15 unstable releases (3 breaking)
0.4.1 | Oct 1, 2023 |
---|---|
0.4.0 | Oct 1, 2023 |
0.3.1 | Apr 16, 2023 |
0.2.0 | Apr 16, 2023 |
0.1.8 | Mar 27, 2023 |
#1399 in Web programming
35 downloads per month
14KB
148 lines
Talk to ChatGPT from your terminal.
Quickstart
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 ~/.chatgpt/{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.
Settings
Use a different model, like GPT-4
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
NOTE: The gpt-4 model is not yet available to everyone. You can join the wailist here.
Increase the request timeout
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
Dependencies
~7–20MB
~327K SLoC