Documentation
This documentation explains the current platform architecture, the roles agents play, how state is persisted, how models are routed, and which capabilities are still on the roadmap.
Current architecture and product boundaries.
Wotai is an AI software engineering platform. The marketing website at wotai.io is separate from the application at app.wotai.io. The application owns accounts, workspaces, projects, project versions, workflow jobs, previews, and source exports.
The current application stack separates the web UI, FastAPI control plane, Redis job dispatch, LangGraph runtime, AI Gateway, MySQL business data, and PostgreSQL checkpoint storage. This separation keeps orchestration state distinct from account and project data.
The workflow uses specialized roles including Supervisor, Requirement, Product, Architect, UX, UI, Frontend, Backend, Database, DevOps, Security, QA, and Reviewer. A project does not necessarily need every agent; the orchestration graph can route work according to project requirements.
The AI Gateway abstracts providers from agents. Role-level chains can use Gemini, DeepSeek, OpenAI, Groq, Mistral, Qwen, OpenRouter, Cerebras, Ollama, or a deterministic Mock provider for platform testing. Profiles such as test, free, economy, quality, and local make routing easier to manage.
SiteSpec: structured websites can use a validated SiteSpec rendered through trusted components. Full App: richer application projects can generate a source repository through planning and file-oriented implementation stages. Source generation and autonomous code execution are separate concerns.
LangGraph checkpoint persistence stores workflow state so eligible jobs can continue from saved execution state. A checkpoint does not guarantee a successful result; it provides durable state for recovery and controlled rework.
AI provider keys remain behind the AI Gateway. Internal databases and service ports should stay private. Agents should not receive unrestricted Docker or host access. Generated code execution should happen only in a dedicated isolated sandbox. Security and QA review are independent workflow stages.
The recommended production topology is Ubuntu with Docker, HTTPS at the edge, and internal-only service networking. The LangGraph runtime is installed inside its container, not directly on the host operating system. The same containerized layout can be used locally and on the VPS.
Current priorities include a dedicated Sandbox Engine for isolated dependency installation, builds, linting, tests, security scans, and runtime health checks. Roadmap items should not be treated as production capabilities until they are enabled and validated.
Is Wotai just a website builder? No. Structured websites are one generation path; the platform is also designed around full web-application workflows. Is Wotai tied to one AI provider? No. The AI Gateway is designed to support multiple providers and local models. Does LangGraph need to be installed separately? In the current architecture, LangGraph is installed inside the dedicated runtime container and the backend communicates with that internal service over the private application network. Can I export generated source? The application architecture includes source export for supported generated projects and versioned outputs.
Wotai / Early access