Skip to content

Latest commit

 

History

History
83 lines (58 loc) · 3.28 KB

README.md

File metadata and controls

83 lines (58 loc) · 3.28 KB
Raw

NOTE

Instead of using ata, I would advice most people to use llm by Simon Willison instead. It supports more models and ways to interact with models.




ata: Ask the Terminal Anything

ChatGPT in the terminal

asciicast

TIP:
Run a terminal with this tool in your background and show/hide it with a keypress.
This can be done via: Iterm2 (Mac), Guake (Ubuntu), scratchpad (i3/sway), or the quake mode for the Windows Terminal.

Productivity benefits

  • The terminal starts more quickly and requires less resources than a browser.
  • The keyboard shortcuts allow for quick interaction with the query. For example, press CTRL + c to cancel the stream, CTRL + ↑ to get the previous query again, and CTRL + w to remove the last word.
  • A terminal can be set to run in the background and show/hide with one keypress. To do this, use iTerm2 (Mac), Guake (Ubuntu), scratchpad (i3/sway), or the quake mode for the Windows Terminal.
  • The prompts are reproducible because each prompt is sent as a stand-alone prompt without history. Tweaking the prompt can be done by pressing CTRL + ↑ and making changes.

Usage

Download the binary for your system from Releases. Or use your package manager to install the application. For example, Brew on MacOS or AUR on Arch Linux have packages available.

To specify the API key and some basic model settings, start the application. It should give an error and the option to create a configuration file called ata.toml for you. Press y and ENTER to create a ata.toml file.

Next, request an API key via https://platform.openai.com/api-keys and update the key in the example configuration file. They key permissions can be "Restricted" to only "Model capabilities".

For more information, see:

$ ata --help

FAQ

How much will I have to pay for the API?

Using OpenAI's API for chat is very cheap. Let's say that an average response is about 500 tokens, so costs about $0.015 (with GPT-4). That means that if you do 50 requests per day, then that will cost you about $0.75 per day ($15 per month assuming you only use it only on workdays). If you use GPT-3.5, then the costs will be much lower.

How does this compare to LLM-based search engines such as You.com or Bing Chat?

At the time of writing, the OpenAI API responds much quicker than the large language model-based search engines and contains no adds. It is particularly useful to quickly look up some things like Unicode symbols, historical facts, or word meanings.

Can I build the binary myself?

Yes, you can clone the repository and build the project via Cargo. Make sure that you have Cargo installed and then run:

$ git clone https://github.com/rikhuijzer/ata.git

$ cd ata/

$ cargo build --release

After this, your binary should be available at target/release/ata (Unix-based) or target/release/ata.exe (Windows).