ThinkingAI

An enterprise-level AI Agent platform that can be deployed privately, allowing enterprises to have a collaborative Agent team

Agent Workflow

https://thinkingai.cn/

Visit Website

In-depth Report

ThinkingAI Agentic Engine is an enterprise-level AI Agent platform that can be deployed privately and will be officially released in April 2026. The core positioning is to allow enterprises to have a team of agents that can collaborate and realize a fully closed loop from problem discovery to problem solving through the three-layer capabilities of "sensing, knowing, and doing". ThinkingAI, formerly known as Shushu Technology, has been working in the field of data intelligence for ten years, serving more than 15,000 companies, and has established a strategic partnership with MiniMax.
ThinkingAI was formerly known as the data intelligence company "Shushu Technology". In 2026, its brand was upgraded and its positioning shifted from a data analysis tool to an enterprise-level Agent platform. Since its establishment in 2015, the company has long focused on user data analysis in gaming, social networking, e-commerce and other industries, and has served more than 15,000 corporate customers and more than 8,000 products. A press conference will be held at the Computer History Museum in Silicon Valley on April 16, 2026, to announce a strategic cooperation with MiniMax. Agentic Engine focuses on the privatized AI deployment needs of medium and large enterprises, forming differentiated competition with general agent platforms such as ByteDance’s “Button Space” and Alibaba’s “Tongyi”.
The design philosophy of Agentic Engine revolves around three dimensions. "Sense" means global awareness, monitoring all channel signals 24/7, including abnormal business data, complaints from social media users, negative reviews on the App Store, etc. "Knowledge" means in-depth understanding. Not only do you know what happened, but you can also break down which channel, which version, which type of user, and understand why it happened. "Action" is a closed loop of action. It directly generates strategies and executes automatic A/B tests without manual scheduling, realizing a complete closed loop from decision-making to action.
The platform equips medium and large enterprises with an entire collaborative Agent team, rather than assigning each person an AI assistant. Specifically, it includes data analysis Agent (the "eyes" of the team), A/B experiment Agent (the "referee" of the team, shortening the experiment cycle from 2-4 weeks to no manual intervention), intelligent operation Agent (the "hand" of the team, compressing the operation cycle from "weekly" to "real-time"), and autonomous creation of Agents (can be created by clicking and dragging, without writing code).
The system realizes collaboration between Agents through the strategy layer, orchestration layer and execution layer. In the strategy layer, the unified Orchestrator is responsible for discovering opportunities and validating hypotheses. The orchestration layer is responsible for task scheduling, status management and context sharing. The execution layer runs multiple business agents in parallel. The execution results are automatically returned to the strategy layer to form a more accurate next round of insights.
This is the core barrier built by ThinkingAI’s ten years of accumulation. The first layer is the Agent memory system, which structures implicit business knowledge through the semantic layer and knowledge graph, allowing the Agent to understand questions such as "How to calculate DAU" without the need for repeated explanations by business personnel. The second layer has more than 100 industry skills preset, covering 8 major areas including user analysis, retention analysis, payment analysis, delivery analysis, operation analysis, report generation, attribution analysis, and public opinion analysis. The third layer of continuous evolution mechanism allows the results of each execution to be precipitated into new knowledge, and the Agent becomes more accurate as it runs.
The platform has multiple enterprise-level security mechanisms built into it. Sandbox isolation ensures that new Agents can be tested in the sandbox, A/B grayscale comparison verifies new and old Agents, full-link observability ensures traceability of every step, and hallucination detection runs through the entire link to prevent erroneous output. Combined with privatized deployment, the entire data and model link is autonomously controllable and compatible with compliance requirements such as GDPR and CCPA.
It has supported mainstream office platforms such as Feishu, DingTalk, Enterprise WeChat, Slack, and Discord. Natively supports MCP and A2A protocols and can be seamlessly connected with any AI platform. The large model base supports MiniMax (private deployment) and other mainstream large model APIs.
Agentic Engine currently does not disclose the specific pricing plan and adopts the "contact sales to obtain a quote" model. The target customers are medium and large enterprises with both standard SaaS and private deployment models.
Judging from public information, enterprise users have high evaluations of its "fully closed-loop" capabilities, believing that it truly solves the "last mile" problem from data to action. Ten years of industry know-how is considered a barrier to differentiation. Preset Skills are considered to lower the barrier to use. On the negative side, the platform has high requirements for enterprise data infrastructure, making it difficult for small and medium-sized enterprises to adapt. Initial Agent training and configuration requires a certain amount of industry knowledge accumulation.
Jiazi Guangnian and other institutions are paying attention to the release of ThinkingAI Agentic Engine. Its "Agentic Engine" positioning and "privatized deployment" strategy are considered to be differentiated competitive strategies in the current AI Agent market. The strategic cooperation with MiniMax is regarded as a typical case of collaboration between domestic AI Agent platforms and large model manufacturers. The enterprise-level AI Agent market is expected to grow from 147.3 billion yuan to 3.3 trillion yuan in 2025-2028, with a compound annual growth rate of more than 100%.
The product will be officially released in April 2026, and the product maturity and stability still need to be verified by the market. Competition in the domestic enterprise-level AI Agent market is fierce, and privatized deployment has certain requirements on the IT capabilities of enterprises.
This product is suitable for medium and large Internet companies and game companies that have a certain scale of operations and data analysis teams and have clear business needs for AI Agents. Typical usage scenarios include refined user operations, real-time data-driven operational intervention, A/B test automation and digital accumulation of industry know-how.
ThinkingAI Agentic Engine is an enterprise-level AI Agent platform with clear positioning and complete capabilities. Its three-layer capability system of "sense, knowledge and action" and ten years of industry know-how accumulation constitute a differentiated advantage. A strategic partnership with MiniMax provides reliable base support for large models. The success of the platform will be verified in the cases of enterprise implementation in the next 1-2 years.

