API Reference

Complete API reference for the AgenticWork SDK.

AgenticWork Class

The main client class for interacting with AgenticWork.

Constructor

new AgenticWork(config: AgenticWorkConfig)

Parameters:

Name	Type	Description
apiKey	string	Your AgenticWork API key
baseUrl	string	API base URL (default: https://chat.agenticwork.io)
timeout	number	Request timeout in ms (default: 30000)

Example:

const client = new AgenticWork({
  apiKey: 'aw_live_xxxx',
  baseUrl: 'https://chat.agenticwork.io',
  timeout: 60000
});

Methods

init()

Initialize the client and verify connection.

await client.init(): Promise<void>

createAgent()

Create a new AI agent instance.

client.createAgent(config: AgentConfig): Agent

AgentConfig:

Name	Type	Description
name	string	Agent identifier
model	string	Model to use (e.g., claude-3-5-sonnet)
intelligenceLevel	number	0-100 cost/quality slider
systemPrompt	string	Optional system prompt
tools	string[]	Allowed MCP tools

getMCPTools()

Get list of available MCP tools.

await client.getMCPTools(): Promise<MCPTool[]>

Returns: Array of available tools with their schemas.

---

Agent Class

Represents a configured AI agent.

complete()

Get a complete response (non-streaming).

await agent.complete(request: CompletionRequest): Promise<CompletionResponse>

CompletionRequest:

Name	Type	Description
messages	Message[]	Conversation messages
tools	string[]	Tools to enable
maxTokens	number	Max response tokens

CompletionResponse:

Name	Type	Description
content	string	Response text
toolCalls	ToolCall[]	Any tool invocations
usage	Usage	Token usage stats

streamComplete()

Stream response tokens.

agent.streamComplete(request: CompletionRequest): AsyncIterable<StreamChunk>

StreamChunk:

Name	Type	Description
type	string	text, tool_use, done
content	string	Chunk content
toolCall	ToolCall	Tool call data (if applicable)

Example:

for await (const chunk of agent.streamComplete({
  messages: [{ role: 'user', content: 'Hello' }]
})) {
  if (chunk.type === 'text') {
    process.stdout.write(chunk.content);
  } else if (chunk.type === 'tool_use') {
    console.log('Tool called:', chunk.toolCall);
  }
}

---

Types

Message

interface Message {
  role: 'user' | 'assistant' | 'system';
  content: string;
}

ToolCall

interface ToolCall {
  id: string;
  name: string;
  input: Record<string, unknown>;
  output?: unknown;
}

Usage

interface Usage {
  inputTokens: number;
  outputTokens: number;
  totalTokens: number;
  estimatedCost: number;
}

MCPTool

interface MCPTool {
  name: string;
  description: string;
  inputSchema: JSONSchema;
  server: string;
}

---

Error Handling

import { AgenticWorkError, RateLimitError, AuthenticationError } from '@agentic-work/sdk';

try {
  await agent.complete({ messages });
} catch (error) {
  if (error instanceof RateLimitError) {
    console.log('Rate limited, retry after:', error.retryAfter);
  } else if (error instanceof AuthenticationError) {
    console.log('Invalid API key');
  } else if (error instanceof AgenticWorkError) {
    console.log('API error:', error.message);
  }
}