--- title: "How Nook works on real stores — Nook" description: "How Nook works on real stores, from the Checkout Bench lab and its results on real checkouts to the way your agent shops around and collects your approval, and the small API developers use to go from quote to order." canonical: https://nookstore.xyz/product last_updated: 2026-11-01 --- # How Nook works on real stores This page brings together the parts of Nook that sit behind every quote, from the agent and model you already use and the store accounts they reach, through the Checkout Bench lab and its results on real store checkouts, to the way your agent shops around, collects your approval and connects through one small API. ## Start with your agent Any agent can register its own partner key, and yours connects your accounts in one link. Registration is self-serve and works at once. ## Any model gets the same checkout Muse, ChatGPT, Grok and open-weights models can all finish the checkout as long as they can call a tool, because the connection belongs to the person and the model is only a detail. - **Same job, three models**: Claude, ChatGPT and Grok each run Connect → Quote → Approve → Bought on one Celsius 12-pack through the same connection. - **It also works with** DeepSeek, Kimi, GLM, ElevenLabs and Exa. ## One-tap integration for everyday purchases and delivery You ask where your agent already is (ChatGPT, Claude, Grok, iMessage, WhatsApp), and Nook turns it into a real order at the store (Amazon, Uber Eats, Instacart, Walmart, Target, Chewy), with the Nook app icon at the center of the diagram. - **Quotes and purchases**: Amazon, Uber Eats, Instacart, Walmart, Target, Chewy. - **Connect only, for now**: DoorDash, Urban Outfitters, Resy, Detoure, Delta, Apple. ## Inside the Checkout Bench lab The lab is where every Checkout Bench run takes place, a signed-in workspace that sends real shopping requests to AI assistants through Nook, records everything each one did and scores the result against known answers. - **Quote suite.** Quote suite sends one shopping request to several AI assistants at once, and each one uses Nook to get a real checkout quote from a real, connected store account. You pick the store, the assistants and what the shopper asks, and every run stops before payment. - **Run view.** Every assistant in a run gets its own verdict card with the quote total or the blocker, whether the request matched, and its time, tokens and cost. Its details open onto the requests it sent, the quote or blocker, the conversation, the checks, the findings in order and the full trace. - **Overview.** Overview shows what the lab is doing and what it can reach, with the lab’s status, the assistants on hand, the stores connected through Nook and a fresh sandbox VM for each model, above the most recent runs. - **Runs.** Runs lists every run the lab kept, newest first and grouped by day, with counts of completed and failed runs and the share of clean verdicts, and each row opens, replays or downloads its evidence. - **Trace replay.** Trace replay opens a single run as “Inside this shopping run,” with what the assistant actually said, the requests it recorded and their outcomes, and a step-by-step timeline you can play, pause and filter by source. - **Model comparison.** Model comparison sends one request to separate assistants, each in its own VM with real tools and responses, and lays their results side by side so the models can be read against each other. - **Bench.** Bench runs fixed shopping tasks with known right answers, starting from an answer sheet, scoring every attempt as a correct or wrong checkout, and collecting the results in a scorecard with the causes behind each failure. - **Accounts.** Accounts lists the stores connected to the lab through Nook and when each one was last verified, next to the stores still available to connect in Nook. - **Retail Profiles.** Every request can run as one consistent shopper, with a short written profile that a simulated shopper answers the assistant’s questions from or that the assistant keeps as its own notes. The profile travels with the real store accounts connected through Nook, so the sizes, addresses and choices an assistant asks about come from the same person each time. - [Open the lab](https://checkoutbench.ai/lab) ## Every purchase adds up to a clear picture Nook keeps the figures from every quote and order, so a month of agent shopping reads like a statement. The page charts weekly spend, spend by store, spend by category, how each purchase was approved, the path from request to doorstep and which banks linked through Plaid paid for it, and those numbers are illustrative sample data rather than anyone's real activity. ## Measured on real store checkouts Checkout Bench runs