Tiangong AI is a new-generation intelligent agent development platform and real-time internet-connected search Q&A tool independently developed by Kunlun Wanwei, integrating cutting-edge technologies such as large language models (LLM), multimodal understanding, knowledge graphs, and autonomous planning and execution. It is not just a traditional search engine, but an AI agent foundation capable of understanding complex instructions, proactively invoking tools, and completing multi-step tasks, aimed at helping individual users and enterprise clients achieve full-chain intelligence from information acquisition to task delivery.

Unlike most conversational AI products on the market, Tiangong AI's core positioning lies in the dual genes of "real-time + agent": on one hand, its search engine accesses real-time data across the entire internet, ensuring information timeliness and traceability; on the other hand, the platform provides an extensible agent development framework that supports users in orchestrating workflows through natural language, attaching private knowledge bases, databases, or external APIs, thereby building customized agents for specific business scenarios. In the AI industry landscape of 2026, Tiangong AI, with its self-developed dual quadrillion-parameter large models and the integrated "search + agent" architecture, has entered the top tier in China, becoming an important infrastructure for digital transformation in industries such as finance, retail, and education.

• Real-time Internet Search - Supports millisecond-level web-wide information retrieval and structured parsing, capable of providing real-time answers with source citations for dynamic data such as news, stocks, academic papers, and automatically identifying information publication time.
• Multimodal Content Understanding - Supports image recognition, OCR text extraction, voice input and generation, can directly read information from charts and images, and output mixed text-image answers or data reports.
• Agent Orchestration Engine - Built-in agent orchestration workbench allows users to define task workflows via drag-and-drop or natural language (e.g., data collection → cleaning → analysis → report generation), automatically invoking built-in search, calculation, and API plugins.
• Private Knowledge Base Integration - Supports uploading documents such as PDF, Word, and web links, building enterprise-specific knowledge bases to enable Q&A and decision assistance based on private data, with data isolation and encryption.
• Long Text Generation and Summarization - Can process contexts of tens of thousands of words at once, automatically generate structured summaries, mind-map-style key points, and support switching among multiple styles (professional/popular/marketing).
• Code and Data Assistant - Built-in Python execution environment, can directly run data analysis code, output interactive charts and visualization results, and provide code explanations and debugging suggestions.

• Step 1: Registration and Login - Visit the Tiangong AI official website (tiangong.cn), register quickly using a phone number or email, or scan to log in via WeChat or an enterprise account. New users automatically receive free credits without immediate payment.
• Step 2: Familiarize Yourself with the Workbench Interface - After logging in, you enter a conversational homepage with a function navigation bar on the left (search, agent, document analysis, API console) and an interactive input box in the center. It is recommended to first use the "guide templates" to understand typical usage of each module.
• Step 3: Input Instructions or Configure Tasks - Basic usage: directly ask questions in the input box, and the system will prioritize triggering real-time search. Advanced usage: enter the "Agent Orchestration" module, set task objectives, input variables, and execution nodes step by step, for example, setting up "automatically collect competitor news at 9 a.m. daily, generate summaries, and send to email."
• Step 4: Result Debugging and Export - If unsatisfied with the answer, you can iteratively optimize through "follow-up questions" or "regenerate"; supports one-click export of conversation records, reports, and charts to PDF, Word, Markdown, or HTML, and can share with collaborators for comments and secondary editing.

Example 1: Real-time Stock Analysis Prompt - "Please search in real time for the closing prices and price changes of BYD and CATL in A-shares today, compare the capital flows over the last 5 days, and finally output an analysis conclusion in table format." Expected output: The system automatically retrieves relevant data in real time, generates a comparison report containing charts and price change statistics, and cites data sources (e.g., Eastmoney, Shanghai Stock Exchange).

Example 2: Automated Agent Weekly Report - "Create an agent that automatically reads the sales flow table I uploaded every Friday at 8 p.m., calculates the Top 3 products by sales this week, and based on inventory levels gives replenishment suggestions, generates a weekly report and sends it to the team group." Expected output: The platform creates a scheduled trigger agent bound to file parsing, data calculation, and text generation nodes; users need to authorize the data source on first run for continuous operation.

Example 3: Complex Document Speed Reading - "Upload this 45-page industry research report PDF, extract core viewpoints, key data, policy risks, and output as a PPT outline." Expected output: AI automatically parses the document structure, generates a structured summary with 5-8 key points, attaches page references from the original document, and can one-click convert to a PPT template.

