sm-aicli 1.6.1

Command-line interface for a number of AI models

A (yet another) GNU Readline-based application for interaction with chat-oriented AI models.

Supported model providers:

• OpenAI via REST API
• GPT4All via Python bindings

Contents

• Install
  • Pip
  • Nix
• Usage
  • Example session
• Vim integration

Install

The following installation options are available:

Pip

$ pip install sm_aicli

Nix

$ git clone --depth=1 https://github.com/sergei-mironov/aicli && cd aicli
# Optionally, change the 'nixpkgs' input of the flake.nix to a more suitable
$ nix profile install ".#python-aicli"

Usage

usage: aicli [-h] [--model-dir MODEL_DIR] [--model [STR1:]STR2]
             [--num-threads NUM_THREADS] [--model-apikey STR]
             [--model-temperature MODEL_TEMPERATURE] [--device DEVICE]
             [--readline-key-send READLINE_KEY_SEND]
             [--readline-prompt READLINE_PROMPT] [--readline-history FILE]
             [--verbose NUM] [--revision] [--version]

Command-line arguments

options:
  -h, --help            show this help message and exit
  --model-dir MODEL_DIR
                        Model directory to prepend to model file names
  --model [STR1:]STR2, -m [STR1:]STR2
                        Model to use. STR1 is 'gpt4all' (the default) or
                        'openai'. STR2 is the model name
  --num-threads NUM_THREADS, -t NUM_THREADS
                        Number of threads to use
  --model-apikey STR    Model provider-specific API key
  --model-temperature MODEL_TEMPERATURE
                        Temperature parameter of the model
  --device DEVICE, -d DEVICE
                        Device to use for chatbot, e.g. gpu, amd, nvidia,
                        intel. Defaults to CPU
  --readline-key-send READLINE_KEY_SEND
                        Terminal code to treat as Ctrl+Enter (default: \C-k)
  --readline-prompt READLINE_PROMPT
                        Input prompt (default: >>>)
  --readline-history FILE
                        History file name (default is '_sm_aicli_history'; set
                        empty to disable)
  --verbose NUM         Set the verbosity level 0-no,1-full
  --revision            Print the revision
  --version             Print the version

The console accepts language defined by the following grammar:

start: (command | escape | text)? (command | escape | text)*
escape.3: /\\./
command.2: /\/ask|\/exit|\/help|\/reset/ | \
           /\/model/ / +/ (/"/ model_string /"/ | /"/ /"/) | \
           /\/apikey/ / +/ (/"/ apikey_string /"/ | /"/ /"/) | \
           /\/nthreads/ / +/ (number | def) | \
           /\/verbose/ / +/ (number | def) | \
           /\/temp/ / +/ (float | def ) | \
           /\/echo/ | /\/echo/ / /
model_string: (model_provider ":")? model_name
model_provider: "gpt4all" -> mp_gpt4all | "openai" -> mp_openai | "dummy" -> mp_dummy
model_name: /[^"]+/
apikey_string: (apikey_schema ":")? apikey_value
apikey_schema: "verbatim" -> as_verbatim | "file" -> as_file
apikey_value: /[^"]+/
number: /[0-9]+/
float: /[0-9]+\.[0-9]*/
def: "default"
text: /(.(?!\/|\\))*./s

Example session

$ aicli

Type /help or a question followed by the /ask command (or by pressing `C-k` key).
>>> /model "~/.local/share/nomic.ai/GPT4All/Meta-Llama-3-8B-Instruct.Q4_0.gguf"
>>> Hi!
>>> /ask
Hello! I'm happy to help you. What's on your mind?^C
>>> What's your name?
>>> /ask
I don't really have a personal name, but you can call me "Assistant"

Vim integration

Aicli is supported by the Litrepl text processor.