@mastra/client-js

Mastra Client

JavaScript/TypeScript client library for the Mastra AI framework. This client provides a simple interface to interact with Mastra AI's APIs for agents, vectors, memory, tools, and workflows.

Installation

npm install @mastra/client-js

Quick Start

import { MastraClient } from '@mastra/client-js';

// Initialize the client
const client = new MastraClient({
  baseUrl: 'http://localhost:4111', // Your Mastra API endpoint
});

// Example: Working with an Agent
async function main() {
  // Get an agent instance
  const agent = client.getAgent('your-agent-id');

  // Generate a response
  const response = await agent.generate({
    messages: [{ role: 'user', content: "What's the weather like today?" }],
  });

  console.log(response);
}

Client Configuration

The client can be configured with several options:

const client = new MastraClient({
    baseUrl: string;           // Base URL for the Mastra API
    retries?: number;          // Number of retry attempts (default: 3)
    backoffMs?: number;        // Initial backoff time in ms (default: 300)
    maxBackoffMs?: number;     // Maximum backoff time in ms (default: 5000)
    headers?: Record<string, string>; // Custom headers
});

Available Methods

Agents

• getAgents(): Get all available agents
• getAgent(agentId): Get a specific agent instance
  • agent.details(): Get agent details
• agent.generate(params): Generate a response
  • agent.generateLegacy(params): Legacy API for generating a response (V1 models)
  • agent.stream(params): Stream a response
  • agent.streamLegacy(params): Legacy API for streaming a response (V1 models)
  • agent.getTool(toolId): Get agent tool details
  • agent.evals(): Get agent evaluations
  • agent.liveEvals(): Get live evaluations

Memory

• getMemoryThreads(params): Get memory threads
• createMemoryThread(params): Create a new memory thread
• getMemoryThread(threadId): Get a memory thread instance
• saveMessageToMemory(params): Save messages to memory
• getMemoryStatus(): Get memory system status

Tools

• getTools(): Get all available tools
• getTool(toolId): Get a tool instance
  • tool.details(): Get tool details
  • tool.execute(params): Execute the tool

Workflows

• getWorkflows(): Get all workflows
• getWorkflow(workflowId): Get a workflow instance
  • workflow.details(): Get workflow details
  • workflow.createRunAsync(): Create workflow run
  • workflow.createRun(): Deprecated - use createRunAsync() instead
  • workflow.startAsync(params): Execute the workflow and wait for execution results
  • workflow.resumeAsync(params): Resume suspended workflow step async
  • workflow.watch({runId},(record)=>{}): Watch the step transitions of the workflow run
  • workflow.start({runId, triggerData}): Start a workflow run sync
  • workflow.resume(params): Resume the workflow run sync

Vectors

• getVector(vectorName): Get a vector instance
  • vector.details(indexName): Get vector index details
  • vector.delete(indexName): Delete a vector index
  • vector.getIndexes(): Get all indexes
  • vector.createIndex(params): Create a new index
  • vector.upsert(params): Upsert vectors
  • vector.query(params): Query vectors

Logs

• getLogs(params): Get system logs
• getLog(params): Get specific log entry
• getLogTransports(): Get configured Log transports

Telemetry

• getTelemetry(params): Get telemetry data

Error Handling

The client includes built-in retry logic for failed requests:

• Automatically retries failed requests with exponential backoff
• Configurable retry count and backoff timing
• Throws error after max retries reached

Internal Implementation

The client uses the native fetch API internally for making HTTP requests. All requests are automatically handled with:

• JSON serialization/deserialization
• Retry logic with exponential backoff
• Custom header management
• Error handling