aictl

When interacting with AI models like gemini-pro there are times when you need to add additional context to to support specific prompts based on data of which the model is not aware. This often includes coping and pasting content into the AI chatbot terminal (e.g. bard). This terminal app allows for easier interaction with external data using prompts that add either local files (e.g. FILE:path) or external resources (e.g. URL:url).

The above content prompts must in upper case and at the be at the at beginning of user input.

Install

You can install aictl CLI using one of the following ways:

Go
Binary

See the release section for aictl checksums and SBOMs.

Binary

You can also download the latest release version of aictl for your operating system/architecture from here. Put the binary somewhere in your $PATH, and make sure it has that executable bit.

The official aictl releases include SBOMs

Go

If you have Go 1.17 or newer, you can install latest aictl using:

go install github.com/mchmarny/aictl@latest

API Token

Create API key: https://makersuite.google.com/app/apikey

Either export the API_KEY environment variable, or pass the key as a flag (see below).

export API_KEY="your-key-goes-here"

Run

To start the AI chat using default values:

aictl

If you haven't defined the API_KEY environment variable, you will have to pass it as a flag:

aictl --api-key your-key-goes-here

Additional parameters that can be passed as flags. More details available here:

temperature (float 0.0 to 1.0, default: 0.9) the lower the number the more predictable the answers, higher numbers result in more creative responses.
tokens (int 1 to 2048, default: 100) the maximum number of output tokens that will be returned from each prompt.
top-p (float 0.0 to 1.0, default: 0.9) combined with temperature changes how the model selects tokens for output. Lower value results in less random responses.
top-k (int, default: 40) combined with temperature changes how the model selects tokens for output. Lower value results in less random responses.

Context

You can add your own context into the chat by inserting file content using FILE: or remote content using URL: references. For example, at the chat prompt:

FILE:content/monthly-gas-price.csv

The chat will ask you first for description of the file to understand its content:

chat: Describe content of content/annual-us-gdp.csv:
you: Annual US Gross Domestic Productivity

So then in chat you can combine that data with the content chat already knows:

chat: How can I help?
you: What was the average gas price in US between 2010 and 2015?
chat: The average gas price in the US between 2010 and 2015 was $3.618 per gallon.

Disclaimer

This is my personal project and it does not represent my employer. While I do my best to ensure that everything works, I take no responsibility for issues caused by this code.