import { Agent } from '@smythos/sdk';
const agent = new Agent({
name: 'Book Assistant',
model: 'gpt-4o',
behavior: 'You are a helpful assistant that can answer questions about the books.',
});
agent.addSkill({
name: 'get_book_info',
description: 'Use this skill to get information about a book',
process: async ({ book_name }) => {
/* ... Skill logic ... */
},
});
const result = await agent.call('get_book_info', { book_name: 'The Great Gatsby' });
console.log(result);
const response = await agent.prompt('Who is the author of "The Great Gatsby"?');
console.log(response);
const response = await agent.prompt('Who is the author of "The Great Gatsby"?', {
files: ['./path/to/image.png', 'https://image.com/myimage.png'],
});
console.log(response);
//First we define the skill that will handle the image
const imgSkill = agent.addSkill({
name: 'ImageAnalyser',
description: 'Any attachment should be processed by this function',
process: async ({ image_url }) => {
/* ... image processing goes here ...*/
return 'Image analysed';
},
});
//very important, we need to force the input type of image_url to Binary
//this will tell our agent to skip the native LLM image processing, and use the agent skill instead
imgSkill.in({
image_url: {
type: 'Binary',
description: 'The image url that we will analyze',
},
});
//now the images will be processed by the agent skill instead of the native LLM
const response = await agent.prompt('Who is the author of "The Great Gatsby"?', {
files: ['./path/to/image.png', 'https://image.com/myimage.png'],
});
const stream = await agent.prompt('Who is the author of "The Great Gatsby"?').stream();
stream.on(TLLMEvent.Content, (data) => {
console.log(data);
});
//all available events: Content, End, Error, Thinking, ToolInfo, ToolCall, ToolResult, Usage
//Simple chat with a session id
const chat = agent.chat('my-chat-session');
//Persisted chat
const persistedChat = agent.chat({ id: 'my-chat-session', persist: true });
//with prompt
const response = await chat.prompt('Hi there');
console.log(response);
//with stream
const stream = await chat.prompt('Hi there').stream();
stream.on(TLLMEvent.Content, (data) => {
console.log(data);
});
//with stdio transport
await agent.mcp(MCPTransport.STDIO);
//with SSE transport, you can define the port number in the second argument
const mcpUrl = await agent.mcp(MCPTransport.SSE, 3388);
const llm = agent.llm.OpenAI('gpt-4o');
//other available models : agent.llm.Anthropic(...), agent.llm.Google(...), agent.llm.Groq(...), agent.llm.TogetherAI(...), agent.llm.Bedrock(...), agent.llm.VertexAI(...), agent.llm.xAI(...), agent.llm.Perplexity(...)
//prompt
const response = await llm.prompt('Who is the author of "The Great Gatsby"?');
console.log(response);
//stream
const stream = await llm.prompt('Who is the author of "The Great Gatsby"?').stream();
stream.on(TLLMEvent.Content, (data) => {
console.log(data);
});
const llm = agent.llm.OpenAI('gpt-4o', {
temperature: 0.5,
max_tokens: 100,
top_p: 1,
frequency_penalty: 0,
presence_penalty: 0,
});
const llm = agent.llm.OpenAI('gpt-4o', {
apiKey: 'your-api-key',
});
const llm = agent.llm.OpenAI(Model.OpenAI('custom-model', { baseURL: 'https://api.custom-model.com' }));
const storage = agent.storage.LocalStorage();
await storage.write('test.txt', 'Hello, world!');
const data = await storage.read('test.txt');
console.log(data);
const storage = agent.storage.S3Storage({
bucket: 'your-bucket',
region: 'your-region',
});
const pinecone = agent.vectorDB.Pinecone('my-namespace', {
indexName: 'my-index',
apiKey: 'your-pinecone-api-key',
embeddings: Model.OpenAI('text-embedding-3-large'),
});
await pinecone.insertDoc('my-doc', 'This is the content of my document');
const results = await pinecone.search('my query');
const milvus = agent.vectorDB.Milvus('my-collection', {
credentials: {
address: 'your-milvus-address',
token: 'your-milvus-token',
},
embeddings: Model.OpenAI('text-embedding-3-large'),
});
await milvus.insertDoc('my-doc', 'This is the content of my document');
const results = await milvus.search('my query');
// RAMVec is a zero-config, in-memory vector DB for quick testing.
const ramvec = agent.vectorDB.RAMVec('my-collection');
await ramvec.insertDoc('my-doc', 'This is the content of my document');
const results = await ramvec.search('my query');
Many capabilities are also available as standalone classes, you can use them without an agent
const openaiGPT4Mini = LLM.OpenAI('gpt-4o-mini');
const anthropicClaude4Sonnet = LLM.Anthropic('claude-4.0-sonnet');
const googleGemini20Flash = LLM.Google('gemini-2.0-flash');
//...
const result = await openaiGPT4Mini.prompt('Hello, how are you?');
const llm = LLM.OpenAI('gpt-4o-mini');
const result = await llm.prompt('Describe these images?', {
files: ['./path/to/image.png', 'https://image.com/myimage.png'],
});
const stream = await openaiGPT4Mini.prompt('Hello, how are you?').stream();
stream.on(TLLMEvent.Content, (data) => {
console.log(data);
});
const storage = Storage.LocalStorage();
await storage.write('test.txt', 'Hello, world!');
const data = await storage.read('test.txt');
console.log(data);
const storage = Storage.S3Storage({
bucket: 'your-bucket',
region: 'your-region',
});
await storage.write('test.txt', 'Hello, world!');
const data = await storage.read('test.txt');
console.log(data);
const pinecone = VectorDB.Pinecone('my-namespace', {
indexName: 'my-index',
apiKey: 'your-pinecone-api-key',
embeddings: Model.OpenAI('text-embedding-3-large'),
});
await pinecone.insertDoc('my-doc', 'This is the content of my document');
const results = await pinecone.search('my query');
const milvus = VectorDB.Milvus('my-collection', {
credentials: {
address: 'your-milvus-address',
token: 'your-milvus-token',
},
embeddings: Model.OpenAI('text-embedding-3-large'),
});
await milvus.insertDoc('my-doc', 'This is the content of my document');
const results = await milvus.search('my query');
// RAMVec is a zero-config, in-memory vector DB for quick testing.
const ramvec = VectorDB.RAMVec('my-collection');
await ramvec.insertDoc('my-doc', 'This is the content of my document');
const results = await ramvec.search('my query');
const parsedDoc = await Doc.auto.parse('./path/to/file.pdf');
//PDF
const parsedPDFDoc = await Doc.pdf.parse('./path/to/file.pdf');
//DOCX
const parsedDocxDoc = await Doc.docx.parse('./path/to/file.docx');
//Markdown
const parsedMarkdownDoc = await Doc.md.parse('./path/to/file.md');
//Text
const parsedTxtDoc = await Doc.text.parse('./path/to/file.txt');
const parsedString = await Doc.text.parse('This is the content of my document');
const parsedDoc = await Doc.auto.parse('./path/to/file.pdf', {
title: 'My Document',
author: 'John Doe',
});