Sentoe
BlogProjectsGalleryAbout
中
Sentoe
© 2026 SentoePowered by Next.js & Notion
March 18, 2026AI Product Analysis13 min read

Coze (Kouzi) Product Research Report

阅读中文版

I. Product Overview

1.1 Product Introduction

Coze (known as "扣子/Kouzi" in China) is a next-generation, all-in-one AI Bot development platform launched by ByteDance. It officially went live on February 1, 2024, with the goal of enabling non-developers to quickly create, debug, and optimize chatbots (intelligent agents). The international version (coze.com) was first launched at the end of 2023, offering AI chatbot creation based on OpenAI's GPT-4 and GPT-3.5 APIs, while the domestic version (coze.cn) is powered by ByteDance's proprietary Doubao large language model.

1.2 Product Positioning

Coze is positioned as a platform that "empowers everyone to create AI assistants." Unlike Dify (a low-code AI application development framework) and n8n (a workflow automation tool), Coze specifically targets non-technical users. Through visual drag-and-drop interfaces and natural language configuration, users can build fully functional AI agents and deploy them with one click to platforms such as Douyin (TikTok China), Feishu (Lark), WeChat, and WeChat Official Accounts.

The upgraded Coze 2.0 inherits the core capabilities of its predecessor while introducing the Agent Skills system and Agent Plan long-term planning features, achieving a transformation from a single tool to an intelligent work companion. The product has evolved from an initial "conversational Bot platform" to a comprehensive "AI intelligent office platform."

1.3 Product Development Timeline

The Coze team's internal project codename was "Bot Studio." It first launched overseas at the end of 2023 under the name "Coze," followed by a domestic launch in early 2024 under the Chinese name "扣子" (Kouzi). Coze started with zero marketing budget but experienced rapid user growth. Its evolution can be divided into the following phases:

Phase 1 (Late 2023 – Early 2024): Conversational Bot Platform. Coze entered the market as a "no-code chatbot building platform," where users only needed to write prompts while the platform handled all the complex backend work.

Phase 2 (Mid-2024 – Early 2025): Workflow-Driven Growth. The Coze team began aggressively developing the Workflow feature. They subsequently introduced clearer monetization paths, achieving their initial six-month targets ahead of schedule.

Phase 3 (April 2025 – Present): AI Office Platform. ByteDance launched Coze Space, a new AI collaborative office platform, which entered beta testing on April 19, 2025. It aims to become the premier workspace for human-AI Agent collaboration.

Phase 4 (July 2025): Open-Source Ecosystem. On July 26, 2025, ByteDance officially open-sourced two core projects of its Coze platform: Coze Studio and Coze Loop.


II. Core Feature Analysis

2.1 Agent Development

The platform encompasses agent development, resource management, and publishing capabilities, supporting three agent modes: autonomous planning, conversational flow, and multi-agent orchestration. It optimizes AI performance through large model principles, parameter configuration, and prompt engineering, while integrating plugins, knowledge bases, and databases to extend functionality.

The platform includes built-in debugging tools to help users identify and correct potential issues, ensuring smooth bot operation, and provides optimization tools to enhance bot capabilities and performance. Coze offers the following core development capabilities:

No-Code Development. Coze provides a visual interface where users can complete AI application development through simple drag-and-drop and configuration operations without writing complex code, significantly lowering the development barrier.

Multimodal Capabilities. Coze integrates text, image, and API call processing capabilities across multiple modalities, supporting intelligent requirements in complex scenarios.

Model Integration. Coze supports virtually all mainstream large language models on the market, including ByteDance's Doubao, Alibaba's Qwen (Tongyi Qianwen), Zhipu's GLM, MiniMax, Moonshot's Kimi, and Baichuan Intelligence's Baichuan4.

2.2 Workflow Design

Coze's workflow feature is designed to handle logically complex tasks with high stability requirements. The platform provides a wide range of flexible, composable nodes including LLM nodes, custom code blocks, conditional logic, and more. Regardless of programming background, users can quickly build workflows through drag-and-drop operations.

