Talk to ChatGPT from your terminal.
First you'll need to install the CLI:
cargo install chatgpt-cli
Then, you'll need to make sure your cargo bin directory is in your path. You can do this by adding the following to your ~/.bashrc or ~/.zshrc:
export PATH="$PATH:$HOME/.cargo/bin"
Finally, you'll need a OPENAI API key (you can get one here), and you'll need to export your API Key as an environment variable:
export OPENAI_API_KEY=<your api key>
Then you can start a conversation with ChatGPT:
chatgpt what is 2 + 2
You can also send multiline messages:
chatgpt '''
Make this sentence more sophisticated:
I like to eat pizza
'''
Your messages in each terminal window are saved to {data_dir}/{OS boot time}/{terminal pid}/chatlog.json. This means you can ask follow-up questions in a terminal window and start a new conversation by opening a new window.
For example on MacOS the data path is $HOME/Library/Application Support/. For other platforms, refer to the directories documentation.
The CLI is configured either by environment variables or the config file.
To find out where the config file is expected to be, run chatgpt -p. Create that file and populate it with the below values to use it.
The CLI will always use CLI flags over environment variables over the config file.
You can specify the API key either via an environment variable or in the config.json file:
export OPENAI_API_KEY=<your api key>
config.json
{
"openai_api_key": "<your api key>"
}
By default, the CLI uses the gpt-3.5-turbo model.
However, you can use a different model by passing the --model flag:
chatgpt --model=gpt-4 Complete this phrase: "Ravioli ravioli, give me the..."
You can also change the default model by setting the CHATGPT_CLI_MODEL environment variable:
export CHATGPT_CLI_MODEL=gpt-4
Or add this to the config file:
config.json
{
"model": "<gpt-4>"
}
NOTE: The gpt-4 model is not yet available to everyone. You can join the waitlist here.
By default, the CLI will wait 120 seconds for a response from the API. You can increase this timeout by setting the CHATGPT_CLI_REQUEST_TIMEOUT_SECS environment variable:
export CHATGPT_CLI_REQUEST_TIMEOUT_SECS=600