TeamAI simulates a company workforce with AI agents. Instead of chatting with a single assistant, you assign work to a structured team: Founder, Marketing Head, SEO Analyst, Content Creator, Data Analyst, Market Researcher, and Evaluator.

The workflow is delegation-first. The Founder routes the task, specialists execute with tools, and an Evaluator acts as a quality gate before final delivery. If quality is low, the system loops with feedback and revision.

Built as a full-stack product:

• Backend: Python + FastAPI for async APIs and orchestration
• Agent system: OpenAI Agents SDK with explicit handoffs
• Frontend: Next.js + TypeScript
• UI: Tailwind CSS + Radix UI
• State: Zustand
• Streaming: Server-Sent Events (SSE) for live agent updates

The backend emits real-time events like AGENT_CHANGE, TOOL_CALL, TOOL_RESULT, DELTA, and COMPLETE, so the frontend can show progress as the task runs. This makes the agent tree and activity feed feel alive instead of waiting for one final blob of text.

Data is company-driven via JSON files. Tools read business context such as SEO data, market research, brand assets, templates, and analytics. Adding a new company is mostly data onboarding; adding a new agent is a clean extension of the team graph and handoff chain.

I also built a CLI on top of the same orchestration core, so the exact same multi-agent engine works in both terminal and web interfaces.