Skip to content

Latest commit

 

History

History
177 lines (115 loc) · 15 KB

CONTRIBUTING.md

File metadata and controls

177 lines (115 loc) · 15 KB
Raw

Contributing to Continue

Table of Contents

❤️ Ways to Contribute

👋 Continue Contribution Ideas

This GitHub project board is a list of ideas for how you can contribute to Continue. These aren't the only ways, but are a great starting point if you are new to the project.

🐛 Report Bugs

If you find a bug, please create an issue to report it! A great bug report includes:

  • A description of the bug
  • Steps to reproduce
  • What you expected to happen
  • What actually happened
  • Screenshots or videos

✨ Suggest Enhancements

Continue is quickly adding features, and we'd love to hear which are the most important to you. The best ways to suggest an enhancement are

  • Create an issue

    • First, check whether a similar proposal has already been made
    • If not, create an issue
    • Please describe the enhancement in as much detail as you can, and why it would be useful

  • Join the Continue Discord and tell us about your idea in the #feedback channel

📖 Updating / Improving Documentation

Continue is continuously improving, but a feature isn't complete until it is reflected in the documentation! If you see something out-of-date or missing, you can help by clicking "Edit this page" at the bottom of any page on docs.continue.dev.

🧑‍💻 Contributing Code

Please make PRs to the preview branch. We use this to first test changes in a pre-release version of the extension.

Environment Setup

VS Code

Pre-requisite: You should have Node.js version 20.11.0 (LTS) or higher installed. You can get it on nodejs.org or, if you are using NVM (Node Version Manager), you can set the correct version of Node.js for this project by running the following command in the root of the project:

nvm use

  1. Clone and open in VS Code the Continue repo https://github.com/continuedev/continue

  2. Open the VS Code command pallet (cmd/ctrl+shift+p) and select Tasks: Run Task and then select install-all-dependencies

  3. Start debugging:

    1. Switch to Run and Debug view
    2. Select Extension (VS Code) from drop down
    3. Hit play button
    4. This will start the extension in debug mode and open a new VS Code window with it installed
      1. The new VS Code window with the extension is referred to as the Host VS Code
      2. The window you started debugging from is referred to as the Main VS Code

  4. To package the extension, run npm run package in the extensions/vscode directory. This will generate extensions/vscode/build/continue-patch.vsix, which you can install by right-clicking and selecting "Install Extension VSIX".

Note: Breakpoints can be used in both the core and extensions/vscode folders while debugging, but are not currently supported inside of gui code. Hot-reloading is enabled with Vite, so if you make any changes to the gui, they should be automatically reflected without rebuilding. Similarly, any changes to core or extensions/vscode will be automatically included by just reloading the Host VS Code window with cmd/ctrl+shift+p "Reload Window".

JetBrains

Pre-requisite: You should use the Intellij IDE, which can be downloaded here. Either Ultimate or Community (free) will work. Continue is built with JDK version 19.

  1. Clone the repository
  2. Run install-dependencies.sh or install-dependencies.ps1 on Windows
  3. Run cd extensions/vscode && node scripts/prepackage.js (this will copy over the built React application to the proper JetBrains directory)
  4. Select the "Run Plugin" Gradle configuration and click the "Run" or "Debug" button
  5. To package the extension, choose the "Build Plugin" Gradle configuration

For changes to Typescript code, the binary/gui will currently need to be rebuilt. Changes to Kotlin code can often be hot-reloaded with "Run -> Debugging Actions -> Reload Changed Classes"

Continue's JetBrains extension shares much of the code with the VS Code extension by utilizing shared code in the core directory and packaging it in a binary in the binary directory. The Intellij extension (written in Kotlin) is then able to communicate over stdin/stdout in the CoreMessenger.kt file.

Formatting

Continue uses Prettier to format JavaScript/TypeScript. Please install these extensions in VS Code and enable "Format on Save" in your settings.

Writing Slash Commands

The slash command interface, defined in core/index.d.ts, requires you to define a name (the text that will be typed to invoke the command), a description (the text that will be shown in the slash command menu), and a run function that will be called when the command is invoked. The run function is an async generator that yields the content to be displayed in the chat. The run function is passed a ContinueSDK object that can be used to interact with the IDE, call the LLM, and see the chat history, among a few other utilities.

export interface SlashCommand {
  name: string;
  description: string;
  params?: { [key: string]: any };
  run: (sdk: ContinueSDK) => AsyncGenerator<string | undefined>;
}

There are many example of slash commands in core/commands/slash that we recommend borrowing from. Once you've created your new SlashCommand in this folder, also be sure to complete the following:

  • Add your command to the array in core/commands/slash/index.ts
  • Add your command to the list in config_schema.json. This makes sure that Intellisense shows users what commands are available for your provider when they are editing config.json. If there are any parameters that your command accepts, you should also follow existing examples in adding them to the JSON Schema.