2.3 Plugin System

Coze features over 60 built-in plugins spanning multiple domains including news reading, travel, productivity, and multimodal models. Users can select appropriate plugins to extend agent capabilities based on their needs. Additionally, Coze supports integration of private APIs into agents, further expanding functional boundaries.

2.4 Knowledge Base

Coze provides an intuitive knowledge base feature to help users manage and store data. Bots can leverage the constructed knowledge base to deliver more precise interactions with users. Users can upload local files or crawl online web content to enrich the bot's knowledge base and improve its response quality.

2.5 Cross-Platform Publishing

Created agents can be published with one click to social platforms such as WeChat, Douyin, Feishu, and Discord, or to enterprise internal systems, making it easy for users to share and distribute their creations.

2.6 Coze 2.0 New Features

Agent Office (Deep Workplace Understanding) provides in-depth analysis and insights when handling complex workplace issues. Agent Coding (Cloud Development Platform) is a one-stop cloud development platform that supports building agents, workflows, and websites through conversational commands, with one-click deployment. Video creation skills are also a significant addition, allowing users to describe their needs in natural language and have AI automatically generate videos.

2.7 Coze Space

Coze Space is a continuation and enhancement of ByteDance's Coze AI Agent platform. If Coze is the "development platform" for AI Bots, then Coze Space is more akin to a "workspace for AI Agents."

The platform integrates the MCP (Model Context Protocol) extension framework, enabling seamless connection and invocation of external services and tools. The first batch of officially supported services includes Feishu Multidimensional Tables, Amap (Gaode Maps), image tools, and text-to-speech, among other high-frequency office utilities.

2.8 Open-Source Components

Coze Studio is the open-source version of ByteDance's next-generation AI Agent development platform. Through visual design and orchestration tools, developers can rapidly build and debug agents, applications, and workflows using zero-code or low-code approaches.

Coze Loop provides full lifecycle management capabilities, helping developers more efficiently develop and operate AI Agents. Whether it's prompt engineering, AI agent evaluation, or post-launch monitoring and optimization, Coze Loop delivers powerful tools and intelligent support.


III. Market Data and User Scale

According to Feifan Research data, as of June 2025, Coze's monthly active visitors reached approximately 4.58 million. As of 2025, over 500,000 developers have created more than 2 million agents on Coze, covering 30+ industries including finance, healthcare, and education. The platform currently hosts over 1 million active agents spanning content creation, office automation, customer service, education, and other domains.

Within just two days of open-sourcing, the project garnered over 6,000 stars on GitHub. Among major tech companies including ByteDance, Baidu, and Tencent competing for developer ecosystems through open platforms, Coze currently holds a leading position.


IV. Business Model and Pricing Strategy

4.1 Pricing Structure

Coze's international pricing structure includes three tiers: Free, Premium, and Premium Plus. The domestic version offers a more granular subscription system, including a free Personal tier, Personal Advanced tier, Team tier, and Enterprise tier.

The platform underwent a transition from free to paid services. In March 2025, the pricing policy took effect, with the base plan increasing from ¥9.9 to ¥39.9 per month, alongside a shift to a credits-based resource system.

4.2 Monetization Path

This transition aims to explore AI service monetization while sustaining platform operations. Coze's commercialization follows the classic "freemium + value-added paid" SaaS model, accumulating users and developer ecosystem through a free tier, then monetizing through premium features and higher-usage paid subscriptions. In 2025, the Enterprise edition introduced private deployment capabilities, supporting integration with domestic large models such as ByteDance's Doubao and Alibaba's Qwen, meeting data compliance requirements in industries like finance and government affairs.


V. Competitive Analysis

5.1 Competitive Landscape

The current AI agent development platform market presents a diverse competitive landscape, with key competitors including Dify, FastGPT, Tencent Yuanqi, Baidu AppBuilder (Lingjing), n8n, RAGFlow, and Manus.

