LEW - your LLM Enhanced Workflow in your cli

A powerfully simple tool for your CLI. Send your last commands to an LLM and get instant help!

Basic example:

lew.mp4

LEW will also request to run commands to further support your request.

2024-03-29_11-01-01.mp4

Use LEW to quickly navigate AWS.

2024-03-29_16-21-47.mp4

Learn more

I will have a blog post up soon at davidlozzi.com

Known Issues / Bugs

• Only works in iTerm2, doesn't work in VSCode terminal or other terminals where iTerm2 isn't in use.
• Will not work when SSH'd into another server.
• If a command run by lew is prompting for user input, you will not see it. Cancel the job and run the command manually.

TODO

The current state of the utility is great, it has helped me a few times already. But there is more I want it to do:

in no particular order

• Switch to Anthropic just to check it out
• Improve logging capture, remove need for iTerm2 (attempted with ZSH natively)
• Allow user to specify which commands to send, instead of starting from the last command, i.e. send commands 2 and 3 I ran, not the immediate last one.
• Explore how to support when SSH'd in another server (is it as simple as following steps below except for the iTerm2 steps?)
• Create an easier install/setup script
• Currently the script only looks at the last log file, expand it to treat all logs as it's "db"
• Support Bash, PowerShell, Windows Command, and whatever else.
• Allow the LLM to provide updated commands and enable the user to run them
• Send contents of a file as desired to the LLM

Set Up

Prerequisites

To get started, you'll need:

1. Python v3 or later
2. iTerm2 installed on your Mac
   1. You can add the config "terminal.external.osxExec": "iTerm.app" to your VSCode settings.json file to use iTerm in VScode.
3. Clone this repo to your local machine
   1. Run pip install -r requirements.txt to install dependencies.
4. Copy the .env.sample file to .env (it will be ignored by git)
5. Get your OpenAI API Key from openai.com and update .env with your key.

Configure

1. Open iTerm2 (prereqs #2), go to Settings

   1. In Settings, go to Profiles, then select your profile on the left (probably Default)
   2. Go to Session on the right.
   3. Under Miscellaneous, check to enable Automatically log session input to files in:
   4. Also check to enable Log plain text on the right
   5. Click the Change button
   6. Currently, the code looks in ~/iterm2_logs/ for the logs from iTerm2.

      • Create a folder in the root called iterm2_logs and select that

        or

      • Create a folder where ever you want, and update the variable ITERM2_LOGS in .env

   7. Update your cli logging delimiter:
      1. In iTerm2, you'll see your cli command prompt prefix, usually something like username@machinename, i.e. david.lozzi@KHH9RGC0WN on my machine. Copy that value.
      2. Edit the .env file and set your CLI_LOG_DELIMITER value.

2. Edit your ~/.zshrc file, you can do this by:

   1. If you have VSCode in your cli, enter code ~/.zshrc
   2. If you like vim, enter vim ~/.zshrc
   3. Or open ~/, then in Finder, press CMD+SHIFT+. to view hidden files then open the file in your editor of choice.

3. Add the following line:

   • alias lew="python ~/git/lew-llm-cli/app/command.py"
     • Update the location of the command.py file (prereqs #3)

Some more notes

1. The python script command.py will clean up the iterm2_logs folder automatically, keeping only the last 7 days
2. This script itself writes to ~/lew.log and will only keep the last 7 days of logs.

Running lew

Real simple, after an error or command run, just enter lew and hit enter!

More detail:

lew "message to include" --cnt 2

• "message to include", optional (default is empty), is a message you'd like to send along with the last command(s). You can explain what you're doing or whatever it is you need.
• --cnt 2, optional (default is 1), allows you to include more than 1 command and output to GPT, if you wanted to include the last 2 or more commands and their output, set the cnt parameter. You may also set cnt to 0 to have a chat with the bot without sending a command.

Examples

Help with a git command

❯ git status
fatal: not a git repository (or any of the parent directories): .git
❯ lew

Calling GPT to get your answer...

The error message you're seeing, `fatal: not a git repository (or any of the parent directories): .git`, indicates that....

Troubleshooting a specific file

lew "my package file is messing up" --cnt 0                                                                                            ─╯

Calling GPT to get your answer...

To help you effectively, I need to know which package file you're referring to and what specific issue you're encountering. If it's a `package.json` for a Node.js project, or a different package file for another language or framework, please specify. Additionally, running a command that shows the error or problem you're facing with the package file can greatly help in diagnosing the issue.

If you can, please run the command that leads to the error, and then use:

---
lew --cnt 2
---

to send me the command along with its output. If it's an issue with the content of the package file itself, consider running:

---
cat path/to/packagefile
---

And then:

---
lew --cnt 2
---

to show me the contents of the file.%

Chat without sending the last command

lew "how do i list my director?" --cnt 0

Calling GPT to get your answer...

To list the contents of your current directory, please run the following command:

---
ls
---

This will show you all the files and directories in your current working directory.%

Troubleshooting

Some common things I've run into:

• Failed to call the API: 401 - make sure your API key is correct.
• Running lew responds to another command, not your last one. Are you running your current command in iTerm2? Can you see your command and output in the iterm2_logs folder you configured?

If you have other issues, post an Issue here.

Contributing

I love contributions from everyone. Go ahead and Fork the repository, perform your work, and then post a PR.