English

简体中文

English

简体中文

Local Model Configuration ​

Local agents are managed through YAML configuration files. The user-level configuration file is located at:

  • macOS / Linux: ~/.vjsp/config.yaml
  • Windows: %USERPROFILE%\.vjsp\config.yaml

To edit this file, click the settings icon next to “Local Config” in the configuration dropdown menu at the top-right corner of the IDE chat input box. This will open config.yaml.

Agent = Models + Rules + MCP Servers

Creating a Local Model ​

  1. In the IDE, select Local Agent, then click the settings button next to it to open config.yaml.
    Add prompt example

  2. Add your model configuration to the file and save it to activate the settings.

yaml
   - name: Qwen3-Coder-30B-A3B-Instruct
    provider: openai
    model: Qwen3-Coder-30B-A3B-Instruct
    apiBase: http://www.example.com/v1
    capabilities:
      - tool_use
    roles:
      - chat
      - edit
	  - apply

Top-Level Structure of the Configuration File ​

Below is a description of all configurable top-level properties in config.yaml. Unless explicitly marked as required, all properties are optional.

Top-level attributes in config.yaml include:

yaml
name: My Config                # Required: Configuration name
version: 1.0.0                # Required: Version number
schema: v1                    # Required: Schema version
models:                       # Optional: Model definitions
context:                      # Optional: Context providers
rules:                        # Optional: System rules
prompts:                      # Optional: Invocable prompts
docs:                         # Optional: Documentation site indexes
mcpServers:                   # Optional: MCP tool servers
data:                         # Optional: Development data reporting destinations

models ​

This section defines the language models used in the configuration. Models can enable functionalities such as conversation, code editing, and content summarization.

Attribute Reference:

  • name (Required): A unique identifier for the model within the configuration.

  • provider (Required): The model provider (e.g., openai).

  • model (Required): The specific model name (e.g., Qwen3-Coder-30B-A3B-Instruct).

  • apiBase: Overrides the default base API URL for the model.

  • roles: An array specifying the roles the model can assume, including chat (conversation), autocomplete (code completion), embed (embedding), rerank (re-ranking), edit (editing), and apply (execution). Default value is [chat, edit, apply].

  • capabilities: A string array declaring model capabilities, overriding auto-detected capabilities based on provider and model.

  • tool_use: Enables function/tool calling capability (required for Agent mode).

  • image_input: Enables image upload and processing.

    Most model capabilities are auto-detected. Manual override is useful for custom deployments or when auto-detection fails.

  • maxStopWords: Maximum number of stop words allowed to prevent API errors due to overly long lists.

  • promptTemplates: Overrides default prompt templates for different model roles. Supported values include chat, edit, apply, and autocomplete. The chat property must reference a valid template name (e.g., llama3 or anthropic).

  • chatOptions: Applied if the model includes the chat role. Settings affect both Chat and Agent modes:

    • baseSystemMessage: Overrides the default system prompt for Chat mode.

    • basePlanSystemMessage: Overrides the default system prompt for Plan mode.

    • baseAgentSystemMessage: Overrides the default system prompt for Agent mode.

  • embedOptions: Applied if the model includes the embed role:

    • maxChunkSize: Maximum token count per document chunk (minimum: 128 tokens).

    • maxBatchSize: Maximum number of chunks per request (minimum: 1).

  • defaultCompletionOptions: Default generation settings for the model:

    • contextLength: Maximum context length (typically in tokens).

    • maxTokens: Maximum tokens generated per request.

    • temperature: Controls randomness (0.0 = deterministic, 1.0 = random).

    • topP: Nucleus sampling cumulative probability threshold.

    • topK: Number of top tokens considered at each step.

    • stop: Array of stop sequences to terminate generation.

  • requestOptions: HTTP request settings specific to the model:

    • timeout: Timeout per LLM request.

    • verifySsl: Whether to verify SSL certificates.

    • caBundlePath: Path to a custom CA bundle.

    • proxy: Proxy URL for HTTP requests.

    • headers: Custom HTTP headers.

