Intent & Entity File Format

Padatious intents are plain text files where each line is a training sample. The syntax supports optional words, alternatives, and named entity slots.

Syntax Elements

Plain words

Any sequence of words is matched literally (case-insensitive).

turn on the lights
play some music

Alternatives (a|b|c)

Parentheses with pipe-separated options expand to multiple training samples.

(turn on|switch on|enable) the lights
play (some|a bit of) music

An empty alternative makes the group optional:

(please |)turn on the lights   # "please" is optional

Entity slots {name}

Curly-brace references are named capture groups. The extracted value is returned in MatchData.matches.

search for {query} on {engine}
remind me to {task} at {time}

Entity names may only contain lowercase letters and underscores.

Combining alternatives and slots

(search|look) for {query} (on|using|with) {engine}
set a {duration} (timer|alarm)

Special tokens

Token	Meaning
#	Matches any digit (\d)
:0	Wildcard, matches one or more words

call extension ##      # matches "call extension 42"
play :0                # matches "play anything here"

Intent files (.intent)

Each line is one sample utterance. Blank lines and leading/trailing whitespace are ignored.

# home.lights.intent
turn on the (living room |bedroom |)lights
switch on (all |)the lights
lights on

Load a file into a container:

container.load_intent('home.lights', 'home.lights.intent')

Or pass samples inline:

container.add_intent('home.lights', [
    'turn on the (living room |bedroom |)lights',
    'switch on (all |)the lights',
    'lights on',
])

Entity files (.entity)

Each line is one example value for the entity. These constrain what the neural network considers a valid match for the slot.

# weekday.entity
monday
tuesday
wednesday
thursday
friday

Load and use:

container.add_intent('weather', ['will it rain on {weekday}?'])
container.load_entity('weekday', 'weekday.entity')
container.train()

Without an entity file the slot will still capture any text, but the neural network match may be less precise.

Blacklisted words

When registering an intent you can pass blacklisted_words to suppress matches that contain specific words:

container.add_intent('play_music', ['play {song}'], blacklisted_words=['video'])

The intent will not fire if the word "video" appears anywhere in the utterance.

Tips

• More varied training samples generally improve accuracy.
• Keep entity files focused — only include realistic values.
• Use alternatives to cover common phrasings rather than writing each one separately.
• Intents are independent. The engine scores all of them and returns the best match.

---

← Installation · Home · API Reference →