Continue capabilities

An autonomous coding assistant that reads files, makes multi-file changes, runs terminal commands, and handles complex multi-step tasks from a single natural-language instruction. Before each tool use the agent requests approval by default; users click Continue to approve or Cancel to reject.

The Continue CLI (cn) runs the AI agent from the terminal in interactive TUI mode or non-interactive headless mode, enabling scripted automation, CI/CD integration, and background coding tasks without an IDE.

In chat and agent mode, users type @ to attach specific context sources — files, code symbols, terminal output, git diffs, and more — so the model has precise, relevant information without pasting manually.

Users supply their own API keys for any supported model provider, or run models locally. There is no locked-in model — every provider and model is user-configurable via config.yaml.

Continue ships as a native extension for Visual Studio Code and as a plugin for JetBrains IDEs. Both surfaces provide autocomplete, chat, inline edit, and agent mode within the editor.

An interactive AI assistant panel embedded in the IDE for asking questions, analyzing code, and getting guidance without leaving the editor. Supports three sub-modes: Chat (no tools), Plan Mode (read-only tools), and Agent Mode (full tool access).

Select a code region, press Cmd/Ctrl+I, describe the change in natural language, and a diff is streamed inline. Designed for focused, single-function changes such as refactoring, adding comments, generating tests, or converting between languages.

Provides AI-powered inline code suggestions as you type, which you accept with Tab. Supports single-line and multi-line completions triggered automatically or manually.

Continue supports the Model Context Protocol standard, allowing any MCP server to be added as a tool extension for the AI agent. This enables integration with external data sources, APIs, databases, and custom tooling.

Rules are persistent instructions injected into the system message for every Agent, Chat, and Edit request. Prompts are reusable slash-command templates invoked with a / prefix for specific tasks.