Class Agent

The core Agent class for creating and managing AI agents.

An Agent combines models, skills, and behaviors to create intelligent assistants that can process prompts, maintain conversations, and execute tasks.

Example

// Create a simple agent
const agent = new Agent({
  name: "Assistant",
  model: "gpt-4",
  behavior: "You are a helpful assistant."
});

// Use the agent
const response = await agent.prompt("Hello, how can you help me?");
console.log(response);

// Add skills to the agent
agent.addSkill({
  name: "calculator",
  description: "Perform mathematical calculations",
  process: (a, b) => a + b
});

Hierarchy (View Summary)

SDKObject
- Agent

Index

Constructors

constructor

new Agent(_settings: TAgentSettings): Agent
Create a new Agent instance.
Parameters
- _settings: TAgentSettings
  Configuration object for the agent
Returns Agent
Example
```
const agent = new Agent({
  name: "Data Analyst",
  model: "gpt-4",
  behavior: "You are an expert data analyst who provides insights."
});
```
Overrides SDKObject.constructor
- Defined in src/Agent/Agent.class.ts:332

Properties

structure

structure: { components: any[]; connections: any[] } = ...

The agent internal structure used for by internal operations to generate the agent data

Accessors

data

get data(): AgentData
The agent data : this is the equivalent of the .smyth file content.

Used for by external operations to get the agent data

Returns AgentData
- Defined in src/Agent/Agent.class.ts:299

llm

get llm(): TLLMProviderInstances

Access to LLM instances for direct model interactions.

Supported providers and calling patterns:

agent.llm.openai(modelId, params) - OpenAI models
agent.llm.anthropic(modelId, params) - Anthropic models

Returns TLLMProviderInstances

Example

// Direct model access
const gpt4 = agent.llm.openai('gpt-4', { temperature: 0.7 });
const response = await gpt4.prompt("Explain quantum computing");

// Using configuration object
const claude = agent.llm.anthropic({
  model: 'claude-3-sonnet',
  maxTokens: 1000
});

// Streaming response
const stream = await claude.prompt("Write a poem").stream();
stream.on('data', chunk => console.log(chunk));

ready

get ready(): ControlledPromise<any>
Returns ControlledPromise<any>
Inherited from SDKObject.ready
- Defined in src/Core/SDKObject.class.ts:19

storage

get storage(): TStorageProviderInstances
Returns TStorageProviderInstances
- Defined in src/Agent/Agent.class.ts:494

team

get team(): Team
Returns Team
- Defined in src/Agent/Agent.class.ts:287

vectorDB

get vectorDB(): TVectorDBProviderInstances
Returns TVectorDBProviderInstances
- Defined in src/Agent/Agent.class.ts:528

Methods

addSkill

addSkill(
    settings?: TSkillSettings,
): {
    in: (inputs: TSkillInputs) => void;
    out: { body: any; headers: any; query: any; [key: string]: any };
}
Add a skill to the agent, enabling it to perform specific tasks or operations.

Skills extend the agent's capabilities by providing custom functions that can be called during conversations or prompt processing.

A skill can be implemented in two ways:
1. With a process function that defines the skill's core logic
2. As a workflow entry point that can be connected to other components to build complex logic
Parameters
- Optionalsettings: TSkillSettings
Returns {
in: (inputs: TSkillInputs) => void;
out: { body: any; headers: any; query: any; [key: string]: any };
}
Example
```
// Add a data fetching skill
agent.addSkill({
  name: "fetch_weather",
  description: "Get current weather for a location",
  process: async (location) => {
    const response = await fetch(`/api/weather?location=${location}`);
    return response.json();
  }
});

// Add a skill that will be used as an entry point in a workflow
agent.addSkill({
  name: "fetch_weather",
  description: "Get current weather for a location",
});

// Attach the skill to a workflow
```
- Defined in src/Agent/Agent.class.ts:582

call

call(skillName: string, ...args: any[]): Promise<any>
Parameters
- skillName: string
- ...args: any[]
Returns Promise<any>
- Defined in src/Agent/Agent.class.ts:588

chat

chat(options?: string | ChatOptions): Chat
Create a new chat session with the agent.

Chat sessions maintain conversation context and allow for back-and-forth interactions with the agent, preserving message history.
Parameters
- Optionaloptions: string | ChatOptions
  The options for the chat session if you provide a string it'll be used as the chat ID and persistence will be enabled by default
  - string
  - ChatOptions
    Optionalid?: string
    The ID of the chat. If not provided, a random ID will be generated.
    
    If provided, it will be used to identify the chat in the storage provider and try to load the previous messages from the storage provider if the chat is not found, a new chat will be created
    
    Optionalpersist?: boolean | ILLMContextStore
    If true, the chat will be persisted in the default SRE storage provider : next time you create a chat with the same chat ID and same agent ID, it will load the previous messages from the storage provider
    
    If false, the chat will not be persisted
    
    If a ILLMContextStore is provided, the chat will be persisted in the provided store
Returns Chat
Chat instance for interactive conversations
Example
```
const chat = agent.chat();

// Send messages in sequence
await chat.send("Hello, I need help with my project");
await chat.send("Can you explain the benefits?");
await chat.send("What are the next steps?");

// Get conversation history
const history = chat.getHistory();
```
- Defined in src/Agent/Agent.class.ts:690

emit

emit(event: string, ...args: any[]): void
Parameters
- event: string
- ...args: any[]
Returns void
Inherited from SDKObject.emit
- Defined in src/Core/SDKObject.class.ts:40

`Protected`init

init(): Promise<void>
Returns Promise<void>
Inherited from SDKObject.init
- Defined in src/Core/SDKObject.class.ts:30

mcp

mcp(transport: MCPTransport, port?: number): Promise<string>
Expose the agent as a MCP (Model Context Protocol) server

