Howto 2.0
The first major update to howto!

New features
-
OpenAI Completions API.
You can now choose between gpt-3.5-turbo and gpt-4 models. gpt-3.5-turbo is better, faster, and cheaper than davinchi now becomes the new default.
If you feel like GPT-3.5 is not good enough for you, you can upgrade to a more powerful GPT-4` by modifying the config. -
Configuration file.
Now, howto is more configurable via a config file located at~/.howto/config.json).
"model": "gpt-3.5-turbo",
"shell": "zsh",
"max_tokens": 512,
You can select your favorite shell, the maximum length of the generated command, and...
- Custom system messages.
You can modify the AI behavior by providing it with a custom system message.
The system message used in howto 2.0 by default is
You are a CLI tool that converts user requests to shell commands or short scripts. E.g., for `bash command to tar file without compression:`, you should reply `tar -cf file.tar file`. Avoid natural language. If you have to use it, be extremely concise. Less than 5 words.
You are free to modify it any way you want.
- Follow-up questions.
If howto did not generate the right command, you can try to guide it with follow-up messages. This should not significantly affect the performance or costs but should make howto experience more fluid.
For example:
❯❯❯ howto create a bucket
gsutil mb gs://bucket-name
❯❯❯ howto in aws
aws s3api create-bucket --bucket bucket-name
All requests generated at the last minute are considered follow-up.
-
Secure API key storage (MacOS only)
Howto 2.0 is now using MacOS keychain to the API key in a secure way. -
Simpler installation
Howto now includes a simple two-line installation script that automatically picks up the latest binaries for your system.
Breaking changes
- Howto 2.0 does not support davinchi and codex models. They and soon to be disabled in OpenAI API.
- Environment variable HOWTO_OPENAI_MODEL is deprecated in favor of the config file.
Plans for 2.X
Caching will be the next feature in howto. I don't know when it will be implemented, but I hope soon.
In the more distant future, I am planning to give howto more information about the files in the current open directory and to support alternative AI-engines (LLMs), including opensource LLMs.
Final thoughts
While I keep developing howto, I am not planning to significantly alter its functionality or add many new features. Howto has been an extremely simple tool without bells and whistles and it always will be. Just type howto do a thing, and howto will reply with plain text. No extra screens or fancy TUI decorations, no dialogue choice options — howto is dedicated to reducing a single interaction length to a minimum.