MiTa AI search provides full-network search, in-depth research, fact checking and voice capabilities, adapting to Chinese information acquisition scenarios.

Secret Tower AI Search: A Chinese AI search engine that trades computing power for manpower

Core parameters and statistics

Metaso AI Search (Metaso) is an AI-driven search engine developed by Shanghai Metaso Network Technology Co., Ltd. It is officially positioned as "exchanging computing power for manpower, empowering work and learning." Its core logic is to break the limitation of traditional search "only returning links", and realize the full-process information processing of "search - integration - creation - management" through the self-developed large language model MetaLLM and deep semantic analysis technology.

Projects	Public Information
Product positioning	Chinese AI search engine with direct results without advertising
Core Technology	Self-developed MetaLLM + RAG + MoE (Hybrid Expert Model)
Search mode	Simple mode (2-3s), in-depth mode (5-8s), research mode (10-15s)
Result form	Structured answers, mind maps, outline documents, topic tables, citation tracing
Place of Belonging	CN
Support platform	Web (main entrance), API
Free quota	100 searches per day for basic functions (limited if not logged in)
Latest version	deep-research-mode (upgraded version of deep research mode, 2026-06)

Business Implications of Search Mode: The essential difference between the three modes is not the speed, but the depth of information processing. The concise mode is suitable for daily quick verification (such as holiday arrangements, encyclopedia queries), the in-depth mode is suitable for learning and work analysis (competing product dismantling, project research), and the research mode is oriented to academic-level tasks, reducing the risk of miscitation through cross-validation of multiple sources. Users can flexibly switch based on task urgency and complexity without having to jump between different tools.

Result structuring capability: Instead of giving users a string of blue links for users to filter, Secret Tower automatically integrates fragmented information into outlines, mind maps or timeline tables. This means that in the Chinese context where information is overloaded, users can save the trouble of "opening 10 tabs → manually comparing → refining key points" and directly get editable structured output.

Users and market recognition of Secret Tower AI search

MiTa AI Search's market recognition mainly comes from word-of-mouth communication among Chinese knowledge workers and inclusion and coverage by third-party navigation sites, rather than publicly disclosed revenue or user numbers.

User Group Positioning: Core users are Chinese knowledge workers who are "sensitive to information quality but have high time costs" - including academic researchers, legal practitioners, financial analysts, and operational strategists. The common pain points of this group of people are: there are many advertisements in traditional search engines, serious information fragmentation, scattered professional information sources, and a lot of man-hours are spent on "searching → filtering → sorting" every day.

Industry coverage depth: According to third-party evaluations, MiTa’s traceability accuracy in the academic field (98.7%), contract risk identification accuracy in legal scenarios (92%), and financial report indicator extraction capabilities in financial scenarios have reached usable levels. This vertical industry penetration makes it different from the general AI search that "can ask everything but cannot answer anything deeply".

Market Signal: As of mid-2026, Secret Tower has been included as a key recommendation tool in the "AI search engine" category by leading AI navigation sites (aigc.cn, etc.), and has gained continued discussion in the domestic knowledge management community. However, the number of MAU or enterprise customers has not been officially disclosed, and third parties need to pay attention to this information gap when making judgments.

The cost advantage of Secret Tower AI search: dismantling of the three-layer structure

• C-side/Individual: Usually a free version is provided to experience the core functions, and high-frequency use requires a paid package subscription.
• API/Developer: Billed by call volume, suitable for development teams that can be flexibly integrated into their own systems.
• Enterprise/Privatized: Contact the business owner for customized quotation and deployment plan. The specific price is subject to the official real-time pricing page.

Main functions of Secret Tower AI search

The functional system of MiTa AI search is built around "information retrieval → integration → application → management", and the core capabilities can be summarized into six categories:

