​
Introduction

Continue hub assistants are defined using the config.yaml specification. Assistants can be loaded from the Hub or locally
  • Continue Hub - YAML is stored on the hub and automatically synced to the extension
  • Locally
    • in your global .continue folder (~/.continue on Mac, %USERPROFILE%\.continue) within .continue/assistants. The name of the file will be used as the display name of the assistant, e.g. My Assistant.yaml
    • in your workspace in a /.continue/assistants folder, with the same naming convention
Config YAML replaces config.json, which is deprecated. View the Migration Guide.
An assistant is made up of:
  1. Top level properties, which specify the name, version, and config.yaml schema for the assistant
  2. Block lists, which are composable arrays of coding assistant building blocks available to the assistant, such as models, docs, and context providers.
A block is a single standalone building block of a coding assistants, e.g., one model or one documentation source. In config.yaml syntax, a block consists of the same top-level properties as assistants (name, version, and schema), but only has ONE item under whichever block type it is. Examples of blocks and assistants can be found on the Continue hub. Assistants can either explicitly define blocks - see Properties below - or import and configure existing hub blocks.

​
Using Blocks

Hub blocks and assistants are identified with a slug in the format owner-slug/block-or-assistant-slug, where an owner can be a user or organization (For example, if you want to use the OpenAI 4o Model block, your slug would be openai/gpt-4o). These blocks are pulled from https://hub.continue.dev. Blocks can be imported into an assistant by adding a uses clause under the block type. This can be alongside other uses clauses or explicit blocks of that type. For example, the following assistant imports an Anthropic model and defines an Ollama DeepSeek one. Assistant models section 
name: Team Agent
version: 1.0.0
schema: v1

models:
  - uses: anthropic/claude-3.5-sonnet # an imported model block
  - model: deepseek-reasoner # an explicit model block
    provider: ollama

​
Local Blocks

It is also possible to define blocks locally in a .continue folder. This folder can be located at either the root of your workspace (these will automatically be applied to all assistants when you are in that workspace) or in your home directory at ~/.continue (these will automatically be applied globally). Place your YAML files in the following folders: Assistants:
  • .continue/assistants - for assistants
Blocks:
  • .continue/rules - for rules
  • .continue/models - for models
  • .continue/prompts - for prompts
  • .continue/context - for context providers
  • .continue/docs - for docs
  • .continue/data - for data
  • .continue/mcpServers - for MCP Servers
You can find many examples of each of these block types on the Continue Explore Page
Local blocks utilizing mustache notation for secrets (${{ secrets.SECRET_NAME }}) can read secret values:
  • globally, from a .env located in the global .continue folder (~/.continue/.env)
  • per-workspace, from a .env file located at the root of the current workspace.

​
Inputs

Blocks can be passed user inputs, including hub secrets and raw text values. To create a block that has an input, use mustache templating as follows:
config.yaml
name: myprofile/custom-model
version: 1.0.0
schema: v1

models:
  - name: My Favorite Model
    provider: anthropic
    apiKey: ${{ inputs.ANTHROPIC_API_KEY }}
    defaultCompletionOptions:
      temperature: ${{ inputs.TEMP }}
Which can then be imported like this:
config.yaml
name: myprofile/custom-assistant
version: 1.0.0
schema: v1

models:
  - uses: myprofile/custom-model
    with:
      ANTHROPIC_API_KEY: ${{ secrets.MY_ANTHROPIC_API_KEY }}
      TEMP: 0.9
Note that hub secrets can be passed as inputs, using a similar mustache format: secrets.SECRET_NAME.

​
Overrides

Block properties can be also be directly overriden using override. For example:
config.yaml
name: myprofile/custom-assistant
version: 1.0.0
schema: v1

models:
  - uses: myprofile/custom-model
    with:
      ANTHROPIC_API_KEY: ${{ secrets.MY_ANTHROPIC_API_KEY }}
      TEMP: 0.9
    override:
      roles:
        - chat

​
Properties

Below are details for each property that can be set in config.yaml. All properties at all levels are optional unless explicitly marked as required. The top-level properties in the config.yaml configuration file are:

​
name

The name property specifies the name of your project or configuration.
config.yaml
name: MyProject

​
version

The version property specifies the version of your project or configuration.

