sixcircuit / ai.cli

ai client

this is a simple terminal program to connect to the openai api and get (gpt-3.5 gpt-4 gpt-x) responses.

• it contains a framework for creating simple prompt helpers.
• it streams the responses (in a nice format).
• it copies the response to the system clipboard (on macos)
• it also prints some stats about how much the request cost, how many tokens it was, and how long it took.

but i'm lying! it's something way more powerful.

it's a terminal program that abstracts a bunch of ai models and supports a plugin framework.

it just ships with an LLM plugin by default.

it's great for rapidly prototyping AI ideas.

i may update it in the future. it does what I need it to do right now.

it has two dependencies: the openai api and tiktoken.

to install

1. clone this repo
2. run npm install
3. copy ./config.example.js to ./config.js
4. add your openai api key to config.js
5. optionally configure a helper file. copy helper.example.js somewhere and update the helper path in config.js
6. run ./bin/llm gpt-3.5-turbo none "recite to be or not to be from hamlet."

usage:

• llm <model | alias | "cont"> [<helper_name> [prompt]]

you can run the command without a helper name and a prompt. if you run the command without a prompt it will open your editor. if you include a helper but no prompt you'll see the helper prompts (commented out) in the editor too it works kinda like git commit if you don't provide a message it's much easier to edit a big prompt in the editor especially if you're include code.

run llm help to see the various models, aliases and helpers supported.

note: if you get errors about access or rate limits, make sure you have API billing enabled. it's separate from the chat account billing. it should live here: https://platform.openai.com/account/billing/overview

examples:

• llm 4 d say hello
• llm gpt-3.5-turbo none "recite to be or not to be from hamlet."
• etc...

todo

1. store history and allow continuation of previous prompts
2. support other llms with llama.cpp

license

Feel free to use and hack on this program. You can't use the lib/_.js code in any other project -- it's a fragment of some proprietary code I license to people.