/sweep

Sweep: AI-powered Junior Developer for small features and bug fixes.

Primary LanguagePythonOtherNOASSERTION

Bug Reports & Feature Requests ⟶  Code Changes

Install Docs

🎉 We recently changed our license to the Elastic License V2 to allow Sweep for commercial usage. Check out https://docs.sweep.dev/deployment

Sweep is an AI junior developer that transforms bug reports & feature requests into code changes.

Describe bugs, small features, and refactors like you would to a junior developer, and Sweep:

  1. Reads your codebase
  2. Plans the changes
  3. Writes a pull request with code ⚡

Features

  • Turns issues directly into pull requests (without an IDE)
  • Addresses developer replies & comments on its PRs
  • Uses embedding-based code & online document search
  • Validates its changes with GitHub Actions and self-review

Why Choose Sweep?

We're not a toy project or a proof-of-concept. We're a production devtool used by startups including ourselves to ship features everyday, with example features here. Unlike Copilot, which only provides IDE-based autocompletion, Sweep handles the entire flow end-to-end. Unlike ChatGPT, Sweep is able to automatically understand and search through your code base, removing the need to tediously copy-and-paste files.

Sweep vs. GPT-Engineer, Smol Developer and AutoGPT Sweep is built to improve on an existing codebase, which is a more frequent and higher need, than generating boilerplate, which is mostly a solved problem since you can just fork existing boilerplates.
Sweep vs. Cody and Bloop We do more than just chat-with-your-code by actually writing code changes.
sweep.mp4

Getting Started

Install Sweep by adding the Sweep GitHub App to your desired repositories.

  • For more details, visit our Installation page.

  • Note: Sweep only considers issues with the "Sweep:" title on creation and not on update. If you want Sweep to pick up an issue after it has been created, you can add the "Sweep" label to the issue.

  • We support all languages GPT-4 supports, including Python, JS/TS, Rust, Go, Java, C# and C++.

Self-hosting

You can self-host Sweep with the Docker image (https://hub.docker.com/r/sweepai/sweep). The setup instructions are at Deployment.

Limitations of Sweep

  • Large-scale refactors: >3 files or >150 lines of code changes (we're working on this!)

    • e.g. Refactor the entire codebase from TensorFlow to PyTorch

  • Editing images and other non-text assets

    • e.g. Use the logo to create favicons for our landing page

  • Performing actions involving a dashboard, including fetching API tokens

    • e.g. Set up sign-in using Ethereum

Development

Starting the Webhook

  1. Install poetry.
  2. Clone the repo with git clone https://github.com/sweepai/sweep.
  3. Install the dependencies with poetry install This will take a few minutes, do the new step in the meantime.
  4. Create .env according to https://docs.sweep.dev/deployment.
  5. Hop into the environment with poetry shell and start the server by running uvicorn sweepai.api:app --port 8080 --workers 3. You can tweak the number of workers but it needs to be at least 2.

Building Docker

To build the docker image, run

docker build -t sweepai/sweep:latest .

Then run docker run --env-file .env -p 8080:8080 sweepai/sweep:latest.

If you are on an ARM-based systems like M1/M2 Mac, you need to install docker buildx and run

docker buildx build --platform linux/amd64 -t sweepai/sweep:latest .

and execute it with docker run --platform linux/amd64 --env-file .env -p 8080:8080 sweepai/sweep:latest.

Story

We were frustrated by small tickets, like simple bug fixes, annoying refactors, and small features. Each task required us to open our IDE to fix simple bugs. So we decided to leverage the capabilities of ChatGPT to address this directly in GitHub.

Unlike existing AI solutions, this can solve entire tickets and can be parallelized + asynchronous: developers can spin up 10 tickets and Sweep will address them all at once.

The Stack

  • GPT-4 32k & GPT-3.5 16k
  • GTE-base embedding model
  • ActiveLoop DeepLake Vector DB
  • Modal Labs for infra + deployment

Highlights

Examine pull requests created by Sweep here.

Pricing

Every user receives unlimited GPT-3.5 tickets and 5 GPT-4 tickets per month. To prevent abuse, users can use 2 GPT-4 tickets a day.

For professionals who want more tickets and priority support/feature requests, check out Sweep Pro now at $480/month. In addition, we're also offering 15 PRs for $60 as a one-time purchase for anyone interested in eventually purchasing Sweep Pro. You can purchase this here.

Roadmap

We plan on rapidly improving Sweep. To see what we're working on, check out our Roadmap.

Star History

Star History Chart

Consider starring us if you're using Sweep so more people hear about us!

Contributing

Contributions are welcome and greatly appreciated! To get set up, clone the repo and run

poetry install
poetry shell

To build the docker image run

For detailed guidelines on how to contribute, please see the CONTRIBUTING.md file.

Contributors

Thank you for your contribution!

and, of course, Sweep!