​
schema

The schema property specifies the schema version used for the config.yaml, e.g. v1

​
models

The models section defines the language models used in your configuration. Models are used for functionalities such as chat, editing, and summarizing. Properties:
  • name (required): A unique name to identify the model within your configuration.
  • provider (required): The provider of the model (e.g., openai, ollama).
  • model (required): The specific model name (e.g., gpt-4, starcoder).
  • apiBase: Can be used to override the default API base that is specified per model
  • roles: An array specifying the roles this model can fulfill, such as chat, autocomplete, embed, rerank, edit, apply, summarize. The default value is [chat, edit, apply, summarize]. Note that the summarize role is not currently used.
  • capabilities: Array of strings denoting model capabilities, which will overwrite Continue’s autodetection based on provider and model. See the Model Capabilities guide for detailed information. Supported capabilities include:
    • tool_use: Enables function/tool calling support (required for Agent mode)
    • image_input: Enables image upload and processing support
    Continue automatically detects these capabilities for most models, but you can override this when using custom deployments or if autodetection isn’t working correctly.
  • maxStopWords: Maximum number of stop words allowed, to avoid API errors with extensive lists.
  • promptTemplates: Can be used to override the default prompt templates for different model roles. Valid values are chat, edit, apply and autocomplete. The chat property must be a valid template name, such as llama3 or anthropic.
  • chatOptions: If the model includes role chat, these settings apply for Agent and Chat mode:
    • baseSystemMessage: Can be used to override the default system prompt for Chat mode.
    • baseAgentSystemMessage: Can be used to override the default system prompt for Agent mode.
    • basePlanSystemMessage: Can be used to override the default system prompt for Plan mode.
  • embedOptions: If the model includes role embed, these settings apply for embeddings:
    • maxChunkSize: Maximum tokens per document chunk. Minimum is 128 tokens.
    • maxBatchSize: Maximum number of chunks per request. Minimum is 1 chunk.
  • defaultCompletionOptions: Default completion options for model settings.
    • contextLength: Maximum context length of the model, typically in tokens.
    • maxTokens: Maximum number of tokens to generate in a completion.
    • temperature: Controls the randomness of the completion. Values range from 0.0 (deterministic) to 1.0 (random).
    • topP: The cumulative probability for nucleus sampling.
    • topK: Maximum number of tokens considered at each step.
    • stop: An array of stop tokens that will terminate the completion.
    • reasoning: Boolean to enable thinking/reasoning for Anthropic Claude 3.7+ models.
    • reasoningBudgetTokens: Budget tokens for thinking/reasoning in Anthropic Claude 3.7+ models.
  • requestOptions: HTTP request options specific to the model.
    • timeout: Timeout for each request to the language model.
    • verifySsl: Whether to verify SSL certificates for requests.
    • caBundlePath: Path to a custom CA bundle for HTTP requests.
    • proxy: Proxy URL for HTTP requests.
    • headers: Custom headers for HTTP requests.
    • extraBodyProperties: Additional properties to merge with the HTTP request body.
    • noProxy: List of hostnames that should bypass the specified proxy.
    • clientCertificate: Client certificate for HTTP requests.
      • cert: Path to the client certificate file.
      • key: Path to the client certificate key file.
      • passphrase: Optional passphrase for the client certificate key file.
  • autocompleteOptions: If the model includes role autocomplete, these settings apply for tab autocompletion:
    • disable: If true, disables autocomplete for this model.
    • maxPromptTokens: Maximum number of tokens for the autocomplete prompt.
    • debounceDelay: Delay before triggering autocomplete in milliseconds.
    • modelTimeout: Model timeout for autocomplete requests in milliseconds.
    • maxSuffixPercentage: Maximum percentage of prompt allocated for suffix.
    • prefixPercentage: Percentage of input allocated for prefix.
    • template: Custom template for autocomplete using Mustache syntax. You can use the {{{ prefix }}}, {{{ suffix }}}, {{{ filename }}}, {{{ reponame }}}, and {{{ language }}} variables.
    • onlyMyCode: Only includes code within the repository for context.
    • useCache: If true, enables caching for completions.
    • useImports: If true, includes imports in context.
    • useRecentlyEdited: If true, includes recently edited files in context.
    • useRecentlyOpened: If true, includes recently opened files in context.
Example:
config.yaml
models:
  - name: GPT-4o
    provider: openai
    model: gpt-4o
    roles:
      - chat
      - edit
      - apply
    defaultCompletionOptions:
      temperature: 0.7
      maxTokens: 1500
  - name: Codestral
    provider: mistral
    model: codestral-latest
    roles:
      - autocomplete
    autocompleteOptions:
      debounceDelay: 250
      maxPromptTokens: 1024
      onlyMyCode: true
  - name: My Model - OpenAI-Compatible
    provider: openai
    apiBase: http://my-endpoint/v1
    model: my-custom-model
    capabilities:
      - tool_use
      - image_input
    roles:
      - chat
      - edit

