iFlytek Star Agent Platform: AI Agent Development Platform by iFlytek
iFlytek Star Agent Platform: AI Agent Development Platform by iFlytekIntroduction
What is the iFlytek Xingchen Agent Platform?
The iFlytek Xingchen Agent Platform is an AI agent development platform launched by iFlytek for developers and enterprises, covering the full process from learning, functional development, and engineering deployment to enterprise applications. The platform focuses on a fully open technical foundation and flexible customization capabilities, aiming to help developers achieve AI application innovation faster.
In January 2026, iFlytek Xingchen completed an important upgrade, with key enhancements in agents' capabilities for physical-world perception, complex context understanding, and multimodal natural interaction. The upgraded platform builds a complete multimodal interaction solution around voice, vision, action, and execution capabilities, with agents evolving from single-function tools into interactive, executable roles. iFlytek Xingchen is now fully open source (Apache 2.0 license), supporting free secondary development and commercial productization.

Core Advantages of the iFlytek Xingchen Agent Platform
Open and flexible architecture: The platform can connect to various models, as convenient as plug-and-play tools. We provide over 16,000 ready-made plugins, and can easily host custom tools, allowing popular applications to be quickly replicated.
Deep customization capabilities: Model performance can be adjusted according to specific needs, with the precision of a tailor-made solution. The platform is also equipped with evaluation tools that can automatically verify effects and optimize models based on scenarios.
Features of the iFlytek Xingchen Agent Platform
Creation methods: Three modes are supported—instruction-based (created via prompts), workflow-based, and interactive digital human. Users can generate agents and workflows through natural language descriptions. The platform also provides numerous templates, enabling complex task orchestration even without programming experience.
Development capabilities: Built-in Xingchen RPA core tools support process recording and drag-and-drop development, and can connect to various enterprise software systems. The platform integrates 870 AI capabilities from the iFlytek Open Platform and over 16,000 MCP Servers, covering common scenarios such as maps, search, payment, and weather queries. It also supports OpenAI-Like API access and custom tool hosting.
Model compatibility: In addition to iFlytek Spark, the platform also supports mainstream large models such as DeepSeek and Qwen, allowing developers to compare and tune prompts across different models.
Effect evaluation: The platform provides a complete chain from development, evaluation, publishing to operations. Developers can create evaluation tasks and evaluation sets to systematically verify agent performance.
Publishing channels: Agents can be published with one click to multiple channels including Spark App/Desk, WeChat Official Accounts, Feishu, API, MCP Server, and the Virtual Human Platform.

