Nlcli is a python package that offers a natural language interface for your programs. It is primarly focused on command line interaction, but it can be used as a chatbot and even as an interface for HTTP based programs; checkout the examples.

Built on top of mycroftai/padatious and libfann/fann.

You can run this tutorial interactively on google colab.

Both https://github.com/libfann/fann and https://github.com/swig/swig are required in order to install nlcli.

On macos you might have to compile FANN from source, swig can be instaled via brew.

To install the libs on linux, run the following command:

!sudo apt-get -qq install libfann-dev swig

Install nlcli using pip:

!pip install -q nlcli

import nlcli


@nlcli.cmd(["hi", "Hi my name is {name}"])
def hi(name=""):
    return f"hi {name}"


@nlcli.cmd(
    ["search for {query} on {engine}", "search {query} on {engine}", "search {query} {engine}"]
)
def search(query, engine="google"):
    return f"query: {query}, engine: {engine}"


if __name__ == "__main__":
    nlcli.interact()

$ python example.py hi my name is joao
hi joao

$ python example.py search python on bing
query: python, engine: bing

Call nlcli.interact to parse the query and call the function intent automatically:

>>> nlcli.interact("search for brazil on google", debug=True)

{'name': 'search', 'sent': ['search', 'for', 'brazil', 'on', 'google'], 'matches': {'query': 'brazil', 'engine': 'google'}, 'conf': 1.0}

query: brazil, engine: google

>>> nlcli.interact("hi", debug=True)

{'name': 'hi', 'sent': ['hi'], 'matches': {}, 'conf': 1.0}

hi

>>> nlcli.interact("hi, my name is joao", debug=True)

{'name': 'hi', 'sent': ['hi', ',', 'my', 'name', 'is', 'joao'], 'matches': {'name': 'joao'}, 'conf': 1.0}

hi joao

By default, nlcli comes with two builtin commands: help and exit. When nlcli fails to match a query with an intent, help command will be automatically called, you can change the default command by passing default=True to the desired @nlcli.cmd decorator.

>>> nlcli.interact("help", debug=True)

{'name': 'help', 'sent': 'help', 'matches': {}, 'conf': 0.49447914140834637}

Heres what I can do:

help - usage: help (|me) with {skill}

bye - usage: Goodbye!

hi - usage: hi

search - usage: search for {query} on {engine}

All custom commands have help automatically, if you want help on a command, simply ask for it:

>>> nlcli.interact("help me with search", debug=True)

{'name': 'help', 'sent': ['help', 'me', 'with', 'search'], 'matches': {'skill': 'search'}, 'conf': 1.0}

here are some examples on how to use search:

search for {query} on {engine}

search {query} on {engine}

search {query} {engine}

!git clone -q https://github.com/joaorafaelm/nlcli.git && cd nlcli

!python -m examples.getting_started hi my name is joao

• Implement data augmentation lib in order to generate more samples, e.g commands with typos, grammar errors, wrong spelling, etc. https://github.com/makcedward/nlpaug & https://github.com/QData/TextAttack
• Support intent files so that intent data can be decoupled from code.
• Response template and i18n.
• Read from stdin.
• tests & ci setup.