The MCP server can be started in two ways:
- STDIO: The MCP server will be started in STDIO mode
- SSE: The MCP server will be started in SSE mode, this is case the listening url will be http://localhost:/mcp
Parameters
- transport: MCPTransport
  The transport for the MCP server
- port: number = 3388
  The port for the MCP server (when using SSE transport)
Returns Promise<string>
MCP instance
Example
```
const agent = new Agent({ /* ... agent settings ... */ });

const stdioMcp = agent.mcp(MCPTransport.STDIO);
const sseMcp = agent.mcp(MCPTransport.SSE, 3389);
```
- Defined in src/Agent/Agent.class.ts:741

off

off(event: string, listener: (...args: any[]) => void): void
Parameters
- event: string
- listener: (...args: any[]) => void
Returns void
Inherited from SDKObject.off
- Defined in src/Core/SDKObject.class.ts:44

on

on(event: string, listener: (...args: any[]) => void): void
Parameters
- event: string
- listener: (...args: any[]) => void
Returns void
Inherited from SDKObject.on
- Defined in src/Core/SDKObject.class.ts:36

once

once(event: string, listener: (...args: any[]) => void): void
Parameters
- event: string
- listener: (...args: any[]) => void
Returns void
Inherited from SDKObject.once
- Defined in src/Core/SDKObject.class.ts:48

prompt

prompt(prompt: string, options?: any): AgentCommand
Send a prompt to the agent and get a response.

The returned command can be executed in multiple ways:
- Promise mode: await agent.prompt("question") - returns final result
- Explicit execution: await agent.prompt("question").run() - same as above
- Streaming mode: await agent.prompt("question").stream() - returns event emitter
Parameters
- prompt: string
  The message or question to send to the agent
- Optionaloptions: any
Returns AgentCommand
AgentCommand that can be executed or streamed
Example
```
// Simple prompt (promise mode)
const answer = await agent.prompt("What is the capital of France?");


// Streaming for long responses
const stream = await agent.prompt("Write a detailed report").stream();
stream.on('data', chunk => console.log(chunk));
stream.on('end', () => console.log('Complete!'));
```
- Defined in src/Agent/Agent.class.ts:660

removeListener

removeListener(event: string, listener: (...args: any[]) => void): void
Parameters
- event: string
- listener: (...args: any[]) => void
Returns void
Inherited from SDKObject.removeListener
- Defined in src/Core/SDKObject.class.ts:51

`Static`import

import(data: TAgentSettings): Agent
Import an agent from a file or configuration object.

Supported import patterns:
- Import from .smyth file: Agent.import('/path/to/agent.smyth')
- Import from configuration: Agent.import(settingsObject)
- Import with overrides: Agent.import('/path/to/agent.smyth', overrides)
Parameters
- data: TAgentSettings
  File path or agent settings object
Returns Agent
New Agent instance
Example
```
// Import from file
const agent1 = Agent.import('./my-agent.smyth');

// Import from configuration object
const agent2 = Agent.import({
  name: "Imported Agent",
  model: "gpt-4"
});

// Import with overrides
const agent3 = Agent.import('./base-agent.smyth', {
  name: "Customized Agent",
  behavior: "Custom behavior override"
});
```
- Defined in src/Agent/Agent.class.ts:405
import(data: string, overrides?: any): Agent
Import an agent from a file or configuration object.

Supported import patterns:
- Import from .smyth file: Agent.import('/path/to/agent.smyth')
- Import from configuration: Agent.import(settingsObject)
- Import with overrides: Agent.import('/path/to/agent.smyth', overrides)
Parameters
- data: string
  File path or agent settings object
- Optionaloverrides: any
  Optional settings to override imported configuration
Returns Agent
New Agent instance
Example
```
// Import from file
const agent1 = Agent.import('./my-agent.smyth');

// Import from configuration object
const agent2 = Agent.import({
  name: "Imported Agent",
  model: "gpt-4"
});

// Import with overrides
const agent3 = Agent.import('./base-agent.smyth', {
  name: "Customized Agent",
  behavior: "Custom behavior override"
});
```
- Defined in src/Agent/Agent.class.ts:406

Class Agent

Example

Hierarchy (View Summary)

Index

Constructors

Properties

Accessors

Methods

Constructors

constructor

Parameters

Returns Agent

Example

Properties

structure

Accessors

data

Returns AgentData

llm

Returns TLLMProviderInstances

Example

ready

Returns ControlledPromise<any>

storage

Returns TStorageProviderInstances

team

Returns Team

vectorDB

Returns TVectorDBProviderInstances

Methods

addSkill

Parameters

Returns { in: (inputs: TSkillInputs) => void; out: { body: any; headers: any; query: any; [key: string]: any };}

Example

call

Parameters

Returns Promise<any>

chat

Parameters

Optionalid?: string

Optionalpersist?: boolean | ILLMContextStore

Returns Chat

Example

emit

Parameters

Returns void

Protectedinit

Returns Promise<void>

mcp

Parameters

Returns Promise<string>

Example

off

Parameters

Returns void

on

Parameters

Returns void

once

Parameters

Returns void

prompt

Parameters

Returns AgentCommand

Example

removeListener

Parameters

Returns void

Staticimport

Parameters

Returns Agent

Example

Parameters

Returns Agent

Example

Settings

On This Page

Returns {
in: (inputs: TSkillInputs) => void;
out: { body: any; headers: any; query: any; [key: string]: any };
}

`Optional`id?: string

`Optional`persist?: boolean | ILLMContextStore

`Protected`init

`Static`import