The Continue.dev config.json Guide

5 min·Jul 18, 2026

AI Summary

Continue.dev config.json wires up its assistant. models is the list of chat models; tabAutocompleteModel is a separate fast model for inline completions; embeddingsProvider powers @codebase retrieval; contextProviders add sources like code, diff and docs; slashCommands add custom /commands. Use a small model for autocomplete, reference API keys as secrets, and prefer local providers for sensitive code.

Continue.dev config.json

A Continue.dev config.json wiring up a chat model, a tab-autocomplete model, an embeddings provider, context providers and a custom slash command — the setup for the open-source AI code assistant.

{
  "models": [
    {
      "title": "Claude Sonnet",
      "provider": "anthropic",
      "model": "claude-sonnet-4-5",
      "apiKey": "${{ secrets.ANTHROPIC_API_KEY }}"
    }
  ],
  "tabAutocompleteModel": {
    "title": "Codestral",
    "provider": "mistral",
    "model": "codestral-latest",
    "apiKey": "${{ secrets.MISTRAL_API_KEY }}"
  },
  "embeddingsProvider": {
    "provider": "openai",
    "model": "text-embedding-3-small",
    "apiKey": "${{ secrets.OPENAI_API_KEY }}"
  },
  "contextProviders": [
    {
      "name": "code"
    },
    {
      "name": "diff"
    },
    {
      "name": "docs"
    },
    {
      "name": "terminal"
    }
  ],
  "slashCommands": [
    {
      "name": "edit",
      "description": "Edit selected code"
    },
    {
      "name": "comment",
      "description": "Write comments for the selected code"
    }
  ]
}

Continue.dev is an open-source AI assistant for VS Code and JetBrains, and config.json is where you tell it which models to use and where to pull context from.

Chat models vs autocomplete

models lists the models available in chat:

"models": [
  { "title": "Claude Sonnet", "provider": "anthropic", "model": "claude-sonnet-4-5", "apiKey": "${{ secrets.ANTHROPIC_API_KEY }}" }
]

tabAutocompleteModel is deliberately separate — inline completion needs to be fast, so people pair a strong chat model with a small one like Codestral for typing.

Embeddings power @codebase

embeddingsProvider builds the index behind @codebase retrieval, so the assistant can answer questions about your whole repo, not just the open file.

Context providers

contextProviders are the @ sources you can pull in:

"contextProviders": [
  { "name": "code" }, { "name": "diff" }, { "name": "docs" }, { "name": "terminal" }
]

Add only the ones you use — each extra provider is more noise in the prompt.

Custom slash commands

slashCommands adds / actions in chat, like /edit or a project-specific /test.

Security notes

Reference API keys as ${{ secrets.NAME }}, never inline.
Local providers such as Ollama keep code on your machine — prefer them for sensitive repos.
Vet third-party context providers before enabling them.

Open the Continue.dev config template and ask the workspace to "add an Ollama model for local inference" or "add the codebase context provider", then diff the result.

FAQ

Why separate chat and autocomplete models?

Chat wants a strong, higher-latency model; inline autocomplete needs to be near-instant. Continue lets you pick a big model for chat and a small fast one (like Codestral) for tabAutocompleteModel.

JSON or YAML?

Newer Continue versions favor config.yaml, but config.json remains supported and is what many existing setups use. The field names map directly between the two.

Related templates

Continue.dev config.json