Skip to content

craigmayhew/pipe-gpt

Repository files navigation

pipe-gpt

Pipe your content to gpt directly from the command line. A concept that allows for a lot of possibilities.

Installation via cargo cargo install pipe-gpt, build it locally via cargo build --release, or download from the releases page.

Please note you will need an OpenAI API Key.

Set the open api key env var

  • in linux: export OPENAI_API_KEY=sk-12345abc
  • in windows powershell $env:OPENAI_API_KEY = 'sk-12345abc'

Use cases

Local command-line usage

  • tail -30 /var/httpd.log | pipe-gpt --p "Is there anything in the http log file I should fix?"
  • cat main.rs | pipe-gpt -p "How would you improve this code? Include line numbers in your comments so I can tell where you mean."
  • cat main.rs | pipe-gpt -p "Is this code production ready? If yes reply 'Yes'. If no, then explain why not. Be concise."
  • cat file.json | pipe-gpt -p "Convert this JSON to YAML" > file.yaml
  • cat french.txt | pipe-gpt -p "Translate this to English please."
  • git diff --staged | pipe-gpt -p "Code review this code change"
  • cat src/main.rs | pipe-gpt -p "improve the code and only output the replacement code as I will pipe the output directly back into a file, no explanations, just pure code please" > src/main.new.rs

pipe-gpt for local dev

user@computer:/GitHub/pipe-gpt$ cat src/main.rs | pipe-gpt -p "how would you improve this code? include line numbers in your comments so I can tell where you mean"

Here are some suggestions to improve your code:

1. Line 2, 3: You can combine these two lines into one as follows:
`use openai_api_rust::{*, chat::*};`

2. Line 6-16: You can simplify the creation of `conversation_messages` by using a `Vec::new()` and then pushing messages as needed. This way, you avoid creating an unnecessary `Message` when `prepend` is empty.

3. Line 18-19: Instead of using `unwrap()`, it's better to handle the error properly. You can use `?` to propagate the error up.

4. Line 20-33: You can create a function to initialize `ChatBody` to make the code cleaner.

5. Line 35-38: Instead of using `unwrap()`, it's better to handle the error properly. You can use `match` to handle the possible `None` case.

pipe-gpt for CI

jobs:
  build:

    runs-on: ubuntu-latest

    steps:
    - uses: actions/checkout@v3
    - name: Build
      run: cargo build --verbose
    - name: Run tests
      run: cargo test --verbose
    - name: GPT Code Review
      env:
        OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
      run: pwd && find . -path './target' -prune -o -name '*.rs' -exec echo {} \; -exec cat {} \; | ./target/debug/pipe-gpt -p "how would you improve this code? include line numbers in your comments so I can tell where you mean"

Which gives this output in Github Actions CI: Pipe GPT used in CI to give code review recomendations

Roadmap

  • graceful API throttling
  • secrets and config files
  • loading custom prompts as short arguments
  • namespaced roles/prompts
  • reduce dependencies

Contributing

If you wish to contribute to the codebase, please see CONTRIBUTE.md

Useful links