Skip to content
forked from tenxstudio/devtale

Document whole code repositories with LLMs

License

Notifications You must be signed in to change notification settings

vwagner/devtale

 
 

Repository files navigation

devtale

Every software product depends on some legacy, undocumented code repositories, whose authors left the company years ago.

Who isn't afraid to make a change, if the code is unreadable?

devtale generates repo documentation with LLMs. Currently supporting Python, Go, PHP, JS, TypeScript and React projects.

devtale-demo.mp4

Installation

Use the following command:

pip install devtale

Usage

Terminal

Note: currently, devtale only supports GPT-4 as LLM.

  • Create a .env file in the root directory and set your OPENAI_API_KEY there.

  • Run the following command to document a file or document all the files in a folder:

devtale -p [path/to/your/code] -o [path/to/docs]

To document an entire repository/directory include the -r (recursive) flag. The program generates a JSON file per code file with the documentation data; If you want to also add the documentation inside a copy of the code file, then please include the -f (fuse) flag. In case you already have the JSON docstrings, you can also fuse them separately using the corresponding script

Workflow

Note: You must check the box "Allow GitHub Actions to create and approve pull requests" in your repository's setting -> actions for this to work.

  • In the repository setting -> Secrets and Variables -> Actions -> Create OPENAI_API_KEY repository secret

  • Add the following step in your workflow. Ensure to specify the write permission to avoid 403 errors when attempting to push the devtale PR.

jobs:
  code-documentation:
    runs-on: ubuntu-latest
    permissions: write-all
    steps:
      - name: Document
        uses: mystral-ai/[email protected]
        with:
          OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
          path: ${{ github.workspace }}
          recursive: true
          target_branch: main
          save_tales: false

The recursive option allows you to document the entire repository. Alternatively, you can specify a specific path to document a single file or folder and set recursive to false. The workflow action will automatically create the devtale/documentation branch and push a new pull request for your review towards the target_branch, including the added documentation.

Cost

You can estimate project documentation costs using the --estimate flag when using the devtale command in terminal, which won't make any GPT calls. Please note that this estimate is approximate, as it doesn't include GPT output tokens nor code items JSON tokens generated by the GPT call in the devtale/utils.py/extract_code_elements function.

If you skip the estimate and simply run 'devtale,' the final log will display the total cost. This total cost considers GPT output tokens and the tokens added by the code items JSON.

For example:

Running the following command:

devtale -r -f -p devtale/devtale --estimate

will provide you with a cost estimate of $2.44733 USD. It won't document anything since it doesn't trigger any GPT calls. On the other hand, running:

devtale -r -f -p devtale/devtale

will give you the total cost of $2.95762 USD. This command triggers GPT calls and, consequently, documents the directory.

Dependency on GPT-4

We found that GPT-3.5 can't extract code components and generate docstring in a reliable manner, while GPT-4 can do so. Hence, devtale currently only works with GPT-4. Beware that the cost associated with running devtale on a large code repository may be prohibitive. To reduce this cost, devtale uses gpt-3.5-turbo-0613 for generating top-level file summaries and README files and gpt-4-1106-preview for the rest.

Pipeline

In a high-level overview, devtale works in the following way:

pipeline

About

Document whole code repositories with LLMs

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 100.0%