User Reviews

Aiden100

—

I just watched the Silicon Valley conference. The idea of perception is so awesome. Even the comments on Discord and App Store can be automatically monitored. Data analysis is finally not dead data.

VAb_en

—

It feels more suitable for enterprises than Button Space, and the privatized deployment is very convenient.

DeltaDefiLarsen

—

I tried it and found that the data analysis agent does have something. I asked it directly why the retention rate dropped. It can break down the specific channels and versions. Unlike before, I just threw a SQL and waited for half a day.

Frank.Hall_2023

—

The three-layer knowledge system is indeed a strong barrier and cannot be copied by others.

ElizabethMiller_2020

—

A/B experiment Agent does not need to be scheduled, and the operation will laugh to death

Sophia_Anderson_2022

—

The cooperation with MiniMax feels very reliable, and the base of the large model is guaranteed.

Jonathan.Cruz0075

—

The skills accumulated over ten years are indeed not fake. Use the preset templates directly without having to adjust them from scratch.

LindaWhite_99

—

To be honest, privatized deployment is very important to enterprises. We have always had the need to prevent data from leaving the intranet, and now someone has finally done it.

孔浩娜

—

The design of sandbox isolation is very stable. The new Agent runs in the sandbox first, and then goes into production. This is very necessary.

石霞

—

The Agent team is not to assign an assistant to each person, but to assign a team to the enterprise. This idea is very correct.

wm68n762u

—

After reading 36 Krypton’s article, I felt that it contained a lot of information. I will study it carefully when I go back.

whitepeacock954

—

The concept of action closed loop is well mentioned. Previously, I had the data but didn’t know how to act.

Cr_yptoTrader198

—

Feishu integration is supported, and it should be much easier to deploy.

KThomasQ

—

Our company's data infrastructure is pretty good and should be able to run

SarahStewartZ

—

I don’t know the price yet, I’ll wait for the official quote.

PatriciaMartinX

—

To be honest, this kind of product may not be affordable for small and medium-sized enterprises, but it is still suitable for large manufacturers.

DeborahCarter_77

—

Fully closed loop yyds

Stephen_JamesJr

—

Upgraded from Shushu Technology, the brand upgrade has been very thorough.

e2af3v0af0

—

The mechanism of continuous evolution is very interesting. The more the Agent runs, the more accurate it becomes. I don’t know how to do it.

ThinkingAI

In-depth Report

User Reviews

Similar Products