​
context

The context section defines context providers, which supply additional information or context to the language models. Each context provider can be configured with specific parameters. More information about usage/params for each context provider can be found here Properties:
  • provider (required): The identifier or name of the context provider (e.g., code, docs, web)
  • name: Optional name for the provider
  • params: Optional parameters to configure the context provider’s behavior.
Example:
config.yaml
context:
  - provider: file
  - provider: code
  - provider: diff
  - provider: http
    name: Context Server 1
    params:
      url: "https://api.example.com/server1"
  - provider: terminal

​
rules

List of rules that the LLM should follow. These are concatenated into the system message for all Agent, Chat, and Edit requests. See the rules deep dive for details. Explicit rules can either be simple text or an object with the following properties:
  • name (required): A display name/title for the rule
  • rule (required): The text content of the rule
  • globs (optional): When files are provided as context that match this glob pattern, the rule will be included. This can be either a single pattern (e.g., "**/*.{ts,tsx}") or an array of patterns (e.g., ["src/**/*.ts", "tests/**/*.ts"]).
"config.yaml"l
rules:
  - Always annotate Python functions with their parameter and return types
  - name: TypeScript best practices
    rule: Always use TypeScript interfaces to define shape of objects. Use type aliases sparingly.
    globs: "**/*.{ts,tsx}"
  - name: TypeScript test patterns
    rule: In TypeScript tests, use Jest's describe/it pattern and follow best practices for mocking.
    globs:
      - "src/**/*.test.ts"
      - "tests/**/*.ts"
  - uses: myprofile/my-mood-setter
    with:
      TONE: concise

​
prompts

A list of custom prompts that can be invoked from the chat window. Each prompt has a name, description, and the actual prompt text.
config.yaml
prompts:
  - name: check
    description: Check for mistakes in my code
    prompt: |
      Please read the highlighted code and check for any mistakes. You should look for the following, and be extremely vigilant:
        - Syntax errors
        - Logic errors
        - Security vulnerabilities

​
docs

List of documentation sites to index. Properties:
  • name (required): Name of the documentation site, displayed in dropdowns, etc.
  • startUrl (required): Start page for crawling - usually root or intro page for docs
  • favicon: URL for site favicon (default is /favicon.ico from startUrl).
  • useLocalCrawling: Skip the default crawler and only crawl using a local crawler.
Example:
config.yaml
docs:
  - name: Continue
    startUrl: https://docs.continue.dev/intro
    favicon: https://docs.continue.dev/favicon.ico

​
mcpServers

The Model Context Protocol is a standard proposed by Anthropic to unify prompts, context, and tool use. Continue supports any MCP server with the MCP context provider. Properties:
  • name (required): The name of the MCP server.
  • command (required): The command used to start the server.
  • args: An optional array of arguments for the command.
  • env: An optional map of environment variables for the server process.
  • cwd: An optional working directory to run the command in. Can be absolute or relative path.
  • connectionTimeout: An optional connection timeout number to the server in milliseconds.
Example:
config.yaml
mcpServers:
  - name: My MCP Server
    command: uvx
    args:
      - mcp-server-sqlite
      - --db-path
      - ./test.db
    cwd: /Users/NAME/project
    env:
      NODE_ENV: production

