Blazing Agents TypeScript SDK

Build production agents with a typed TypeScript client for the Blazing Agents API.

Documentation · npm

The official TypeScript SDK is a client for the Blazing Agents /v1 API that your backend uses to manage Agents and run them for your users. It requires Node.js 24 or newer.

Features

• A typed method for every /v1 operation an API key can call: Agents, Skills, Sessions, Memories, Prompts, Tasks, Workspaces, Artifacts, Providers, MCP Connections, Chat Connections, usage, spending limits, Tenant settings, and merchant monetization.
• Chat streams in the AI SDK UI message stream format, with steering, Tool approvals, and backend functions that run in your process.
• Text and structured-object generation.
• End-user scoped clients through forUser.
• Cursor pagination, uploads, and downloads.
• Request IDs and typed errors.
• BlazingAgentsChatTransport and createChatRelay connect AI SDK useChat to your backend.

Installation

npm install @blazingagents/sdk ai@^7.0.84 zod@^4.5.4

ai and zod are required peer dependencies. The SDK re-exports AI SDK message types and Zod schemas, so your app and the SDK share one copy of each.

Quick start

Create a Full API key in the Blazing Agents dashboard and set BLAZING_AGENTS_API_KEY in your server environment. The key acts for your whole Tenant, so keep it on your backend and never send it to a browser.

import { BlazingAgents } from "@blazingagents/sdk";

const client = new BlazingAgents({
  apiKey: process.env.BLAZING_AGENTS_API_KEY!,
});

const result = await client.completion({
  agentId: "ag_...",
  prompt: "Write a friendly welcome message.",
});

console.log(await result.text);

Replace ag_... with an Agent ID from the dashboard.

For requests from your users, scope the client to the user ID that your backend verified at sign-in, never to an ID from the request body. A scoped client reaches only that End-user's Agents, Sessions, and Workspace. An unscoped Agent shares one Workspace across all your users, so give each user an Agent of their own:

export async function createAgentForUser(userId: string): Promise<string> {
  const agent = await client.forUser(userId).agents.create({
    name: "Assistant",
    providerId: "prv_...",
    model: "openai/gpt-6-luna",
  });
  return agent.id;
}

export async function chatAsUser(
  userId: string,
  agentId: string,
  text: string
): Promise<Response> {
  const chat = await client.forUser(userId).chat({
    agentId,
    message: {
      id: crypto.randomUUID(),
      role: "user",
      parts: [{ type: "text", text }],
    },
  });
  return chat.toResponse();
}

Call createAgentForUser once per user and store the Agent ID with the user. A scoped chat with an Agent created without the scope fails with 404. chatAsUser returns the chat as a streaming Response. To chat from the browser with useChat, serve createChatRelay from your backend and use BlazingAgentsChatTransport. The relay scopes each request to the user that your resolveContext returns.

Examples

The examples repository has complete TypeScript and Python apps built on the Blazing Agents SDKs.

Documentation

Read the TypeScript SDK documentation for client setup, End-user scoping, chat and generation, a guide to each resource, and error handling.

Development

npm ci
npm run check
npm run test:consumer

See CONTRIBUTING.md for the changelog convention and how to release.

License

MIT