Coze
Build and publish AI agents with tools, knowledge bases and visual workflows, plus bundled office features.
Quick facts
| Pricing | Freemium |
|---|---|
| Free tier | Yes — free tier with limited messages and bot count |
| Platforms | Web, API |
| Developer | ByteDance |
| Website | https://www.coze.com/ |
What Coze is
Coze is a platform for building AI agents and chatbots without writing a back end. You define an agent with instructions, give it tools — web search, image generation, a knowledge base, custom APIs — and publish it to a chat interface, a messaging app or an API. A visual workflow builder handles multi-step logic, and a plugin marketplace supplies integrations. The current product also bundles office features: writing assistance, slide generation and simple web page design. It is aimed at people who want an assistant that does a specific job rather than a general chat window. The limits are depth and portability: complex logic becomes awkward in a visual builder, published agents depend on the platform runtime, and data terms differ between the international and Chinese versions. Agents can be tested against sample conversations before publishing, and each bot keeps its own knowledge base separate from other projects in the workspace.
Best for
- Publishing a support or FAQ bot quickly
- Agents that need web search and a knowledge base without code
- Teams testing an assistant before investing in engineering
Not ideal for
- Complex stateful logic that needs real code
- Deploying the same agent outside the platform runtime
Pros
- Agent, tools and knowledge base in one builder
- Publish targets include messaging apps and APIs
- Free tier covers real prototypes
Cons
- Visual workflows get awkward for complex logic
- Agents depend on the platform runtime
- Data terms differ between regional versions
Pricing in detail
The free tier includes a monthly message allowance, a limited number of agents and basic model access. Paid tiers are billed monthly and raise message quotas, add stronger models, more knowledge-base storage and team collaboration; enterprise terms are quoted. Because the meter is messages rather than seats, a popular published bot can exhaust a plan quickly. Check the quota against expected conversation volume before launch.