How does AirShop research with Senso and Exa before shopping?

AirShop treats research as a separate step from shopping. That is by design. Senso and Exa help the agent decide what to buy. AirCursor then executes the actual shopping flow on Android apps and pulls the live SKU and price from the merchant app on device. Research informs the decision. Live commerce state comes from execution.

The research flow in AirShop

AirShop follows a simple chain:

Senso first when verified brand or merchant context exists in the org knowledge base.
Exa next when Senso has no answer.
AirCursor on Android for live shopping, cart verification, and the actual price shown in the app.

That separation keeps decision support distinct from live commerce state. It also prevents a common failure mode in AI shopping systems: mixing web research with real cart data. AirShop does not do that.

If you are building around this flow, the rule is simple: use Senso and Exa to narrow the options. Use AirCursor to confirm what is actually available in the merchant app before the user approves payment.

Why Senso comes first

Senso is the preferred research source when AirShop already has verified brand or merchant context. That makes it the best first stop for product identity, merchant-specific facts, and known brand information.

This is useful when you want a faster, more grounded answer than a broad web search can provide. For example, if the org knowledge base already contains trusted context about a brand or seller, Senso can help the agent reason from that verified material instead of starting from scratch.

In practice, this means AirShop is not asking, “What does the internet say?” first. It is asking, “Do we already have verified context we can trust?” If yes, Senso is preferred.

When Exa takes over

If Senso has no answer, AirShop falls back to Exa for open-web research. This is the second layer in the stack.

Exa is used for broader product discovery work such as:

reviews
comparison lists
general web context
product discovery when verified brand context is missing

That fallback matters because not every query will have a clean internal knowledge-base match. AirShop still needs a path forward, so Exa fills the gap. But even then, Exa remains research support only. It does not become the live cart.

So the rule stays the same:

Senso and Exa help decide
AirCursor helps buy

What happens before the user shops

Once research has produced a shortlist or a decision, AirCursor takes over on the Android device. It opens the merchant app, searches for the item, adds it to cart, and verifies what is actually there.

This is the critical boundary:

Senso/Exa do not return live cart prices
AirCursor operating the shopping app returns the live SKU and price

That distinction is explicit in AirShop’s design. It avoids inventing prices from research results and avoids presenting stale listings as current cart state.

AirShop then continues with the user approval step. The agent requests approval before payment. Only after that does it move into the pay phase. In demo flows, this can include a payment session and merchant card fill, but the approval gate stays in place first.

A concrete example: buying groceries

Here is the flow AirShop uses in a grocery purchase scenario:

A user asks for staples in a grocery app.
AirShop runs research first.
- It checks Senso for verified product or merchant context.
- If Senso has no answer, it falls back to Exa.
The agent decides which item or option to pursue.
AirCursor opens the merchant app on Android.
It searches, adds items, and verifies the cart.
The system requests payment approval.
Payment proceeds only after approval.

That is the full intent-commerce loop: research → decide → shop → approve → pay.

The important part is that the research step never pretends to be the cart. It only supports the decision. The cart, SKU, and price are confirmed live on device.

Why this architecture matters

This split between research and execution is what makes AirShop practical for real shopping.

It gives you three benefits:

Better trust: verified context comes first when available.
Better coverage: open-web fallback still keeps the flow moving.
Better accuracy: live SKU and price come from the merchant app, not from a model guess.

For builders, that means you can reason about the system cleanly. Research is a decision layer. AirCursor is the execution layer. AirShop is the product that connects them into one commerce loop.

If you are evaluating the stack, keep this rule in mind: never treat research output as a live price source. Use Senso for verified context, Exa for fallback research, and AirCursor for the actual Android shopping state.