    • extraBodyProperties: Additional properties merged into the request body.

    • noProxy: Hostnames bypassing the proxy.

    • clientCertificate: Client certificate for HTTPS requests:

      • cert: Path to certificate file.

      • key: Path to private key file.

      • passphrase: Optional passphrase for the key.

  • autocompleteOptions: Applied if the model includes the autocomplete role:

    • disable: If true, disables autocomplete for this model.

    • maxPromptTokens: Max tokens in the autocomplete prompt.

    • debounceDelay: Delay before triggering autocomplete (ms).

    • modelTimeout: Model request timeout for autocomplete (ms).

    • maxSuffixPercentage: Max percentage of suffix in prompt.

    • prefixPercentage: Percentage allocated to prefix.

    • transform: If false, disables trimming of multi-line completions (default: true). Useful for models that generate better multi-line completions without post-processing.

    • template: Custom Mustache template using variables:

      {{{ prefix }}}, {{{ suffix }}}, {{{ filename }}}, {{{ reponame }}}, {{{ language }}}

    • onlyMyCode: If true, only includes code from the current repository.

    • useCache: If true, enables caching of completions.

    • useImports: If true, includes import statements in context.

    • useRecentlyEdited: If true, includes recently edited files.

    • useRecentlyOpened: If true, includes recently opened files.

Example:

yaml
name: My Config
version: 1.0.0
schema: v1
models:
  - name: Qwen3-Coder-30B-A3B-Instruct
    provider: openai
    model: Qwen3-Coder-30B-A3B-Instruct
	apiBase: http://www.example.com/v1
	capabilities:
	  - tool_use
    roles:
      - chat
      - edit
      - apply
    defaultCompletionOptions:
      temperature: 0.7
      maxTokens: 1500
  - name: Qwen2-5-VL-32B-Instruct
    provider: openai
    model: Qwen2-5-VL-32B-Instruct
	apiBase: http://www.example.com/v1
	capabilities:
	  - tool_use
	  - image_input
    roles:
      - chat
      - edit
      - apply
    defaultCompletionOptions:
      temperature: 0.3

context ​

Defines context providers that supply additional information to language models. Each provider can have custom parameters.

Attribute Reference:

  • provider (Required): Identifier of the context provider (e.g., code, docs, web).

  • name: Optional display name for the provider.

  • params: Optional parameters to configure provider behavior.

Example:

yaml
name: My Config
version: 1.0.0
schema: v1
context:
  - provider: file
  - provider: code
  - provider: diff
  - provider: http
    name: Context Server
    params:
      url: "https://api.example.com/server"
  - provider: terminal

rules ​

Rules are appended to the system prompt in all Agent, Chat, and Edit mode requests.

Configuration Example:

yaml
name: My Config
version: 1.0.0
schema: v1
rules:
  - uses: sanity/sanity-opinionated # Rule file stored in Task Control Center
  - uses: file://user/Desktop/rules.md # Local rule file

Rule File Example:

md
---
name: Language Style Rule (Pirate rule)
---

Respond in Chinese

See Rules Deep Dive for more details.

prompts ​

Prompts can be invoked using the / command.

Configuration Example:

yaml
name: My Config
version: 1.0.0
schema: v1
prompts:
  - uses: supabase/create-functions # Prompt stored in Task Control Center
  - uses: file://user/Desktop/prompts.md # Local prompt file

Prompt File Example (prompts.md):

md
---
name: Generate Chinese Comments
invokable: true
---

Rewrite all comments in the current file into Chinese

See Prompts Deep Dive for more details.

docs ​

Specifies documentation sites to be indexed.

Attribute Reference:

  • name (Required): Display name of the documentation site.

  • startUrl (Required): Starting URL for the crawler (usually the root or intro page).

  • favicon: URL to the site’s favicon (defaults to /favicon.ico under startUrl).

  • useLocalCrawling: If true, skips the default crawler and uses only local crawling.