Writing Context Providers

A ContextProvider is a Continue plugin that lets type '@' to quickly select documents as context for the language model. The IContextProvider interface is defined in core/index.d.ts, but all built-in context providers extend BaseContextProvider.

Before defining your context provider, determine which "type" you want to create. The "query" type will show a small text input when selected, giving the user the chance to enter something like a Google search query for the GoogleContextProvider. The "submenu" type will open up a submenu of items that can be searched through and selected. Examples are the GitHubIssuesContextProvider and the DocsContextProvider. The "normal" type will just immediately add the context item. Examples include the DiffContextProvider and the OpenFilesContextProvider.

After you've written your context provider, make sure to complete the following:

Adding an LLM Provider

Continue has support for more than a dozen different LLM "providers", making it easy to use models running on OpenAI, Ollama, Together, LM Studio, and more. You can find all of the existing providers here, and if you see one missing, you can add it with the following steps:

  1. Create a new file in the core/llm/llms directory. The name of the file should be the name of the provider, and it should export a class that extends BaseLLM. This class should contain the following minimal implementation. We recommend viewing pre-existing providers for more details. The LlamaCpp Provider is a good simple example.
  • providerName - the identifier for your provider
  • At least one of _streamComplete or _streamChat - This is the function that makes the request to the API and returns the streamed response. You only need to implement one because Continue can automatically convert between "chat" and "raw completion".
  1. Add your provider to the LLMs array in core/llm/llms/index.ts.
  2. If your provider supports images, add it to the PROVIDER_SUPPORTS_IMAGES array in core/llm/index.ts.
  3. Add the necessary JSON Schema types to config_schema.json. This makes sure that Intellisense shows users what options are available for your provider when they are editing config.json.
  4. Add a documentation page for your provider in docs/docs/reference/Model Providers. This should show an example of configuring your provider in config.json and explain what options are available.

Adding Models

While any model that works with a supported provider can be used with Continue, we keep a list of recommended models that can be automatically configured from the UI or config.json. The following files should be updated when adding a model:

  • config_schema.json - This is the JSON Schema definition that is used to validate config.json. You'll notice a number of rules defined in "definitions.ModelDescription.allOf". Here is where you write rules that can specify something like "for the provider 'anthropic', only models 'claude-2' and 'claude-instant-1' are allowed. Look through all of these rules and make sure that your model is included for providers that support it.
  • modelData.ts - This file defines that information that is shown in the model selection UI in the side bar. To add a new model:
    1. create a ModelPackage object, following the lead of the many examples near the top of the file
    2. add the ModelPackage to the MODEL_INFO array if you would like it to be displayed in the "Models" tab
    3. if you would like it to be displayed as an option under any of the providers, go to the PROVIDER_INFO object and add it to the packages array for each provider that you want it to be displayed under. If it is an OS model that should be valid for most providers offering OS models, you might just be able to add it to the osModels array as shorthand.
  • index.d.ts - This file defines the TypeScript types used throughout Continue. You'll find a ModelName type. Be sure to add the name of your model to this.
  • LLM Providers: Since many providers use their own custom strings to identify models, you'll have to add the translation from Continue's model name (the one you added to index.d.ts) and the model string for each of these providers: Ollama, Together, and Replicate. You can find their full model lists here: Ollama, Together, Replicate.
  • Prompt Templates - In this file you'll find the autodetectTemplateType function. Make sure that for the model name you just added, this function returns the correct template type. This is assuming that the chat template for that model is already built in Continue. If not, you will have to add the template type and corresponding edit and chat templates.

Adding Pre-indexed Documentation

Continue's @docs context provider lets you easily reference entire documentation sites and then uses embeddings to add the most relevant pages to context. To make the experience as smooth as possible, we pre-index many of the most popular documentation sites. If you'd like to add new documentation to this list, just add an object to the list in preIndexedDocs.ts. startUrl is where the crawler will start and rootUrl will filter out any pages not on that site and under the path of rootUrl.

📐 Continue Architecture

Continue consists of 2 parts that are split so that it can be extended to work in other IDEs as easily as possible:

  1. Continue GUI - The Continue GUI is a React application that gives the user control over Continue. It displays the current chat history, allows the user to ask questions, invoke slash commands, and use context providers. The GUI also handles most state and holds as much of the logic as possible so that it can be reused between IDEs.

  2. Continue Extension - The Continue Extension is a plugin for the IDE which implements the IDE Interface. This allows the GUI to request information from or actions to be taken within the IDE. This same interface is used regardless of IDE. The first Continue extensions we have built are for VS Code and JetBrains, but we plan to build clients for other IDEs in the future. The IDE Client must 1. implement IDE Interface, as is done here for VS Code and 2. display the Continue GUI in a sidebar, like here.

Continue VS Code Extension

The starting point for the VS Code extension is activate.ts. The activateExtension function here will register all commands and load the Continue GUI in the sidebar of the IDE as a webview.

Continue JetBrains Extension

The JetBrains extension is currently in alpha testing. Please reach out on Discord if you are interested in contributing to its development.