Continue

The @continuedev/llm-info package is a lightweight utility used to manage metadata about various Large Language Models (LLMs). It is distinct from @continuedev/openai-adapters, which handles API translation.

@continuedev/llm-info is specifically responsible for:

• Templates: Model-specific prompt templates.
• Capabilities: Defining what a model can do (e.g., tool use, image support, streaming, predicted outputs).
• Model Aliases: Mapping common names to specific model identifiers.

Developers building or extending model support in Continue should use this package to centralize model capabilities and metadata.

Continue is an open source AI code assistant that allows developers to create, share, and use custom AI code agents. It is available as extensions for both VS Code and JetBrains IDEs.

Key capabilities include:

• Agent mode: Collaborative development tasks with AI.
• Chat mode: General questions and code clarification.
• Edit mode: Inline code modifications within the current file.
• Autocomplete: Real-time inline code suggestions.
• Continue CLI (cn): Terminal-native AI coding assistance.

The @continuedev/hub-api@0.0.1 client is a TypeScript/JavaScript client generated to use the Fetch API.

Supported Environments

• Node.js
• Webpack
• Browserify

Language and Module Support

• Language Levels: ES5 (requires a Promises/A+ library) and ES6.
• Module Systems: CommonJS and ES6 module system.
• Typing: Works in both TypeScript and JavaScript. In TypeScript, definitions are automatically resolved via package.json.

Continue provides five core feature sets to enhance your workflow:

• Agent Mode: Equips the Chat model with tools to handle wide-ranging coding tasks.
• Chat Mode: Allows asking for help from an LLM without leaving the IDE.
• Plan Mode: A safe environment with read-only tools for exploring code and planning changes.
• Edit Mode: A convenient way to modify code without leaving your current file.
• Autocomplete: Provides inline code suggestions as you type.

The Continue plugin for JetBrains IDEs provides AI-powered development capabilities including an Agent, Chat, Edit, and Autocomplete.

Important Note: The JetBrains plugin is currently community-maintained. For the best experience and access to active development, it is strongly recommended to use the Continue CLI instead. The CLI provides the same agent capabilities, model access, and configuration in any environment without the limitations of the plugin framework.

A 'chat model' is an LLM trained for conversational interaction. In Continue, these models are primarily used for the standard Chat interface.

Important Fallback Behavior: If you do not explicitly specify a model for the edit or apply roles in your configuration, Continue will fall back to using your selected chat model for those tasks.

The automated documentation workflow uses the Continue CLI in headless mode to analyze code changes and update documentation. The process follows these steps:

1. Environment Setup: Validate environment, install CLI, and set CONTINUE_API_KEY.
2. Change Analysis: Generate git diff context to identify new functionality between branches.
3. Branch Creation: Create a new branch named {original-branch}-docs-update-{timestamp}.
4. AI Documentation Generation: Run Continue CLI with custom rules (via an agent configuration) to analyze changes and update documentation files.
5. Review & Commit: Review the AI-generated docs, commit them to the docs directory, and push to origin.
6. Cleanup: Remove temporary files and provide a summary.

Next Edit is an experimental, proactive AI-powered code prediction feature in Continue. Unlike traditional autocomplete which reacts to your typing, Next Edit analyzes your recent edits and coding patterns to suggest entire code modifications (diffs) across multiple lines before you start typing them.

Key Differences:

• Scope: Predicts entire code modifications vs. completing a single line/statement.
• Interaction: Displays diff overlays (before/after) vs. simple ghost text.
• Nature: Proactive (anticipates next steps) vs. Reactive (responds to input).
• Context: Analyzes recent edit patterns and broader context vs. immediate code completion.

A reranking model is a specialized, typically small and fast model trained to take two pieces of text (such as a user question and a document) and return a relevancy score between 0 and 1.

In Continue, you designate a model with the rerank role to improve codebase awareness. The reranker is used to select the most relevant code snippets from the results of an initial vector search, providing higher precision than vector search alone.