Building Your Pipeline¶

In a nutshell

The intents.pipeline list in mycroft.conf is the whole arbitration model for intent matching. OVOS tries each stage in the order you list, and the first one that claims the utterance wins. This page shows three fully worked pipeline configs, built only from stages documented elsewhere in this manual, so you can see how the ordering rules play out in practice. See the Intent Pipeline Overview for the full stage reference.

How the pipeline list works¶

Every entry in intents.pipeline is a pipeline ID for a plugin already installed and initialized. OVOS walks the list in order and hands the utterance to each stage until one returns a match. There is no cross-plugin score comparison. Each plugin's own conf_high / conf_med / conf_low thresholds are its private accept gate, not a rank the orchestrator compares between plugins.

Reordering the list changes which stage gets first refusal on an utterance. Removing an ID from the list stops that stage from being tried, but the plugin is still loaded at startup unless you uninstall its package. See Pipelines Overview for the full list of available stage IDs and their confidence tiers.

The three configs below are complete and directly runnable. Nothing is left as "add remaining stages as needed."

Example A: the stock voice-assistant default¶

This is the bundled default pipeline shipped in mycroft.conf, reproduced exactly as listed in the Pipelines Overview:

{
  "intents": {
    "pipeline": [
      "ovos-stop-pipeline-plugin-high",
      "ovos-converse-pipeline-plugin",
      "ovos-ocp-pipeline-plugin-high",
      "ovos-padatious-pipeline-plugin-high",
      "ovos-adapt-pipeline-plugin-high",
      "ovos-m2v-pipeline-high",
      "ovos-ocp-pipeline-plugin-medium",
      "ovos-fallback-pipeline-plugin-high",
      "ovos-stop-pipeline-plugin-medium",
      "ovos-padatious-pipeline-plugin-medium",
      "ovos-adapt-pipeline-plugin-medium",
      "ovos-fallback-pipeline-plugin-medium",
      "ovos-fallback-pipeline-plugin-low"
    ]
  }
}

Why each stage sits where it does:

ovos-stop-pipeline-plugin-high: stop is a dedicated, always-on pipeline plugin, not a skill. It must sit first so an exact "stop" or "cancel" is intercepted before any other stage sees the bare word. See Stop Pipeline.
ovos-converse-pipeline-plugin: converse depends on session state (which skills are active, who is awaiting a reply), so it must run before the general matchers so a skill mid conversation can claim "yes", "the kitchen", or "stop" before a generic parser would. See Converse Pipeline.
ovos-ocp-pipeline-plugin-high: OCP's high tier claims explicit media intents and, while it holds paused media for the session, bare control words like "resume" or "next". It needs to run early for the same session-aware reason as converse. See OCP Pipeline.
ovos-padatious-pipeline-plugin-high: Padatious's high tier runs before Adapt's, so a strong example-sentence match wins over a keyword rule. See Padatious Pipeline.
ovos-adapt-pipeline-plugin-high: Adapt's high tier runs right after Padatious's, so a strong keyword match still beats most fallbacks even when no example sentence matched. See Adapt Pipeline.
ovos-m2v-pipeline-high: placed after the high tiers of Padatious and Adapt, so exact and keyword matches win first and Model2Vec only catches the paraphrases the deterministic engines miss. See Model2Vec Pipeline.
ovos-ocp-pipeline-plugin-medium: catches utterances classified as media queries by keyword rather than an explicit intent, once the higher-confidence stages have passed. See OCP Pipeline.
ovos-fallback-pipeline-plugin-high: the fallback plugin's high-priority range (0 < p ≤ 5) is reserved for critical fallback handlers, tried before medium-confidence matchers get another pass. See Fallback Pipeline.
ovos-stop-pipeline-plugin-medium: a fuzzy (non-exact) stop match, for stop phrases that are not a literal "stop" or "cancel". See Stop Pipeline.
ovos-padatious-pipeline-plugin-medium (in the default list since ovos-config 2.3.9a1): catches template matches that score below conf_high, most importantly utterances whose auto-registered .entity slot values are out-of-list and therefore cap below the high threshold. Without this stage those utterances go unmatched on a default install.
ovos-adapt-pipeline-plugin-medium: Adapt's medium tier, a second, looser pass over keyword rules.
ovos-fallback-pipeline-plugin-medium and ovos-fallback-pipeline-plugin-low: general and catch-all fallback skills, tried last, in ascending priority order, only once every regular matcher has passed.

Note what is deliberately absent: Padatious/Adapt's low tier, Model2Vec's medium/low tiers, OCP's low tier, Common Query, Persona, and stop/padatious's low tiers are not in the default list. You add them yourself if you want them.

Example B: a media-focused device¶

A device used mainly for music or podcasts can afford to promote OCP further and add its low tier, which the OCP Pipeline page warns should only run "on devices used mainly for media playback" because it can fire on any phrase containing a known artist or show name.