• Semantic search and three modes: Based on MetaLLM's natural language understanding capabilities, it supports fuzzy intent recognition and context questioning. Three modes (concise/in-depth/research) cover the full spectrum of needs from second-level verification to multi-source cross-validation, and users do not need to switch between multiple search tools.
• Structured results presentation: Automatically generate mind maps (showing logical relationships of information), outline documents (hierarchically presenting core ideas), topic tables (structured data of people/events), and citation tracing (all information is marked with source links, with an accuracy of 98.7% in academic scenarios). Linked value: Structured output not only lowers the threshold for information reading, but also supports one-click export of Word/PDF/PPT, directly entering the creation section, avoiding the gap of "manually organizing the format after searching".
• Vertical scene deep adaptation: The academic scene supports hundreds of millions of document screening (including PubMed, SCI journals), automatic reference typesetting (APA/GB/T7714), and experimental data visualization; the legal scene supports contract risk scanning (term recognition accuracy 92%) and similar case retrieval; the financial scene supports the extraction of key indicators of financial reports and the automatic generation of analysis briefings.
• Topic and bookshelf knowledge management: Users can create customized knowledge bases ("topics"), support uploading local documents (PDF/DOCX/PPTX), collect search results, and manually enter notes. Subsequent searches can be limited to topics, greatly improving accuracy and supporting team collaboration and sharing. The bookshelf function is used for long-term accumulation and offline use.
• Multi-format export and ecological linkage: Search results support exporting to Word, PDF, and PPT, and can be connected to the Secret Tower ecological tool "Secret Tower Writing Cat" to achieve the integration of "search → creation → polishing". For example, enter "2025 AI Search Industry Report", and AI will automatically generate a slide frame and style template.
• Fact checking and multi-modal input: built-in

Real-time verification capabilities, assisting in reducing information bias through multi-source consistency comparison. Supports multi-type input of text, image PDF, covering web pages, images, videos, podcasts and other resources across the entire network.

Model and version evolution of Secret Tower AI search

Since its launch in early 2024, Secret Tower AI Search has experienced an evolution from "basic AI search" to "in-depth research + knowledge management". Since the official has not fully disclosed the precise release date of each version, the following context is based on public pages and third-party collection records.

Mainline release

• AI search initial public version (~2024-01): Chinese AI search capabilities are launched, and the core selling point is "ad-free direct results". At that time, the functions were mainly based on basic semantic search and concise answer generation, and did not yet support in-depth research mode and topic management.
• In-depth research mode upgraded version (~2025 second half - 2026-06): Gradually launch in-depth mode and research mode, adding capabilities such as multi-source cross-validation, mind map generation, and access to academic literature databases. Around June 2026, the homepage continued to strengthen the three keywords of "in-depth research", "fact verification" and "voice input and output", marking a shift in product focus from "fast search" to "deep search and accurate verification".

Technology stack evolution

• MetaLLM Base: The self-developed large language model continues to iterate, supports multi-modal input (text, pictures, speech), and the semantic understanding accuracy is claimed to be over 95%.
• RAG and MoE introduced: The hallucination rate is reduced through enhanced retrieval generation (RAG), and the academic scene error rate is reduced to 1.3%; the hybrid expert model (MoE) allocates exclusive expert modules for different fields to improve the accuracy of professional problem processing.
• Hierarchical attention mechanism: When processing long documents of more than 50 pages, prioritize key modules such as executive summaries and data tables to avoid context loss.

Version information boundary: The above version nodes are mostly a comprehensive collection of public pages and third-party reviews. The official has not released a precise version release date. If you need version records accurate to the date, it is recommended to refer to the official real-time page or official announcement.

Technical advantages of Secret Tower AI search

The technical competitiveness of MITA AI Search does not lie in the parameter scale of a single model, but in the architectural design of the "full link of information processing" - from the data layer to the algorithm layer to the application layer, each section is solving the specific pain points of information overload in the Chinese scenario.

MetaLLM + RAG synergy mechanism: The self-developed model MetaLLM is responsible for semantic understanding and generation, and the RAG module is responsible for anchoring the generated content to the real source. The advantage of this "generation + retrieval" dual engine is that users do not get answers "recited" by the model, but verifiable conclusions with reference links. The traceability accuracy rate in academic scenarios reaches 98.7%, which means that less than 2 out of 100 citations are biased, which has practical engineering significance in scenarios where information verification is sensitive.

MoE Hierarchical Reasoning: The hybrid expert model architecture allocates dedicated reasoning modules to different fields (academic, legal, financial, general). The effect is: the term recognition accuracy in legal scenarios is 92%, and the financial reporting indicator extraction accuracy in financial scenarios is higher than that of the general model. This "specialized" design makes Secret Tower perform more stably in vertical scenarios than general-purpose AI search (such as the Chinese version of Perplexity).

