Configuration · docmd assistant

docmd-assistant accepts a flexible AssistantOptions configuration object upon initialisation. All settings can also be updated dynamically at runtime.

Options Schema Reference

interface AssistantOptions {
  provider?: string;
  model?: string;
  apiKey?: string;
  baseURL?: string;
  relayUrl?: string;
  endpoint?: string;
  projectId?: string;
  systemPrompt?: string;
  history?: ChatMessage[];
  tools?: AssistantTool[];
  temperature?: number;
  maxTokens?: number;
  reasoning?: boolean | 'none' | 'low' | 'medium' | 'high';
  headers?: Record<string, string>;
  contextWindow?: number;
}

Parameter Details

Field	Type	Description	Default
`provider`	`string`	Target AI provider (`'openai'`, `'anthropic'`, `'gemini'`, `'deepseek'`, `'groq'`, `'minimax'`, `'ollama'`)	Dynamic
`model`	`string`	Target model identifier (e.g. `'gpt-4o-mini'`, `'claude-3-5-haiku-20241022'`)	Dynamic
`apiKey`	`string`	Provider API key for direct connectivity via `aiplug`	`undefined`
`baseURL`	`string`	Custom base API gateway URL	`undefined`
`relayUrl`	`string`	Cloud Relay endpoint URL for keyless proxy routing	`undefined`
`endpoint`	`string`	Alias for `relayUrl`	`'https://api.docmd.io/v1/ai/chat'`
`projectId`	`string`	Project or site identifier sent with relay requests	`undefined`
`systemPrompt`	`string`	Base instructions guiding assistant identity and behaviour	System default prompt
`history`	`ChatMessage[]`	Pre-populated conversation history	`[]`
`tools`	`AssistantTool[]`	Initial array of registered tools	`[]`
`temperature`	`number`	Sampling temperature (0.0 to 1.0)	Provider default
`maxTokens`	`number`	Maximum tokens returned per response turn	Provider default
`reasoning`	`boolean \| string`	Toggle extended reasoning mode (`false`, `'low'`, `'medium'`, `'high'`)	`false`
`headers`	`Record<string, string>`	Custom HTTP headers sent with relay requests	`{}`
`contextWindow`	`number`	Maximum characters of retrieved documentation context passed to the model per turn. `0` = uncapped (full content delivered for highest response fidelity).	`0` (uncapped)

Default System Prompt

If no systemPrompt is provided, the engine applies a default prompt that enforces docmd assistant identity rules, search-first tool calling, concise Markdown output, and clickable citation links.

Updating Options at Runtime

Modify configuration options dynamically during an active session using updateOptions():

assistant.updateOptions({
  provider: 'anthropic',
  model: 'claude-3-5-haiku-20241022',
  temperature: 0.2
});

System Prompt Management

docmd-assistant provides dedicated methods to update or append system instructions:

// Replace system prompt completely
assistant.setSystemPrompt('You are a technical support specialist for a cloud developer platform.');

// Append additional context or instructions
assistant.appendSystemPrompt('Always respond using British English spelling and provide step-by-step code snippets.');

// Retrieve current active system prompt
const currentPrompt = assistant.getSystemPrompt();

Reasoning Mode Support

For models supporting extended reasoning (such as DeepSeek-R1 or OpenAI o3-mini), set the reasoning option to control reasoning depth:

const assistant = new DocmdAssistantEngine({
  provider: 'deepseek',
  model: 'deepseek-reasoner',
  apiKey: process.env.DEEPSEEK_API_KEY,
  reasoning: 'medium'
});

Custom Relay Headers

Pass custom headers when routing turns through enterprise API gateways or authenticated relay endpoints:

const assistant = new DocmdAssistantEngine({
  relayUrl: 'https://internal-ai-gateway.company.com/v1/chat',
  headers: {
    'Authorization': 'Bearer my_enterprise_token',
    'X-Custom-Tenant-ID': 'tenant_12345'
  }
});