler-lifecycle events (ovos.intent.handler.start / .complete / .error). In OVOS this role is filled by ovos-core, but the spec is implementation-agnostic: any conformant orchestrator works. OVOS-PIPELINE-1.
OVOS (OpenVoiceOS)
The whole open-source, privacy-respecting voice assistant platform this manual documents.
ovos-core
The "brain" service that runs skills and decides which one should answer an utterance. See ovos-core.
ovos-installer
The guided installer (a TUI wizard) that sets OVOS up for you: the recommended way to install.

P

Padatious
An example-based intent parser that uses a small neural network to match utterances against sample sentences.
Persona
A configurable AI personality (backed by an agent engine / LLM) that answers open-ended questions.
PHAL
Platform & Hardware Abstraction Layer. A service that handles low-level hardware interactions (volume, buttons, etc.).
pip
Python's package installer: the command you run (pip install ...) to add a library or skill to a Python environment. See virtual environment (venv) below for which environment it installs into.
Pipeline Plugin
A matcher that exposes the match contract: match(utterances, lang, session) → Match \| None. The orchestrator iterates the installed pipeline plugins in session.pipeline order, first-match-wins. A plugin returns a Match to claim the utterance, or None to pass. Adapt, Padatious, Converse, Fallback, OCP and Persona are all pipeline plugins. OVOS-PIPELINE-1.
Plugin
A swappable building block (a STT, TTS, wake word, VAD… engine). Plugins change how OVOS works. Skills add what it can do.

Q

QML
A declarative language used for designing the user interface of OVOS skills.

R

RAG (Retrieval-Augmented Generation)
Fetching relevant remembered/retrieved context and giving it to an LLM before it answers. See Persona Memory.
Recency-targeted stop
How the stop plugin picks a target: among positive responders to the stoppability poll it stops the entry with the highest activated_at in session.active_handlers. When no handler answers positively it still falls back to recency over the full active_handlers set rather than escalating; a global stop is reserved for explicit global-stop vocabulary and the empty-active_handlers case. OVOS-STOP-1 §4.1.
Release Channel
A stability track, either stable, testing, or alpha, that controls how new the installed packages are. See Release Channels.
Reserved intent_name
A small set of intent names (stop, converse, response, fallback, common_query) that skills must not register, because the orchestrator dispatches them on <skill_id>:<intent_name> on behalf of a pipeline plugin (e.g. Converse emits response, Stop emits stop). Each is leased to its owning spec. OVOS-PIPELINE-1 §7.3.

S

Satellite
In HiveMind, a lightweight device (e.g. a small Pi) that captures voice and forwards it to hivemind-core.
Session
A data structure representing a specific user interaction or conversation state across multiple turns. See Session (carrier) below. Its wire shape is OVOS-SESSION-1, its lifecycle OVOS-SESSION-2.
Session (carrier)
The per-conversation JSON state that rides inside Message.context.session on every bus message: the voice OS's "shared memory". OVOS-SESSION-1 fixes its wire shape and the field registry other specs extend (e.g. session.pipeline, session.response_mode, session.persona_id). OVOS-SESSION-2 fixes its lifecycle: who owns it, when it may be mutated, and the reserved "default" device session.
Skill
A modular add-on that gives OVOS a new ability (e.g., "Weather Skill").
Skill Settings
Per-skill configuration, optionally editable by the user. See Skill Settings.
skill_id
A skill's unique identifier (e.g. ovos-skill-weather.openvoiceos), taken from its entry point. Used to namespace its settings, events, and resources.
Solver
The former name for an Agent Engine. You'll still see it in older configs and code.
SSML (Speech Synthesis Markup Language)
Tags that control how TTS speaks a sentence: pauses, emphasis, pitch, rate. Experimental and engine-dependent: most TTS plugins ignore it. See SSML.
STT (Speech-to-Text)
The process of transcribing spoken audio into text.
systemd
The Linux service manager most OVOS installs run under. It starts each OVOS component (ovos-core, ovos-audio, …) as a background service and restarts it if it crashes. Its unit files describe how each service is launched (which command, which environment).

T

terminal / shell
The text-only command window where you type commands instead of clicking things. On Linux this is usually bash or zsh. Most OVOS install and troubleshooting steps happen here.
Transformer
A plugin that modifies audio, text, or metadata as it moves through the pipeline. Specified by OVOS-TRANSFORM-1. See Transformer Chain below.
Transformer Chain
One of six ordered chains (audio, utterance, metadata, intent, dialog, tts) that reshape an artifact at a fixed point in the utterance lifecycle. Unlike a pipeline plugin (which decides whether to claim an utterance), every transformer in a chain always runs at its injection point. OVOS-TRANSFORM-1.
TTS (Text-to-Speech)
The process of synthesizing spoken audio from text.

U

Utterance
The text transcription of a user's spoken command.

V

VAD (Voice Activity Detection)
The process of detecting when a user starts and stops speaking.
virtual environment (venv)
An isolated Python install with its own set of packages, kept separate from the system Python. You activate a venv (e.g. source path/to/venv/bin/activate) so that pip installs into that isolated copy instead of somewhere else. This matters for OVOS because a skill installed into the wrong environment won't be found by ovos-core.
Visemes
Visual representations of phonemes used for lip-sync animations on a GUI.
Voice Operating System
The framing the formal specs are built around: OVOS is a platform, not a product. Like a general-purpose OS, it defines the boundary between user input and computation, arbitrates which application handles each utterance, and exposes a stable ABI (the match contract) that third-party skills and plugins run against without knowing about each other. See the spec index for the OS-concept ↔ voice-OS mapping. OVOS-PIPELINE-1.
Vocabulary (.voc)
A file of keywords / sentence fragments (not full sentences) used by keyword (rule-based) intents: what Adapt and skills match against.

W

Wake word
The specific phrase (e.g., "Hey Mycroft") that triggers the listener to start recording.
Wyoming
A simple voice protocol (from the Home Assistant world) that OVOS can bridge to and from.

X

XDG
The freedesktop standard for where config / data / cache files live (~/.config, ~/.local/share, …). OVOS follows it, so your settings aren't scattered.

Deprecated & alias terms

Older docs, code comments, and community posts sometimes use terms that have since been replaced. These still mean the same thing. You don't need to update anything working, but prefer the current name in new writing:

Old term Current term
Solver Agent Engine
Intent engine / intent parser Pipeline plugin
Hotword / wake word (as a standalone term) Wake word. Note: this stays as written inside real identifiers, such as the wakeword config keys and the recognizer_loop:wakeword bus message.

Read next: Start Here Related: Concepts Overview · Formal Specifications · Project Timeline & History