Skip to content

Rift: an AI-native language server for your personal AI software engineer

License

Notifications You must be signed in to change notification settings

ZeroXClem/rift

 
 

Repository files navigation

Rift

Rift is open-source infrastructure for AI-native development environments. Rift makes your IDE agentic. Software will soon be written mostly by AI software engineers that work alongside you. Codebases will soon be living, spatial artifacts that maintain context, listen to, anticipate, react to, and execute your every intent. The Rift Code Engine implements an AI-native extension of the language server protocol. The Rift VSCode extension implements a client and end-user interface which is the first step into that future.

rift.mp4

Features

Conversational code editing

code edit screencast

Codebase-wide edits

aider screencast

Contextual codebase generation

smol screencast

Tips

  • Press Command+M to focus the Rift Omnibar.
    • Once focused, you can either engage with the current chat or use a slash-command (e.g. /aider) to spawn a new agent.
  • Each instance of a Rift Chat or Code Edit agent will remain attached to the open file / selection you used to spawn it.
    • To switch to a new file or request a code edit on a new selection, spawn a new agent by pressing Command+M and running a slash-command (e.g. /edit)
    • Both Rift Chat and Code Edit see a window around your cursor or selection in the currently active editor window. To tell them about other resources in your codebase, mention them with @.
    • Code Edit
  • You can @-mention files and directories to tell your agents about other parts of the codebase.
    • @-mentioning files currently only works with Aider if those files are tracked by git.
  • Currently, Rift works best when the active workspace directory is the same as the root directory of the git project.
  • Command+Shift+P -> "Rift: Start Server" restarts the server if it has been auto-installed.

Getting started

Install the VSCode extension from the VSCode Marketplace. By default, the extension will attempt to automatically start the Rift Code Engine every time the extension is activated. During this process, if a rift executable is not found in a virtual environment under ~/.morph, the extension will ask you to attempt an automatic installation of a Python environment and the Rift Code Engine. To disable this behavior, such as for development, go to the VSCode settings, search for "rift", and set rift.autostart to false.

When rift.autostart is true, the extension will attempt to automatically start the Rift Code Engine. You can set rift.riftPath to change the path of the Rift executable, which may be necessary due to interactions with WSL on Windows.

When rift.autostart is false, the extension will display a loading indicator while it waits for a server instance to connect to rift.riftServerPort (default 7797). In this scenario, you will have to start the Rift server instance manually by running it in a terminal, e.g. with

source ~/.morph/env/bin/activate
rift --port 7797

If the automatic installation of the Rift Code Engine fails, follow the below instructions for manual installation.

Manual installation

Rift Code Engine:

  • Set up a Python virtual environment for Python 3.10 or higher.
    • On Mac OSX:
      • Install homebrew.
      • brew install python@3.10
      • mkdir -p ~/.morph/ && cd ~/.morph/ && python3.10 -m venv env
      • source ./env/bin/activate
    • On Linux:
      • On Ubuntu:
        • sudo apt install software-properties-common -y
        • sudo add-apt-repository ppa:deadsnakes/ppa
        • sudo apt install python3.10 && sudo apt install python3.10-venv
        • mkdir -p ~/.morph/ && cd ~/.morph/ && python3.10 -m venv env
        • source ./env/bin/activate
      • On Arch:
        • yay -S python310
        • mkdir -p ~/.morph/ && cd ~/.morph/ && python3.10 -m venv env
        • source ./env/bin/activate
    • On Windows:
      • We recommend that you use WSL with Ubuntu. Once inside a WSL shell, follow the Ubuntu installation instructions above.
      • Make sure inbound connections over port 7797 from WSL to Windows are allowed (e.g. try following this guide but for port 7797 instead of 3000).
      • On Windows we recommend that users disable rift.autostart in VSCode and run Rift manually as ~/.morph/env/bin/rift after following the installation instructions below.
  • Install Rift. We recommend that you pip install Rift in a dedicated Python >=3.10 virtual environment from this repository.
    • Make sure that which pip returns a path whose prefix matches the location of a virtual environment, such as the one installed above.
    • using pip from GitHub:
      • pip install "git+https://github.com/morph-labs/rift.git@main#egg=pyrift&subdirectory=rift-engine"
    • From source:
      • cd ~/.morph/ && git clone git@github.com:morph-labs/rift && cd ./rift/rift-engine/ && pip install -e .

Rift VSCode Extension (via code --install-extension, change the executable as needed):

  • From the repository root: cd ./editors/rift-vscode && npm i && bash reinstall.sh. Make sure your OpenAI API key is set in the VSCode settings (open with Ctrl + , then search for "rift").

The road ahead

Existing code generation tooling is presently mostly code-agnostic, operating at the level of tokens in / tokens out of code LMs. The language server protocol (LSP) defines a standard for language servers, objects which index a codebase and provide structure- and runtime-aware interfaces to external development tools like IDEs.

The Rift Code Engine is an AI-native language server which will expose interfaces for code transformations and code understanding in a uniform, model- and language-agnostic way --- e.g. rift.summarize_callsites or rift.launch_ai_swe_async should work on a Python codebase with StarCoder as well as it works on a Rust codebase using CodeGen. Within the language server, models will have full programmatic access to language-specific tooling like compilers, unit and integration test frameworks, and static analyzers to produce correct code with minimal user intervention. We will develop UX idioms as needed to support this functionality in the Rift IDE extensions.

Contributing

We welcome contributions to Rift at all levels of the stack, for example:

  • adding support for new open-source models in the Rift Code Engine
  • implementing the Rift API for your favorite programming language
  • UX polish in the VSCode extension
  • adding support for your favorite editor.

See our contribution guide for details and guidelines.

Programming is evolving. Join the community, contribute to our roadmap, and help shape the future of software.

Feedback

We'd love to hear your feedback on Rift! Share your thoughts with us here.

About

Rift: an AI-native language server for your personal AI software engineer

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 76.8%
  • TypeScript 13.1%
  • Svelte 8.6%
  • Other 1.5%