Full-link privacy design: User search records are not tracked, private documents are end-to-end encrypted (TLS protocol + 512-bit RSA algorithm), and private deployment version data does not go through the public cloud. The output content is original and reorganized to prevent copying of the original text, and is equipped with a blockchain certification function for copyright ownership. This is a compliance bonus in enterprise knowledge management scenarios, but it still needs to be noted: the legal effect of blockchain certificates may differ in different jurisdictions.

Technical bottleneck: It is still recommended to split tasks for long document processing (more than 50 pages); the recall rate of multi-lingual mixed search (such as Chinese, English and Japanese) is not disclosed; pictures/podcasts do not currently support uploading to the topic function, and only PDF/DOCX/PPTX and other document formats are supported.

How to use Secret Tower AI search

MiTa AI search uses the Web as the main entrance and also provides API interfaces for enterprise-level integration. The usage path can be broken down into four stages:

How to use	Suitable scenarios	Key configuration	Threshold
Web version	Personal daily search and research	Visit metaso.cn, support WeChat/mobile phone number registration	Zero threshold
Browser default search engine	Seamless replacement of Baidu/Google for high-frequency users	Chrome settings: URL format is https://metaso.cn/, q=%s	One-time configuration
API interface	Enterprise knowledge base docking and batch search	Business required to obtain API Key and frequency control strategy	Business confirmation required
Privatized deployment	Industries with high compliance requirements	Business confirmation required for deployment plan and resource requirements	Business confirmation required

Typical steps:

1. Input question: Enter a natural language question (such as "China's AI search market size in 2025") in the search box, and select the corresponding search mode (concise/in-depth/research).
2. View structured results: Browse AI-generated answers, mind maps, outlines, or tables, and click on the source link to verify the information.
3. In-depth questioning and scope limitation: You can refine your requirements by asking questions, or limit the search scope to the created "topic" knowledge base to obtain more accurate answers.
4. Export and Creation: Export the results to Word/PDF/PPT, or directly connect to Secret Tower Writing Cat for polishing and expansion.

Human-machine collaboration boundary (mandatory confirmation point):

• Information retrieval that can be 100% automated** includes daily encyclopedia queries, news summaries, preliminary literature screening, and data table extraction. These segmented outputs can be directly output by AI, and humans only need to quickly browse and confirm.
• Scenarios where manual confirmation points must be set: legal contract risk scanning results (AI-marked risk clauses need to be reviewed by lawyers one by one), financial analysis indicator extraction (key figures need to be cross-checked with source documents), academic reference citations (AI-generated citations are in the correct format, but the original sources of references need to be manually confirmed). In these chapters, the role of the secret tower is an "accelerator" rather than a "decision-maker".

Product Pricing of Secret Tower AI Search

The pricing model is subject to the official real-time page. Usually a freemium or subscription system is used, and basic functions can be used for free. Advanced functions or high-frequency use require paid subscriptions, and users are advised to evaluate the optimal solution based on actual usage.

Application scenarios of Secret Tower AI search

The implementation scenarios of MITA AI search are concentrated in Chinese knowledge workflows with "large amount of information, high timeliness, and need for structured output". The core can be summarized into four categories:

• Academic Research and Literature Review: Covers hundreds of millions of Chinese and English academic documents (including PubMed, SCI journals), supports literature screening, review framework generation, automatic reference typesetting and experimental data visualization. Typical Benefits: The time for initial literature screening and first draft review is reduced from 6 hours to approximately 2 hours. Verification focus: Although the AI-generated review framework can be used directly, it is still recommended to manually confirm its relevance and timeliness in the original papers cited.
• Legal and Compliance Scenario: Legal practitioners can use Secret Tower to scan contract risk clauses (term recognition accuracy is 92%), search for similar case precedents, and query the latest regulations. Typical benefits: The efficiency of similar case retrieval is increased by about 85%, and the initial contract review time is reduced from 2-3 hours to about 0.5 hours. Key points of verification: Contract risk scan results cannot replace a lawyer's professional judgment, and the "risk clauses" marked by AI need to be reviewed item by item.
• Financial and Market Analysis: Supports the extraction of key financial report indicators (revenue growth rate, gross profit margin, debt ratio, etc.), automatically generates analysis briefings, and aggregates competitive product information. Typical benefits: The first draft of the financial report analysis briefing can be completed in 15-20 minutes, and traditional manual editing takes 2-3 hours. Key points of verification: Key financial figures are recommended to be cross-checked with the original financial reports.
• Operation and Content Creation: Operators can use in-depth mode to search market trends and competitive product dynamics, and quickly complete content output through the link of "Search→Export PPT→Write Cat Polish". Typical Benefits: Competitor analysis report reduced from 4 hours to 1-1.5 hours. Key points of verification: Key arguments such as market share figures of data sources need to be manually confirmed for their statistical caliber and timeliness.