​
data

Destinations to which development data will be sent. Properties:
  • name (required): The display name of the data destination
  • destination (required): The destination/endpoint that will receive the data. Can be:
    • an HTTP endpoint that will receive a POST request with a JSON blob
    • a file URL to a directory in which events will be dumpted to .jsonl files
  • schema (required): the schema version of the JSON blobs to be sent. Options include 0.1.0 and 0.2.0
  • events: an array of event names to include. Defaults to all events if not specified.
  • level: a pre-defined filter for event fields. Options include all and noCode; the latter excludes data like file contents, prompts, and completions. Defaults to all
  • apiKey: api key to be sent with request (Bearer header)
  • requestOptions: Options for event POST requests. Same format as model requestOptions.
Example:
config.yaml
data:
  - name: Local Data Bank
    destination: file:///Users/dallin/Documents/code/continuedev/continue-extras/external-data
    schema: 0.2.0
    level: all
  - name: My Private Company
    destination: https://mycompany.com/ingest
    schema: 0.2.0
    level: noCode
    events:
      - autocomplete
      - chatInteraction

​
Complete YAML Config Example

Putting it all together, here’s a complete example of a config.yaml configuration file:
config.yaml
name: MyProject
version: 0.0.1
schema: v1
models:
  - uses: anthropic/claude-3.5-sonnet
    with:
      ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}
    override:
      defaultCompletionOptions:
        temperature: 0.8
  - name: GPT-4
    provider: openai
    model: gpt-4
    roles:
      - chat
      - edit
    defaultCompletionOptions:
      temperature: 0.5
      maxTokens: 2000
    requestOptions:
      headers:
        Authorization: Bearer YOUR_OPENAI_API_KEY
  - name: Ollama Starcoder
    provider: ollama
    model: starcoder
    roles:
      - autocomplete
    autocompleteOptions:
      debounceDelay: 350
      maxPromptTokens: 1024
      onlyMyCode: true
    defaultCompletionOptions:
      temperature: 0.3
      stop:
        - "\n"
rules:
  - Give concise responses
  - Always assume TypeScript rather than JavaScript
prompts:
  - name: test
    description: Unit test a function
    prompt: |
      Please write a complete suite of unit tests for this function. You should use the Jest testing framework.
      The tests should cover all possible edge cases and should be as thorough as possible.
      You should also include a description of each test case.
  - uses: myprofile/my-favorite-prompt
context:
  - provider: diff
  - provider: file
  - provider: code
mcpServers:
  - name: DevServer
    command: npm
    args:
      - run
      - dev
    env:
      PORT: "3000"
data:
  - name: My Private Company
    destination: https://mycompany.com/ingest
    schema: 0.2.0
    level: noCode
    events:
      - autocomplete
      - chatInteraction

​
Using YAML anchors to avoid config duplication

You can also use node anchors to avoid duplication of properties. To do so, adding the YAML version header %YAML 1.1 is needed, here’s an example of a config.yaml configuration file using anchors:
config.yaml
%YAML 1.1
---
name: MyProject
version: 0.0.1
schema: v1
model_defaults: &model_defaults
  provider: openai
  apiKey: my-api-key
  apiBase: https://api.example.com/llm
models:
  - name: mistral
    <<: *model_defaults
    model: mistral-7b-instruct
    roles:
      - chat
      - edit
  - name: qwen2.5-coder-7b-instruct
    <<: *model_defaults
    model: qwen2.5-coder-7b-instruct
    roles:
      - chat
      - edit
  - name: qwen2.5-coder-7b
    <<: *model_defaults
    model: qwen2.5-coder-7b
    useLegacyCompletionsEndpoint: false
    roles:
      - autocomplete
    autocompleteOptions:
      debounceDelay: 350
      maxPromptTokens: 1024
      onlyMyCode: true