5.2 Coze vs. Dify

Coze is more user-friendly, while Dify offers greater functional depth. Users seeking to deeply explore workflows would prefer Dify, whereas those looking to easily embed workflows into other platforms would find Coze ideal.

Coze comprises two main components: Coze Studio for designing and building AI applications, and Coze Loop for application runtime monitoring and continuous optimization. Coze adopts a modular microservices architecture based on microservices and Domain-Driven Design (DDD) principles. In contrast, Dify's architecture is designed as a tightly integrated system, with all core functions bundled together and served through a unified API and dashboard. This monolithic design significantly simplifies deployment and management complexity.

5.3 Coze vs. n8n

As a ByteDance product, Coze's greatest advantage lies in "ByteDance ecosystem integration" and a "lightweight user experience." It is deeply integrated with ByteDance products such as Douyin, Feishu, and CapCut (Jianying), making the onboarding process extremely low-friction for users within the ByteDance ecosystem. In contrast, n8n focuses on general-purpose workflow automation orchestration and is better suited for technically proficient users.

5.4 Coze vs. Tencent Yuanqi and Baidu AppBuilder

Coze offers the most extensive skill set, including plugins, workflows, image flows, triggers (scheduled tasks), knowledge bases, and memory capabilities. Tencent Yuanqi's skills include plugins, workflows, and knowledge bases. Both ByteDance's Coze and Tencent's Yuanqi maintain aggressive update cycles to stay competitive, with version iterations averaging 2–3 updates per month.

5.5 Competitive Positioning Summary

DimensionCozeDifyFastGPTTencent Yuanqi
Core PositioningZero-code AI agent platformLLMOps development frameworkKnowledge base Q&A systemAI agent creation & distribution
Target UsersNon-technical users / PMsDevelopers / Technical teamsTechnical usersCreators / Enterprise users
DeploymentSaaS-primary + Enterprise privateOpen-source self-hosted + SaaSOpen-source self-hostedSaaS
Ecosystem AdvantageByteDance (Feishu / Douyin)Open-source communityOpen-source communityWeChat / QQ social ecosystem
Learning Curve★☆☆☆☆★★★☆☆★★★★☆★★☆☆☆

VI. User Profiles and Application Scenarios

6.1 Target Users

Coze serves a broad user base. Developers can focus on tuning AI models and prompts to optimize performance. Enterprises can integrate AI chatbots into internal systems through the platform. General users, even those without coding knowledge, can quickly create, debug, and deploy their own AI chatbots.

The target audience includes AI beginners, product managers, operations personnel, creators looking to quickly build personalized AI agents, and individuals and small teams with limited budgets and technical resources.

6.2 Typical Application Scenarios

As ByteDance's intelligent agent development platform, Coze supports both single-agent and multi-agent modes, capable of efficiently handling complex logic chains. Typical scenarios encompass enterprise customer service automation, content moderation pipelines, educational tutoring bots, content creation assistants, and office automation.

Finance, government affairs, and education have emerged as the top three sectors for agent applications, accounting for 38%, 25%, and 18% of market share, respectively.


VII. SWOT Analysis

7.1 Strengths

Powerful ByteDance Ecosystem Synergy. Coze not only supports distribution of developed agents to ByteDance products such as Feishu, Juejin, and Douyin, but also allows distribution to WeChat Customer Service, WeChat Service Accounts, and WeChat Subscription Accounts, forming a comprehensive distribution network.

Extremely Low Barrier to Entry. Coze emphasizes "building automated workflows with natural language," allowing users to quickly create tasks through conversational commands.

Rich Plugin and Template Ecosystem. Coze's plugin marketplace is highly active, with a comprehensive range of functional plugins readily available.

Open-Source Strategy Enhances Influence. The open-sourcing marks a significant step in ByteDance's AI strategy and provides both enterprises and individual developers with low-barrier, highly flexible AI agent development tools.

7.2 Weaknesses

Model Capability Limitations. Coze Space relies on domestic models (Doubao/DeepSeek), which are noticeably weaker than Claude/GPT on complex tasks.