{
  "intents": {
    "ovos-ocp-pipeline-plugin": {
      "min_score": 50
    },
    "pipeline": [
      "ovos-stop-pipeline-plugin-high",
      "ovos-converse-pipeline-plugin",
      "ovos-ocp-pipeline-plugin-high",
      "ovos-padatious-pipeline-plugin-high",
      "ovos-adapt-pipeline-plugin-high",
      "ovos-ocp-pipeline-plugin-medium",
      "ovos-fallback-pipeline-plugin-medium",
      "ovos-ocp-pipeline-plugin-low"
    ]
  }
}

Why each stage sits where it does:

ovos-stop-pipeline-plugin-high and ovos-converse-pipeline-plugin: stop and converse stay first for the same session-state reasons as Example A. A media device still needs "stop" and an in-progress conversation to win before playback control does.
ovos-ocp-pipeline-plugin-high: promoted right after converse, ahead of Padatious/Adapt, because on this device explicit playback commands and stateful "resume"/"next" control words are the primary use case.
ovos-padatious-pipeline-plugin-high and ovos-adapt-pipeline-plugin-high: regular skill intents (setting timers, asking the weather, and so on) still get their normal high-confidence pass.
ovos-ocp-pipeline-plugin-medium: catches keyword-classified media queries that the explicit-intent high tier missed, per the OCP Pipeline quick-start example.
ovos-fallback-pipeline-plugin-medium: a general fallback pass before the most permissive OCP tier gets a turn, so a misheard non-media utterance is not swallowed by a loose media keyword hit.
ovos-ocp-pipeline-plugin-low: last, and only justified on a media-focused device, exactly as the OCP page recommends. It keys off any skill-registered media keyword, so placing it last limits the damage from a false positive.

Example C: a persona/chatbot-first assistant¶

This reproduces the Hybrid Mode configuration from the Persona Pipeline page: skills get first refusal, and the persona (an LLM-backed conversational agent) fills in wherever they fall short, before the catch-all fallback skills get a turn.

{
  "intents": {
    "persona": {
      "handle_fallback": true,
      "default_persona": "Remote Llama"
    },
    "pipeline": [
      "ovos-stop-pipeline-plugin-high",
      "ovos-converse-pipeline-plugin",
      "ovos-padatious-pipeline-plugin-high",
      "ovos-adapt-pipeline-plugin-high",
      "ovos-persona-pipeline-plugin-high",
      "ovos-padatious-pipeline-plugin-medium",
      "ovos-adapt-pipeline-plugin-medium",
      "ovos-persona-pipeline-plugin-low",
      "ovos-fallback-pipeline-plugin-high",
      "ovos-fallback-pipeline-plugin-medium",
      "ovos-fallback-pipeline-plugin-low"
    ]
  }
}

Why each stage sits where it does:

ovos-stop-pipeline-plugin-high and ovos-converse-pipeline-plugin: unchanged from Examples A and B. Stop must always win over a chatty persona, and an in-progress skill conversation still gets priority.
ovos-padatious-pipeline-plugin-high and ovos-adapt-pipeline-plugin-high: regular skills get first refusal, which is the point of Hybrid Mode: "preserves traditional voice assistant behavior" while the persona "fills in where skills fall short." See Persona Pipeline.
ovos-persona-pipeline-plugin-high: placed after the deterministic high tiers but before their medium tiers. It engages only for a persona already summoned in the session; it never consults default_persona.
ovos-persona-pipeline-plugin-low: the stage where handle_fallback: true and default_persona actually take effect. Without it those keys are dead config and a fresh session never reaches the persona: the low stage is what catches open-ended chat and unmatched questions before the fallback ladder.
ovos-padatious-pipeline-plugin-medium and ovos-adapt-pipeline-plugin-medium: a second, looser skill-matching pass in case the persona missed something a skill could still answer.
ovos-fallback-pipeline-plugin-high, -medium, -low: the standard fallback ladder closes out the pipeline, in ascending priority order, in case neither skills nor the persona claimed the utterance.

For a persona-first deployment where the chatbot should override skills entirely, see Persona Pipeline: Full Control mode instead.

Ordering rules of thumb¶

Stop goes first. STOP-1 only works because a high-confidence stop stage placed first in the pipeline can intercept the bare word "stop" before any other plugin claims it. See Stop Pipeline.
Converse goes early, before the general matchers. It depends on session state (is a skill still active, is someone awaiting a reply), so it needs first refusal on utterances that continue an existing exchange. See Converse Pipeline.
High tiers before medium tiers before low tiers. A high-confidence match is tried before any medium-confidence stage, and a medium-confidence stage before any low-confidence one. This is the arbitration model itself, not an optional convention. See Pipelines Overview.
Statistical/semantic stages (Model2Vec, Hierarchical KNN) go after the deterministic engines' high tiers. They generalize well, which also means they can claim an utterance a more precise parser would have nailed if placed too early. See Model2Vec Pipeline and Hierarchical KNN Pipeline.
Fallback goes last. By design it only runs "after all other matchers fail," and its own high/medium/low tiers further order fallback skills by priority range. See Fallback Pipeline.
A persona's placement encodes its strategy. First and ahead of skills for a full-control chatbot, after the skills' high tiers for a hybrid assistant, or last (a -low stage) for a pure "default persona when nothing else matched" setup. See Persona Pipeline.

Read next: Debugging Intent Matching Related: Pipelines Overview · Adapt Pipeline · Padatious Pipeline · OCP Pipeline