Continue

GitHub: continuedev/continue Docs: docs.continue.dev License: Apache 2.0

[Source: docs.continue.dev, github.com/continuedev/continue, 2026-05-03]

Key Facts

• VS Code extension: actively maintained by Continue team
• JetBrains plugin: community-maintained (as of 2025)
• Model agnostic: any model via config — Anthropic, OpenAI, Gemini, local Ollama, Azure OpenAI
• Context providers: modular system for injecting context (files, docs, web, terminal, codebase)
• Autocomplete: multi-line tab completion with configurable model
• Chat: sidebar chat with context awareness (cmd/ctrl+L VS Code, cmd/ctrl+J JetBrains)
• Config format: ~/.continue/config.json (JSON) or config.yaml

Configuration

All configuration lives in ~/.continue/config.json. The file controls models, context providers, rules, and slash commands.

{
  "models": [
    {
      "title": "Claude Sonnet",
      "provider": "anthropic",
      "model": "claude-sonnet-4-6",
      "apiKey": "$ANTHROPIC_API_KEY"
    },
    {
      "title": "Ollama Llama 3",
      "provider": "ollama",
      "model": "llama3"
    }
  ],
  "tabAutocompleteModel": {
    "title": "Starcoder2",
    "provider": "ollama",
    "model": "starcoder2:3b"
  },
  "contextProviders": [
    { "name": "code" },
    { "name": "docs" },
    { "name": "diff" },
    { "name": "terminal" },
    { "name": "problems" },
    { "name": "folder" },
    { "name": "codebase" }
  ],
  "slashCommands": [
    { "name": "edit", "description": "Edit selected code" },
    { "name": "comment", "description": "Write comments for selected code" },
    { "name": "share", "description": "Export conversation" }
  ],
  "rules": "Always use TypeScript. Prefer functional components. Never use `any`."
}

Context Providers

Context providers are the core differentiator. They let you @mention specific context sources in the chat.

The @codebase provider builds an embedding index of the repo locally — the closest equivalent to Cursor's codebase context.

Usage Patterns

Chat with file context

@src/models/order.py why does the status field allow null?

Inline edit (cmd+I)

Select code → cmd+I → describe the change. Continue rewrites the selection and shows a diff.

Tab autocomplete

Continue shows multi-line completions as you type. Tab to accept, Escape to dismiss. Uses a dedicated smaller model (e.g., Starcoder2 via Ollama) to keep latency low.

Slash commands

/edit add error handling to this function
/comment explain what this function does
/share export this conversation as markdown

Custom slash commands can run any prompt template against the selected context.

Comparison with Other AI Coding Tools

Continue is the right choice when:

• You want full control over which model powers which task (different models for chat vs autocomplete)
• Your team needs to work air-gapped or with local models
• You want to use Claude or another non-OpenAI model for coding assistance
• You are an existing JetBrains shop and want Copilot-style assistance

Continue is not the right choice when:

• You need deep autonomous multi-file editing (use Claude Code or Cursor)
• You want the highest-quality codebase awareness without configuration (use Cursor)
• You're running one-shot tasks from the terminal (use Aider or Claude Code)

Docs Indexing

Continue can index any documentation site and make it available as @docs:

{
  "contextProviders": [
    {
      "name": "docs",
      "params": {
        "sites": [
          { "startUrl": "https://docs.pydantic.dev/latest/", "title": "Pydantic" },
          { "startUrl": "https://fastapi.tiangolo.com/", "title": "FastAPI" },
          { "startUrl": "https://docs.anthropic.com/", "title": "Anthropic" }
        ]
      }
    }
  ]
}

Then use @Pydantic how do I use model_validator? in chat.

Common Failure Cases

@codebase returns irrelevant results for large repos Why: the local embedding index quality depends on the embedding model and chunk size; large repos with many similar files produce noisy retrieval. Detect: @codebase mentions clearly wrong files in its response. Fix: configure a better embedding model in config.json (e.g., voyage-code-2 via Voyage AI instead of the default); add .continueignore to exclude test fixtures, generated code, and vendored libraries.

Autocomplete latency too high Why: the default or configured chat model is used for autocomplete — large models have too much latency for inline suggestions. Detect: completions appear more than 1–2 seconds after stopping typing. Fix: configure a dedicated tabAutocompleteModel using a small fast model (Starcoder2 3B via Ollama, or Codestral via Mistral API).

JetBrains plugin out of date with VS Code features Why: the JetBrains plugin is community-maintained; new features land in VS Code first and may lag by weeks or months. Detect: a feature described in docs.continue.dev doesn't appear in JetBrains. Fix: check the JetBrains plugin changelog; the VS Code extension is the reference implementation.

Connections

• ai-tools/claude-code — autonomous agentic coding from the terminal; complement to Continue for multi-file changes
• ai-tools/cursor-copilot — Cursor has deeper codebase indexing; Continue has more model flexibility
• ai-tools/aider — terminal-based, similar model agnosticism, stronger for autonomous commits
• protocols/mcp — Continue has MCP support via config; can expose local tools to the model

Open Questions

• Does Continue support MCP servers as context providers in config.json, and if so, what is the stability of that integration?
• How does @codebase embedding quality compare to Cursor's proprietary codebase index on a 100k+ line repo?