Insufficient Technical Depth. Coze's model configuration process requires users to possess considerable technical knowledge, which presents a barrier for non-technical users. At the same time, for advanced developers, the platform's customization capabilities and flexibility are relatively limited.

Free-to-Paid Transition Growing Pains. For users who heavily relied on Coze's free access to high-end models, the shift to a paid model represents a significant change.

7.3 Opportunities

AI Agent Market Explosion. 2024 was hailed as the "Year of the Agent," and 2025 marks the beginning of practical agent deployment. The intelligent workflow tools market is projected to reach $50 billion by 2027.

Enterprise Market Expansion. In 2025, Coze launched its Enterprise edition with private deployment support and customizable model integration, signaling an upgrade from a "personal tool" to an "enterprise-grade AI agent platform."

Open-Source Ecosystem Building. In the face of intense competition, the industry has reached a broader consensus: the future Agent ecosystem will be defined by thousands of developers, not dominated by a single company. Open source is the best path to directly engage global frontline developers for feedback, code contributions, and invaluable innovative ideas.

7.4 Threats

Intense Competitive Landscape. Coze's open-sourcing places it in direct competition with agent development platforms such as Dify and FastGPT.

User Retention Challenges. Novelty alone cannot sustain long-term value. The number of bots that are genuinely used on an ongoing basis remains very limited, and retention is clearly suboptimal.

Insufficient Industry Maturity. General-purpose AI agents appear capable of handling everything but excel at nothing specific. Agent deployment in real-world applications continues to face numerous challenges.


VIII. Strategic Recommendations

8.1 Deepen Vertical Scenarios and Enhance Delivery Value

It is recommended that Coze develop industry-specific solution templates for high-value sectors such as finance, education, and government affairs, built upon its general platform foundation. This would reduce cold-start costs for enterprise clients and improve the delivery quality of agents in real business scenarios.

8.2 Strengthen Open-Source Ecosystem Development

The Coze Studio project rapidly accumulated thousands of GitHub stars, demonstrating high developer enthusiasm. It is recommended to increase community operations investment, establish developer incentive mechanisms (such as contribution leaderboards and developer competitions), and promote the proliferation of third-party plugins and templates.

8.3 Optimize the Free-to-Paid Conversion Funnel

The current credits-based pricing model presents a relatively high cognitive cost for average users. It is recommended to simplify billing explanations, introduce more intuitive pricing models such as "pay-per-result" (e.g., per agent invocation), and maintain a competitive free tier to continuously attract new users.

8.4 Deepen Integration with the ByteDance Ecosystem

Coze should further deepen its integration with ByteDance's core products — Douyin e-commerce, Feishu office suite, and CapCut content creation — to build a differentiated "AI-native workflow" experience, making agents an indispensable foundational capability layer within the ByteDance product family.

8.5 Enhance Model Capabilities and Multi-Model Orchestration

Domestic models still lag behind international leading models on certain complex tasks. It is recommended that Coze continue expanding the range of integrable models and implement intelligent multi-model routing at the platform level, automatically selecting the optimal model based on task type to improve overall agent output quality.


IX. Conclusion

The ultimate challenge Coze must address is making agents truly usable and deployable in real-world scenarios. As the frontrunner among domestic AI agent development platforms, Coze has established a significant first-mover advantage, leveraging ByteDance's powerful technology infrastructure and traffic ecosystem.

From a product evolution perspective, Coze has undergone multiple strategic upgrades — from "conversational Bot → workflow platform → AI office workspace → open-source ecosystem" — demonstrating exceptional adaptability and iteration speed. By delivering genuine value to users, Coze has transformed its "early start" into a "sustained lead."

Looking ahead, against the backdrop of the continued AI Agent market explosion, whether Coze can establish a firm foothold in the enterprise market, build technical influence through its open-source ecosystem, and maintain product innovation amid increasingly fierce competition will determine whether it can transition from "leading" to truly "winning."


Report Date: March 18, 2026

PreviousBuilding a Knowledge Management System with NotionNextDoubao (豆包) Product Research Report

Table of Contents

0%