ellie

$ echo "It's dangerous to go alone" | ellie
Take this!

ellie is a Rust-based command-line interface to ChatGPT, featuring support for function calling through external providers.

Functions

Function calling is supported by delegating to external providers. All you have to do is configure a function provider in ~/.config/ellie/functions.toml (or the equivalent path in your platform).

Provider configuration

To configure a function provider, add the following information to the functions.toml file:

[[provider]]
name = "get_current_weather"
command = "python"
args = ["get_current_weather.py"]

This example configures a provider named "get_current_weather" that uses a Python script called "get_current_weather.py".

Provider behavior

A function provider reads from the standard input and writes results to the standard output. When given an extra spec argument, it writes a specification to the standard output.

For example, running the provider with the spec argument would output the following specification:

$ python get_current_weather.py spec
{
  "name": "get_current_weather",
  "description": "Get the current weather in a given location",
  "parameters": {
    "type": "object",
    "required": ["location"],
    "properties": {
      "location": {
        "type": "string",
        "description": "The city and state, e.g. San Francisco, CA"
      },
      "unit": {
        "type": "string",
        "enum": ["celsius", "fahrenheit"]
      }
    }
  }
}

To call the function, you can provide the required input as a JSON object through the standard input. The provider will then output the result as a JSON object.

For example:

$ echo '{"location":"Boston, MA"}' | python get_current_weather.py
{
  "location": "Boston, MA",
  "temperature": "72",
  "unit": null,
  "forecast": ["sunny", "windy"]
}

Template implementation

Here is a template implementation in Python:

from sys import argv
import json

match argv[1:]:
    case []:
        print(json.dumps({
            # ...
        }))
    case ["spec"]:
        print(json.dumps({
            # ...
        }))

You can write function providers in any programming language. For more information on function specifications, refer to the OpenAI official guide.

Detailed output

TL;DR: use logging for now.

If you just want to see functions being called, use RUST_LOG=info:

$ echo 'What is the weather like in Boston?' | RUST_LOG=info ellie
 INFO  ellie > get_current_weather {"location":"Boston, MA"}
The weather in Boston is currently sunny and windy with a temperature of 72 degrees.

For debugging information (e.g., the exact request payloads), use RUST_LOG=debug:

$ echo 'What is the weather like in Boston?' | RUST_LOG=debug ellie
 DEBUG ellie > {"model":"gpt-3.5-turbo","messages":[{"role":"user","content":"What is the weather like in Boston?"}],"functions":[{"name":"get_current_weather","description":"Get the current weather in a given location","parameters":{"properties":{"location":{"description":"The city and state, e.g. San Francisco, CA","type":"string"},"unit":{"enum":["celsius","fahrenheit"],"type":"string"}},"required":["location"],"type":"object"}}],"temperature":0.0,"max_tokens":null}
 INFO  ellie > get_current_weather {"location":"Boston, MA"}
 DEBUG ellie > {"model":"gpt-3.5-turbo","messages":[{"role":"user","content":"What is the weather like in Boston?"},{"role":"assistant","content":"","function_call":{"name":"get_current_weather","arguments":"{\"location\":\"Boston, MA\"}"}},{"role":"function","content":"{\"forecast\":[\"sunny\",\"windy\"],\"location\":\"Boston, MA\",\"temperature\":\"72\",\"unit\":null}","name":"get_current_weather"}],"functions":[{"name":"get_current_weather","description":"Get the current weather in a given location","parameters":{"properties":{"location":{"description":"The city and state, e.g. San Francisco, CA","type":"string"},"unit":{"enum":["celsius","fahrenheit"],"type":"string"}},"required":["location"],"type":"object"}}],"temperature":0.0,"max_tokens":null}
The weather in Boston is currently sunny and windy with a temperature of 72 degrees.