Applicable groups for Secret Tower AI search

The "search + knowledge management" dual engine design of Secret Tower AI Search serves four types of core roles. Each type of role has different entry points and value focus for using Secret Tower:

• Academic researchers (college students, graduate students, scientific researchers): The value of Secret Tower lies in "search is review". Traditional academic search requires manual switching of databases, filtering documents, and organizing citation formats. Secret Tower compresses these steps into a single interface. Unfit Boundary: Scenarios that require in-depth analysis of original experimental data or professional statistical modeling, Secret Tower is not a tool to replace SPSS/Matlab.
• Legal and Financial Practitioners (Lawyers, Legal Affairs, Analysts): Terminology identification and structured output in vertical fields make Secret Tower an "assistant-level" tool, but as mentioned above, AI output cannot replace professional judgment. Prerequisite: The team needs to have established a standard process of "AI assistance + manual review", otherwise compliance risks may arise.
• Operations and Strategy Staff (New Media Operations, Market Research, Competitive Product Analysis): The high-frequency information monitoring and report writing scenarios are highly consistent with the "Search→Structure→Export" link of Secret Tower. Not suitable for the boundary: In scenarios that require first-hand interview data or field research, the Secret Tower cannot replace original data collection.
• Knowledge management practitioners (individual learners, team knowledge base managers): The topic function provides a hybrid retrieval mode of "private knowledge base + full network search", which is suitable for long-term tracking of a certain field. Not suitable for boundaries: For in-depth development scenarios that require complete API orchestration, complex integration of multiple systems, or enterprise-level RBAC permissions governance, Secret Tower's current enterprise delivery capabilities still require business confirmation.

Summary and Outlook

The core competitiveness of MITA AI Search lies in "using AI to reconstruct the complete link of Chinese information acquisition" - from semantic search to structured output, from academic review to knowledge management, it attempts to accomplish within one product what traditional search engines require the cooperation of multiple tools. Its self-developed MetaLLM base, combined with RAG and MoE architecture, provides verifiable information processing capabilities in Chinese vertical scenarios (academic, legal, financial). This is its core difference from general AI search.

Current limitations and uncertainties: ① API pricing and frequency control strategies are not open and transparent, and developer integration has unpredictable costs; ② The delivery conditions SLA and compliance certification (such as SOC2, MLA) for enterprise-level privatized deployment are not fully disclosed; ③ The capability boundaries of multi-lingual hybrid search and image analysis are not yet clear; ④ The open source ecosystem and third-party plug-in system have not yet been established, and there is a gap in scalability compared with competing products.

Purchase and Adoption Risk Assessment: For academic researchers, individual knowledge workers, or operational teams, Secret Tower’s free quota can cover daily needs and can be used directly from the web without any upfront investment. For teams that plan to embed it into formal business processes or corporate knowledge bases, it is recommended to conduct a small-scale pilot with 1-2 processes (such as preliminary review of legal contracts, automation of weekly competitive product reports). During the pilot period, focus on verifying three dimensions: the accuracy of AI output on real business data, the stability and delay of the API during peak periods, and the agreement on data deletion and compliance auditing in the business terms. Before the official disclosure of enterprise-level delivery details, it is not recommended to use it as the only information entry for core compliance links.

Related tools: perplexity, you-com

How to use Secret Tower AI Search

• Web client: You can use it by visiting the official website and registering an account. Most functions do not require installation.
• API Access: Provides RESTful API, developers can obtain the API Key and integrate it into their own applications.

Version Info

• In-depth research mode upgraded version ：The homepage continues to emphasize in-depth research, fact-checking, and voice input and output capabilities. （2026-06）
• AI search initial public version ：Launched Chinese AI search capabilities, providing an ad-free direct results experience. （2024-01）