TypeScript · Node.js · Electron

An AI Agent SDK.
Built for your product.

The choice of over 1300+ developers

The shortest path to intelligent software

A leader in Agent SDK standardization

Tansr Agent Harness SDK brings the agent loop, tools, context and permission controls into your application. Build the experience you have in mind, with the runtime already connected.

Start integratingnpm install @tansr/sdk

Run the SDK in Node.js or the Electron main process. Connect your own UI.

Our MIT open-source commitment
import { createSession } from '@tansr/sdk';

// Obtain a short-lived token from your authenticated backend.
const session = await createSession({ token, baseUrl });
session.send('Summarize this contract');
for await (const event of session.events) {
  if (event.type === 'msg.text.delta') process.stdout.write(event.text);
}

For SDK applications only

SDK application pricing

Choose an operating scale for agent applications built with the SDK. Adjudication, memory, context, tools and media capabilities span the plans; individuals and organizations each hold their own plan and wallet.

Standalone use of CLI, serve, ACP or headless does not require this subscription. Services delivered through platform applications are subject to their individual or organization owner's application limits. Model and media calls are still billed by usage.

Trial

限时五折!全额返还!

Was ¥98.00 / year50% off

¥49.00/ year

Annual subscription

Included operating limits:

  • 10 concurrency slots
  • Up to 5 applications
  • Up to 3 developer seats per organization
  • Up to 100 distinct end users per month
  • ¥49.00 in model credit per purchase (CNY, not withdrawable)

Administrator-supplied catalog text is shown in its original language; translations are not available for these fields.

Choose SDK plan

Individual developer

限时五折!全额返还!

Was ¥198.00 / year50% off

¥99.00/ year

Annual subscription

Included operating limits:

  • 1,000 concurrency slots
  • Up to 10 applications
  • Up to 10 developer seats per organization
  • No monthly distinct end-user limit
  • ¥99.00 in model credit per purchase (CNY, not withdrawable)

Administrator-supplied catalog text is shown in its original language; translations are not available for these fields.

Choose SDK plan

Team developer

限时五折!全额返还!

Was ¥1980.00 / year50% off

¥990.00/ year

Or ¥99.00 / month

Included operating limits:

  • 10,000 concurrency slots
  • Up to 32 applications
  • Up to 100 developer seats per organization
  • No monthly distinct end-user limit
  • ¥990.00 in model credit per purchase (CNY, not withdrawable)

Administrator-supplied catalog text is shown in its original language; translations are not available for these fields.

Choose SDK plan

Flagship

限时五折!全额返还!

Was ¥19800.00 / year50% off

¥9900.00/ year

Or ¥990.00 / month

Choose SDK application scale

Included operating limits:

  • 100,000 concurrency slots
  • Up to 100 applications
  • Up to 1,000 developer seats per organization
  • No monthly distinct end-user limit
  • ¥9900.00 in model credit per purchase (CNY, not withdrawable)

Administrator-supplied catalog text is shown in its original language; translations are not available for these fields.

Choose SDK plan

Create your first 5 applications free of subscription charges. More than 5 applications requires Individual developer (small) or above. Select individual or organization ownership before purchase. All prices are in Chinese yuan (CNY).

Bonus tokens offset settled usage and cannot be redeemed for cash. Applies only to platform-paid LLM usage settled in Chinese yuan (CNY). Input, output, cache-read and cache-write tokens are counted without duplication. Images, video, speech recognition (ASR), speech synthesis (TTS), search, bring-your-own-key (BYOK) calls and client-reported usage are excluded.

See full pricing and billing details →

MIT · OPEN SOURCE COMMITMENT

Build with a future you can own.

We commit to releasing the Tansr Agent SDK source under the MIT license on .

Review the source

Inspect how the SDK runs, handles data and applies permissions. Make technical decisions with the implementation in view.

Adapt it to your product

Modify and integrate the released source under the MIT license, including in commercial products.

Keep the code in your hands

Retain and maintain your own copy of the released code, giving your team more control over its future.

The commitment concerns the SDK source to be released. MIT permits use, modification and redistribution while requiring copyright and license notices to be retained. Hosted platform services and model usage remain governed by their own terms.

Agent Harness SDK

The runtime behind your AI product

Build a desktop assistant, a file workflow or an agent inside your service. The harness handles the execution loop while you build the product experience.

Agent loop and model switching

An Agent Harness SDK connects model calls, tool execution and results in a multi-step agent loop. Use platform-configured, authorized models and switch through aliases within a session.

Explore the integration →

Electron, TypeScript and Node.js integration

Install the npm package with TypeScript definitions. Run the SDK in the Electron main process and connect your interface over IPC, or embed it in a Node.js server.

Explore the integration →

Function calling, MCP and Agent Skills

Connect your business functions with defineTool, existing services through MCP, and domain knowledge through defineSkill or skill directories. File operations remain subject to capabilities and runtime permissions.

Explore the integration →

Multi-turn sessions, recovery and streaming

Resume conversations with history snapshots and an injected store. Stream text and tool status into your interface with SessionView. Your application chooses where to persist conversation history.

Explore the integration →

Tool approvals and client key protection

Connect capability switches, runtime rules and approval callbacks. In token mode, keep appkey credentials on your backend, issue short-lived tokens to clients and let the platform manage upstream keys.

Explore the integration →

Token usage and per-user metering

Track consumption with live usage events, per-user queries and developer cost reconciliation. The SDK provides metering; developers implement their own end-user pricing and payment flow.

Explore the integration →

One harness, five ways to use it

Embed the SDK in a desktop product or a Node.js service. The same agent core also supports terminal, service, protocol and unattended workflows.

SDK

Embed the agent runtime in your application and connect your own interface.

CLI

Use the same core from the terminal for interactive tasks.

serve

Expose the runtime through a service interface.

ACP

Connect compatible clients through the Agent Client Protocol.

headless

Run the agent without the interactive terminal interface.

Explore Electron and Node.js integration →

Before you integrate

What does an Agent Harness SDK handle?

It runs the model and tool loop, manages conversational context and sessions, and applies configured permission rules. Your application provides its business functions, user interface, identity and storage choices.

Can I keep long-lived API keys out of my desktop client?

In token mode, your backend keeps the appkey credentials and issues short-lived tokens after authenticating a user. Upstream model keys stay on the platform. Follow the token-server example for the complete integration.

Can I bill my own users from Token usage?

Use live usage events, per-user counters and developer cost reconciliation as inputs. You implement end-user prices, invoicing and payments. SDK application plans and model or media usage are separate charges.