• Global Real-time Search + Credible Citations - Self-developed indexing architecture achieves second-level updates, and answers are accompanied by original links, effectively mitigating the hallucination problem of large models; compared to pure generative models like ChatGPT, it has better timeliness and verifiability.
• Deep Integration of Agent and Search - Search, calculation, and generation capabilities can be invoked in agent workflows without additional integration, lowering the entry barrier for complex task construction; compared to platforms like Coze, it is more lightweight.
• Deep Optimization for Chinese Scenarios - Special training for Chinese contexts, policy terminology, and internet neologisms, resulting in significantly better knowledge coverage and response naturalness for domestic users compared to general international models in similar scenarios.
• Enterprise-grade Private Deployment Capability - Supports public cloud, private cloud, and hybrid cloud deployment, with data isolation and audit logs, meeting compliance requirements for industries with high security needs such as finance and government.
• Cost-effectiveness and Flexible Pricing - Provides a free basic version with unlimited usage (including ads); paid plans are priced far lower than international products with equivalent computing power, and supports billing based on API calls, suitable for uses from individuals to enterprises of various sizes.

Comparison with Baidu's ERNIE Bot and Alibaba's Tongyi Qianwen: Tiangong AI is stronger in real-time information and knowledge retrieval; its search engine naturally has the latest data, while ERNIE Bot and Tongyi Qianwen tend to focus more on closed-source conversations and ecosystem integration, with real-time search relying on third-party plugins. In agent development, Tiangong provides a more intuitive orchestration interface and more built-in tools, but its ecosystem richness may be slightly inferior to Tongyi Qianwen's ModelScope.

Comparison with Microsoft Bing Chat / ChatGPT (with browsing): Tiangong AI has significantly better retrieval coverage for Chinese content and domestic data sources (such as WeChat, Zhihu, government websites), and its compliance better suits domestic enterprises; ChatGPT's browsing function is limited by international network access and does not support private deployment. On Tiangong AI, users can directly use the Python execution environment, while ChatGPT requires the Code Interpreter plugin, which is more cumbersome.

Comparison with specialized agent platforms like Coze / Dify: Tiangong AI's advantage lies in "search as a plugin," where real-time data can be obtained without manually registering API keys, greatly reducing setup costs. However, for highly complex multi-agent collaboration and deep custom development, platforms like Coze are richer in node diversity and community component sharing.

Business Model: Tiangong AI adopts a dual-tier model of "freemium + enterprise customization," providing free basic services to C-end users and paid subscriptions and API usage-based billing to B-end clients.

• Free Tier - No payment required; daily use of real-time search, conversation, document analysis (up to 10 pages per session), and basic agents (3 tasks per day). The free tier includes brand advertising or watermarks, suitable for personal learning and light usage.
• Pro - Approximately ¥29/month, with an annual discount to ¥24/month. Includes high-speed search priority channel, unlimited agent runs, long document parsing (200 pages per session), code execution environment, full-format export, and no ads or watermarks.
• Team - Charged per seat at ¥69/month per seat, providing unified management console, knowledge base sharing, permission levels, SSO login, and audit logs; suitable for teams of 10 or more.
• API and Enterprise Customization - Provides standard RESTful APIs, billed per token, approximately ¥0.002 per 1,000 tokens (input) and ¥0.005 per 1,000 tokens (output). Private deployment and custom model fine-tuning require contacting sales for project-based quotes.

• Software Engineers and Data Analysts - Need real-time technical documentation queries, code debugging, SQL optimization, and data visualization; Tiangong AI's code explanation and Python execution can be directly integrated into development workflows, improving efficiency.
• Academic Researchers and University Students - Used for literature reviews, data verification, paper grammar polishing, and cross-language translation; real-time retrieval ensures access to the latest research findings, and built-in source citations facilitate academic tracing.
• Self-media and Marketing Professionals - Skilled in trend tracking, pain point analysis, and creative copy generation; use agents to automatically complete competitor monitoring and viral topic mining, and output scripts adapted for multiple platforms with one click.
• Enterprise Operations and Decision Makers - Market intelligence collection, user feedback analysis, and automatic generation of weekly and monthly reports; suitable for operations staff without technical backgrounds, configure agents through natural language, lowering the barrier to using digital tools.

• Workplace Copywriting - Automatically generate professional emails, meeting minutes, and work summaries based on given key points, with tone switching (formal/colloquial/motivational), greatly reducing time spent on paperwork.
• Programming Assistance and Code Refactoring - Supports code completion, bug localization, and performance tuning suggestions; can automatically migrate old code from Python 2 to Python 3 and generate unit test cases.
• Intelligent Customer Service - Connect to a private knowledge base to build 24/7 automated customer service, answering product FAQs, order queries, and after-sales policies in real time, with a handoff interface to human agents.
• Multilingual Localization Translation - Provides translation between 20+ languages including Chinese, English, Japanese, and Korean, preserving industry terminology and contextual style, and exports bilingual comparison tables after translation, suitable for foreign trade and multinational enterprises.
• Market Intelligence Monitoring - Use scheduled agents to capture news, social media, and academic articles for specified keywords daily, automatically aggregate them into briefs and send by email, helping enterprises quickly respond to public opinion changes.