How Seeded AI Apps Work
The architecture behind AI tools without infrastructure
The Architecture
A Seeded AI App has no backend. There's no server, no database, no API you need to maintain. Instead, you define a seed that transforms any AI into a specialized assistant.
1. Context
Background information the AI needs to understand your domain:
- Who you are
- What you offer
- Who your customers are
- Any terminology or concepts unique to your business
This gets loaded into the AI's context window when the user starts a conversation.
Example:
Acme Corp provides industrial automation tools for manufacturing plants. Our customers are typically plant managers and operations engineers looking to reduce downtime and improve throughput.
2. Instructions
Behavioral guidance for how the AI should operate:
- What types of questions to answer
- How to route different request types
- When to escalate to human support
- What tone and style to use
Example:
## Instructions - For technical questions, fetch the relevant help page and guide the user - For pricing questions, provide current plan information from the pricing page - For account-specific issues (order status, billing), direct users to login at acme.com/account - Keep responses concise and practical - Don't speculate about features not documented on the site
3. Data Sources
URLs to public pages the AI can fetch for current information:
- Help documentation
- Product pages
- Pricing information
- Policy documents
- FAQs
Example:
## Resources Technical support: https://acme.com/help Pricing and plans: https://acme.com/pricing Product documentation: https://acme.com/docs Shipping policies: https://acme.com/shipping Contact information: https://acme.com/contact
How It Runs
- 1.User gets the seed — Via shared link, QR code, or pasted context
- 2.Seed loads into their AI — Claude, ChatGPT, Gemini, etc.
- 3.User asks a question — "How do I configure the API?"
- 4.AI matches intent to instructions — This is a technical question
- 5.AI fetches relevant URL — Gets content from acme.com/docs/api
- 6.AI responds — Using fetched content plus seed context
The entire interaction runs on the user's AI. You don't see the conversation, you don't pay for compute, and you don't maintain any infrastructure.
Data Constraints
Seeded AI Apps work with public data only. No authentication is involved, which means:
Works:
- Product information
- Documentation
- Pricing
- Policies
- Public FAQs
- Help articles
Doesn't work:
- Order status (requires login)
- Account details (requires auth)
- Personalized recommendations (requires user data)
For auth-required information, your seed instructs the AI to direct users to the appropriate portal.
Comparison to Traditional Approaches
| Aspect | Chatbot | Seeded AI App |
|---|---|---|
| Infrastructure | You host | User's AI |
| Cost per conversation | You pay | Free |
| Training data | Upload and manage | Point to URLs |
| Updates | Deploy new version | Edit seed |
| User trust | Company bot | Their own AI |
| Scaling | Costs increase | Costs stay zero |
Getting Started
- 1. Identify your use case — Support, onboarding, product advice?
- 2. Audit your public content — What URLs answer common questions?
- 3. Write your instructions — How should the AI behave?
- 4. Create your InstantContext profile — Package it as a shareable seed
- 5. Distribute — Share the link wherever your users are