Example:

yaml
name: My Config
version: 1.0.0
schema: v1
docs:
  - name: VJSP Official Docs
    startUrl: https://docs.VJSP.dev/intro
    favicon: https://docs.VJSP.dev/favicon.ico

mcpServers ​

The Model Context Protocol (MCP), proposed by Anthropic, standardizes prompts, context, and tool usage. All MCP servers are supported via MCP context providers.

Attribute Reference:

  • name (Required): Name of the MCP server.

  • command (Required): Command to launch the server.

  • args: Optional command arguments.

  • env: Optional environment variables for the server process.

  • cwd: Optional working directory (absolute or relative path).

  • requestOptions: Optional request settings for SSE/HTTP servers (same format as model requestOptions).

  • connectionTimeout: Optional initial connection timeout.

Example:

yaml
name: My Config
version: 1.0.0
schema: v1
mcpServers:
  - name: My MCP Server
    command: uvx
    args:
      - mcp-server-sqlite
      - --db-path
      - ./test.db
    cwd: /Users/NAME/project
    env:
      NODE_ENV: production

data ​

Configures destinations for development data reporting.

Attribute Reference:

  • name (Required): Display name of the data destination.

  • destination (Required): Endpoint receiving data, supporting two types:

    • HTTP endpoint: Receives POST requests with JSON data.

    • File URL: Directory path where events are stored as .jsonl files.

  • schema (Required): Schema version for JSON data (0.1.0 or 0.2.0).

  • events: Array of event names to include (defaults to all if unspecified).

  • level: Predefined field filter level:

    • all: Includes all fields.
    • noCode: Excludes code-related data (file content, prompts, generations).
      Default: all.

  • apiKey: API key sent via Bearer header.

  • requestOptions: Request settings for POST events (same format as model requestOptions).

Example:

yaml
name: My Config
version: 1.0.0
schema: v1
data:
  - name: Local Data Repository
    destination: file:///Users/dallin/Documents/code/VJSPdev/VJSP-extras/external-data
    schema: 0.2.0
    level: all
  - name: Enterprise Private Endpoint
    destination: https://mycompany.com/ingest
    schema: 0.2.0
    level: noCode
    events:
      - autocomplete

Complete YAML Configuration Example ​

Below is a full config.yaml integrating all components:

yaml
name: My Config
version: 1.0.0
schema: v1
models:
  - name: Qwen3-Coder-30B-A3B-Instruct
    provider: openai
    model: Qwen3-Coder-30B-A3B-Instruct
    roles:
      - chat
      - edit
    defaultCompletionOptions:
      temperature: 0.5
      maxTokens: 2000
    requestOptions:
      headers:
        Authorization: Bearer YOUR_OPENAI_API_KEY
  - name: Qwen2-5-VL-32B-Instruct
    provider: openai
    model: Qwen2-5-VL-32B-Instruct
    roles:
      - autocomplete
    autocompleteOptions:
      debounceDelay: 350
      maxPromptTokens: 1024
      onlyMyCode: true
    defaultCompletionOptions:
      temperature: 0.3
      stop:
        - "\n"
rules:
  - Responses must be concise and clear.
  - Prefer TypeScript over JavaScript by default.
prompts:
  - name: Unit Test Generator
    description: Generate unit tests for functions
    prompt: |
      Write a complete unit test suite for this function using Jest.
      Cover all edge cases thoroughly, and add explanations for each test case.
  - uses: myprofile/my-favorite-prompt
context:
  - provider: diff
  - provider: file
  - provider: code
mcpServers:
  - name: Dev Server
    command: npm
    args:
      - run
      - dev
    env:
      PORT: "3000"
data:
  - name: Enterprise Private Endpoint
    destination: https://mycompany.com/ingest
    schema: 0.2.0
    level: noCode
    events:
      - autocomplete
      - chatInteraction

Using YAML Anchors to Avoid Repetition ​

yaml
%YAML 1.1
---
name: My Config
version: 1.0.0
schema: v1
model_defaults: &model_defaults
  provider: openai
  apiKey: my-api-key
  apiBase: https://api.example.com/llm
models:
  - 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