Configuration

August 11, 2026 · View on GitHub

Nebulento's OVOS pipeline plugin is configured under the nebulento key inside intents in mycroft.conf (~/.config/mycroft/mycroft.conf).

The two-stage HierarchicalNebulentoPipeline is configured the same way under the nebulento_hierarchical key, and accepts every key below plus domain_threshold. The two plugins can run side by side in one OVOS instance.

Full Example

{
  "lang": "en-US",
  "secondary_langs": ["es-ES", "fr-FR"],
  "intents": {
    "pipeline": [
      "ovos-nebulento-pipeline-plugin"
    ],
    "nebulento": {
      "conf_high": 0.95,
      "conf_med": 0.80,
      "conf_low": 0.50,
      "max_words": 50,
      "strategy": "DAMERAU_LEVENSHTEIN_SIMILARITY"
    }
  }
}

Config Key Reference

`strategy`

Property	Value
Type	`str`
Default	`"DAMERAU_LEVENSHTEIN_SIMILARITY"`
Source	`nebulento/opm.py:72`

Name of the MatchStrategy enum value to use for all intent matching. Must match an enum member name exactly (case-sensitive).

Valid values:

Value	Notes
`"SIMPLE_RATIO"`	Character-level Levenshtein ratio (difflib)
`"RATIO"`	rapidfuzz ratio, effectively same as `SIMPLE_RATIO`
`"PARTIAL_RATIO"`	Substring match, high FP risk
`"TOKEN_SORT_RATIO"`	Sorted tokens, word-order insensitive
`"TOKEN_SET_RATIO"`	Set intersection, highest recall, most FP
`"PARTIAL_TOKEN_RATIO"`	Not recommended for intent gating
`"PARTIAL_TOKEN_SORT_RATIO"`	Not recommended for intent gating
`"PARTIAL_TOKEN_SET_RATIO"`	Not recommended for intent gating
`"DAMERAU_LEVENSHTEIN_SIMILARITY"`	Default, zero FP on benchmark

An unrecognised value falls back to "DAMERAU_LEVENSHTEIN_SIMILARITY" with a warning logged.

`conf_high`

Property	Value
Type	`float`
Default	`0.95`
Source	`nebulento/opm.py:65`

Confidence threshold for match_high. Intents with conf > conf_high are returned at this tier. The OVOS pipeline tries this tier first. A high-confidence match prevents lower-priority engines from running.

`conf_med`

Property	Value
Type	`float`
Default	`0.80`
Source	`nebulento/opm.py:66`

Confidence threshold for match_medium. Intents with conf > conf_med are returned at this tier (after match_high returned nothing).

`conf_low`

Property	Value
Type	`float`
Default	`0.50`
Source	`nebulento/opm.py:67`

Confidence threshold for match_low. This is the lowest tier. Matches above this threshold are returned only if match_high and match_medium both returned nothing. Setting this too low increases the risk of spurious intent matches.

`max_words`

Property	Value
Type	`int`
Default	`50`
Source	`nebulento/opm.py:68`

Maximum number of space-separated tokens an utterance may contain. Utterances exceeding this limit are silently dropped before matching. This guards against extremely long inputs (e.g. dictated paragraphs) which would produce meaningless fuzzy scores.

If all utterances in a request exceed max_words, calc_intent returns None and an error is logged.

`domain_threshold` (hierarchical plugin only)

Property	Value
Type	`float`
Default	`0.0`
Source	`nebulento/opm.py:284`

Only read by HierarchicalNebulentoPipeline under the nebulento_hierarchical key. Minimum confidence the top-level domain classifier must reach for a query to be routed to a domain. When the best domain scores below this value the query is rejected with no intent match. 0.0 (default) disables the gate. Every query is routed to its best domain and rejection is left to the conf_* tiers. Raising it trades recall for precision. See Hierarchical Matching.

`lang` (global key)

Property	Value
Type	`str` (BCP-47)
Default	`"en-US"`
Source	`nebulento/opm.py:59`

The primary language. One IntentContainer is created for this language. Read from the top-level ovos_config.config.Configuration(), not from the nebulento sub-key.

`secondary_langs` (global key)

Property	Value
Type	`List[str]` (BCP-47)
Default	`[]`
Source	`nebulento/opm.py:60-63`

Additional languages to support. One IntentContainer is created per language. Skills that register intents with a lang field matching a secondary language are handled by the corresponding container. Intents registered with a language not in this list (or not close enough by langcodes.closest_match) are silently ignored.

Read from the top-level ovos_config.config.Configuration().

Standalone Library

When using IntentContainer directly (outside OVOS), there is no config file. Strategy and case-folding are set via constructor arguments:

from nebulento import IntentContainer, MatchStrategy

c = IntentContainer(
    fuzzy_strategy=MatchStrategy.TOKEN_SET_RATIO,
    ignore_case=True,
)

See Intent API for the full constructor reference.

← Hierarchical Matching · Home · Benchmark →