Build new product ideas with an AI "studio" that works like a small team. You chat in everyday language; the team moves through clear steps and asks for your approval before continuing.

What you get

• A quick viability check with sources
• One simple visual mockup to align on a direction
• A short product spec and bill of materials (BOM)
• A first list of ingredients and manufacturer leads

How it works (Agno under the hood)

• Agent teammates: We built four specialists with Agno and put them in one Team.
  • ResearchAgent: checks the market and competition with web citations.
  • VisualAgent: proposes one mockup and shares a real image link.
  • ProductAgent: turns the idea into a short, buildable spec.
  • SourcingAgent: drafts ingredients and a shortlist of makers.
• One coordinator: A friendly Coordinator keeps everyone on track and only moves forward when you say it’s OK (simple "looks good, continue" is enough).
• Simple stages: intake → viability → visuals → spec → sourcing → final. You can say "revise visuals" or "go back to spec" at any time.
• Memory: The team remembers choices and approvals using a tiny SQLite file, so your session survives refreshes.
• Tools: Research uses Perplexity’s Search API for grounded sources. Visuals use Replicate to generate a real image URL. We add small helper tools to save your choices (e.g., selected visual, spec snapshot).

Quick start

• Requirements: Python 3.11+, Node 18+, and a browser
• Keys you’ll need (place in a .env at the repo root):
  • OPENROUTER_API_KEY (required) – lets the agents think and write clearly.
  • PERPLEXITY_API_KEY (required) – powers web search for up-to-date sources.
  • REPLICATE_API_TOKEN (optional) – enables image generation for mockups.

Start the backend (AgentOS)

• cd backend
• pip install agno fastapi[standard] uvicorn httpx replicate
• python agentos_app.py
• The AgentOS FastAPI server runs at http://localhost:7777

Start the frontend (Next.js)

• cd frontend
• npm install
• npm run dev
• Open http://localhost:3000
• If your backend isn’t on the default URL, set NEXT_PUBLIC_AGENTOS_URL in frontend/.env.local (for example http://localhost:7777).

Try it

• "I want to make a calming lavender soap for Gen Z."
• "Approve viability" or "revise viability: check competitors in EU."
• "Show one playful mockup."
• "Approve visuals; write the v1 spec."

What Agno is doing for you

• Agents: Each teammate is an Agno Agent. We group them into an Agno Team with one coordinator. The Team handles delegation and keeps everyone’s replies visible.
• Shared state: We enable shared session state so the coordinator can enforce stage gates and remember approvals. The data is stored in backend/team/data/product_studio.db.
• Tools with guardrails:
  • perplexity_search: a web search tool that the research and sourcing agents use for fresh sources (we nudge them and also double-check they used it).
  • generate_media: creates one mockup image via Replicate so you get a real URL and inline preview.
  • Small helper tools like set_stage, mark_approval, record_visual_choice, and record_spec to save decisions as you go.
• AgentOS app: We wrap the Team in Agno’s AgentOS, which exposes a lightweight FastAPI service. The frontend streams updates from http://localhost:7777 so you see responses live.

Approvals are casual

• You can type "yeah, looks good, continue" or "go ahead" to advance.
• If you’re unsure, say "revise visuals" or ask questions—the coordinator will loop the right teammate back in.

Troubleshooting

• Frontend can’t connect: make sure the backend is running on http://localhost:7777 or update NEXT_PUBLIC_AGENTOS_URL.
• No web sources: set PERPLEXITY_API_KEY in .env.
• Image not showing: set REPLICATE_API_TOKEN in .env (or skip images and continue).
• Model issues: confirm OPENROUTER_API_KEY is valid and has quota.

Want more detail?

• See the fuller plan and roles in backend/docs/TEAM_PLAN.md.
• The team code lives in backend/team/innovation_team.py if you’re curious.