Coze

Coze is a no-code platform for building artificial intelligence agents and chatbots, developed by the Chinese technology company ByteDance, best known as the owner of TikTok and Douyin. Coze lets users assemble conversational assistants from natural-language prompts, tools, knowledge sources and visual workflows, and publish them to messaging platforms and websites without writing application code. It is one of the most widely used consumer-accessible agent builders, alongside platforms such as OpenAI's GPT suite, Microsoft Copilot Studio and the open-source Dify.^[1]

History

ByteDance launched Coze in China in 2023 under the name 扣子 (Kòuzi), and released an international version at coze.com in early 2024 with English and other language support. The international launch extended the platform's model choices — in addition to ByteDance's own Doubao models, international users could connect frontier models from providers such as OpenAI and Anthropic — and positioned Coze as a direct competitor to Western no-code AI platforms. Through 2024 and 2025 ByteDance expanded the product with richer workflow tooling, enterprise features, an expanded plugin ecosystem and deeper integration with its own agent and model stack, while the Chinese version evolved in parallel with features tailored to local platforms such as Feishu and WeChat. By 2026 Coze reported a large base of creators and businesses building and publishing agents on the platform.^[1][2]

Key Concepts and Technology

Coze is organised around a visual, template-driven development model. A builder defines an agent's persona and instructions in natural language, then adds capabilities from a marketplace of plugins — prebuilt connectors to web search, image generation, translation, messaging, productivity and third-party services — or writes custom plugins. For domain-specific answers, agents connect to knowledge bases of uploaded documents that are retrieved at runtime using retrieval-augmented generation (RAG), so a customer-service agent can answer from a company's manuals and policies.

More complex behaviour is assembled in a workflow canvas, where users connect nodes — model calls, conditionals, knowledge retrieval, code execution, HTTP requests — into a graph, and in multi-agent arrangements where a primary agent routes requests to specialist sub-agents. Conversations can be persisted in a built-in database for long-running memory, and agents expose variables that let developers track user state across sessions. Finished agents are published to channels such as web chat widgets, Discord, Telegram, Line and other messaging services, or called through an API from existing applications. The platform also provides testing consoles, analytics and version management, aiming to cover the full lifecycle from prototype to production without a dedicated engineering team.^[2][3]

Applications and Impact

Coze is used for a wide range of conversational and automation workloads: customer-service bots that answer from company knowledge bases, marketing and lead-qualification assistants, community moderators on messaging platforms, personal productivity agents, and internal tools that wrap company data for employees. Its accessibility has made it a common entry point for non-programmers — small-business owners, marketers and students — who want an AI assistant without hiring developers, while its API and enterprise tiers serve companies that need governance, scale and integration with existing systems. The platform's main limitations, relative to code-based frameworks, are reduced control over the underlying orchestration and dependence on a hosted cloud service, considerations that matter for organisations with strict data-residency or customisation requirements.^[1][3]

>See Also

• Dify
• Agentic AI
• Chatbot
• Retrieval-Augmented Generation
• Doubao
• ByteDance