Application Scenarios of the Xingchen Agent Development Platform
Enterprise office and process automation: Through RPA capabilities, cross-system connections are achieved, enabling legacy systems to be improved non-intrusively via Agents—for example, scenarios such as financial reimbursement and multi-platform content publishing.
Voice interaction and digital humans: Based on iFlytek's AIUI system, integrating virtual humans, voice replication, and role-playing models, it can be used for smart lobby greeters, entertainment character design, companion products, and more.
Learning and knowledge management: Supports web search summarization and information gathering and organization, making it suitable for academic research and knowledge Q&A applications.
Content creation: Supports building multi-platform content generation Agents, such as copywriting and news summarization.
How to Use the Xingchen Agent Development Platform?
Creating an Instruction-Based Agent
After entering the iFlytek Xingchen Agent Development Platform (https://agent.xfyun.cn/agentbuilder) homepage, click the "Create Application" button on the left. In the pop-up dialog, select "Create with Prompt".
Enter the agent's setup description in the dialog box, then click "Quick Create".
After entering the instruction-based agent setup page, you need to fill in four basic pieces of information: agent name (up to 20 characters), category (choose from Workplace, Creation, Learning, Social, Health, Lifestyle, Other), brief introduction (within 100 characters), and avatar image.
Next, edit the prompt in three parts: Role, Task, and Reply Requirements. For example, the Role can be set as "You are an experienced interviewer", the Task written as "List common interview questions based on the position I provide", and the Reply Requirements written as "Give 5 questions, presented in separate bullet points".
Configure advanced features as needed—web search, AI drawing, and code generation can be enabled; a knowledge base can be added to upload proprietary materials; or conversation enhancement options such as greetings, persona, voice, and background image can be set.
Select the model version on the page, supporting Spark V4.0 Ultra, DeepSeek, Qwen, etc.
After completing, click the "Create" button in the upper right corner. You can input text in the debugging area to test the effect, and click "Publish" once it passes.
Creating a Workflow Agent
After logging into the platform, click "Create" in the upper left corner, then click "Workflow Creation" on the right side of the creation interface.
There are two options here: one is "Custom Creation", which opens a blank canvas for you to build yourself; the other is "Template Creation", which uses the platform's existing templates to generate directly, such as the "AI Podcast Generation Template".
After entering the canvas, first click the edit icon in the upper left corner to set the workflow's name and function description. Then drag the required nodes from the left-side node list onto the canvas, such as "Large Model" nodes, "Code" nodes, "Knowledge Base" nodes, "Q&A nodes", etc., and connect these nodes in order.
Click on each node to configure specific parameters, such as setting the prompt and selecting the model version in the Large Model node. After configuration, click "Debug" to test the running result, and publish once it passes.
Publishing Channels
After the agent is created, you can choose the publishing method as needed:
Personal use only
Publish to iFlytek Spark App, open to Spark users
Publish to WeChat Official Accounts, open to self-media followers
Publish as an API interface, for your own use or shared with specified developers
Publish as an MCP Server, for your own use or shared with other developers
Publish as a Feishu robot, for use by Feishu members of your enterprise
Publish to the Virtual Human Interaction Platform

Competitor Comparison
| Comparison Dimension | iFlytek Xingchen Agent | ByteDance Coze | Baidu Wenxin AgentBuilder | Alibaba Cloud Bailian |
|---|---|---|---|---|
| Core Advantage | Voice interaction + multimodal | Douyin ecosystem + low barrier to entry | Wenxin large model + search ecosystem | Alibaba Cloud infrastructure | Model Support | Spark/DeepSeek/Qwen etc. | Proprietary + Third-party | ERNIE Bot Series | Tongyi Series |
| Voice Capability | Strong (iFlytek's core strength) | Average | Medium | Average |
| Open Source Strategy | Fully open source under Apache 2.0 | Partially open source | Partially open source | Partially open source |
| Number of MCP Tools | 16000+ | Relatively rich | Relatively rich | Rich |
| Pricing | Free / Starting from 9.9 yuan | Free / Paid | Free / Paid | Pay-as-you-go |
| Distribution Channels | Spark App / Official Account / Feishu / API, etc. | Douyin / Feishu, etc. | Baidu ecosystem products | Alibaba ecosystem products |
| Typical Use Cases | Voice assistants, digital humans, enterprise RPA | Content creation, customer service | Search enhancement, content generation | Enterprise-level application development |
FAQ
Q: What are the two main ways that agents on the Xunhao Agent Development Platform serve users?
A: Agents on the Xunhao Agent Development Platform mainly serve users through two approaches: instruction-based and workflow-based. The instruction-based approach allows users to input instructions, and the agent directly returns results; the workflow-based approach requires users to follow a specific process step by step, with the agent providing corresponding services along the way.
Q: Which agent on the Xunhao Agent Development Platform should I use if I want to polish an English SCI journal paper?
A: You should use the English Paper Polishing agent, which is specifically designed for polishing and revising English SCI journal papers.
Q: What news-related agents are available on the Xunhao Agent Development Platform, and what are the differences between them?
A: The news-related agents available include Hot News and Daily News. Hot News aggregates news from authoritative media based on user-input keywords, including news titles, summaries, sources, and publication dates. Daily News provides illustrated daily news with detailed explanations, placing greater emphasis on image-and-text presentation and covering a comprehensive roundup of daily news.
Last updated: August 5, 2026Related Information:
Tutorial Guide on Building Workflow Agents on the iFlytek Xunhao Agent Development Platform
Script Planner Assistant▸
AI Apps
AI News