real models through Nook against signed-in stores and a replica of Amazon built from 211 real runs, and its actual results are charted on the page rather than sample data: 9 of 10 tasks land within 15 points of live Amazon, the grader matches three outside readers on 44 of 44 calls, 6 of 7 task kinds score 88 to 100 percent for all five models, and the replica is built from 211 recorded live Amazon runs. - **The Amazon replica predicts what live Amazon does**: for DeepSeek V4.1 Flash, nine of ten tasks land within 15 points of live Amazon, and the one that misses is the DJI Mic Mini kit, where live Amazon showed three competing listings and the replica's September data had gone stale. - **See how your checkout does against different models**: on lookalike-product tasks, shown one model at a time, Grok 4.7 quoted the right product 88% of the time and never quoted a wrong one 97% of the time, ahead of Gemini 3.8 Flash (78/81), GPT-5.6 Sol (66/100), DeepSeek V4.1 Flash (63/66) and GPT-5.6 Luna (47/98). - **The rebuilt replica ranks models the way live Amazon does**: on held-out tasks GPT-5.6 Luna goes from 69% to 88% to 84% and DeepSeek V4.1 Flash from 80% to 73% to 74% across the September replica, the rebuilt replica and live Amazon, and the live column is interim because those batches are still running. - **Run real checkouts with payment or without**: a run can go all the way through an approved payment or end at a verified quote, and the 31 graded runs on real stores on Oct 25 used quote-only mode (DeepSeek V4.1 Flash: 7 quotes ready, 7 blocked, 2 stopped by the operator; GPT-5.6 Luna: 4 ready, 9 blocked, 2 asked the person). - **See the top reasons different AI agents stopped before recommending a transaction**: of 94 blocked quotes from Sep 26 to Oct 27, 46 needed the person to pick, 22 had a checkout title mismatch, 8 were over budget, 7 never reached the cart, 4 had charges that didn't add up, and the rest split across unexpected items, no saved address, no Buy Now button and the card list. ## Train a model on your own site Checkout Bench turns your site's own sign-ins, carts and checkouts into graded tasks and fine-tunes a model on the attempts that pass, and reinforcement learning can then reward the runs that ask when a request is unclear, so the model learns how your checkout actually behaves instead of guessing from every other store. The loop is run your tasks, grade every attempt, train on what passed and measure again on live, and the pass-rate-by-round chart beside it is labeled illustrative because no trained result has been measured yet. ## Send your agent through as many connected accounts as you want. The same cart is rarely the same price everywhere, so Nook quotes it across every store you connected at once, watches the prices you care about and speaks up when one moves, and the agent always asks before it buys. ## Keeps up with your life Your agent handles the standing errands, like the restock before the bag runs out and the Friday list before the weekend, and it learns the rhythm so each one comes to you as a single approval. ## Launch a connected store for any agent. Each store you connect becomes available to whichever agent you authorize, and the connection keeps its owner, status, start date and last order on record, so nothing about your accounts is a black box. ### Sign in with iMessage Your phone number is your sign-in, because Nook texts you a code and, once you enter it, your agents can reach every account you connected without passwords to hand out or store-by-store logins. Face ID and fingerprint work too, and the code never leaves your phone. ## Approve anywhere Approval lives in the conversation you already have with your agent rather than in a separate app, so any clear affirmative reply to the quote's total, such as "approve", "confirm", "buy it" or "I'm down", counts in your own words. Nook passes your words along exactly as you said them and keeps them with the order, and if you'd rather look first, Buy on Nook's review page works too because the review doesn't expire. The page draws this as an iPhone at real proportions (393 by 852 points, with its bezel and Dynamic Island) holding an iMessage thread: an itemized Amazon quote for three items totalling $50.00 with an Approve button, your Approve reply, and Nook confirming the order, beside one list of the same approval in WhatsApp, Telegram and on Nook's review page. ## With your people When a buy takes more than one approval, Nook collects each person's approval for their share (Split it) in the chat apps they already use, including WhatsApp, Telegram, Discord and nine more. ## Let agents finish real purchases in under 60 seconds Your agent sends one connect link, gets a real quote back from the store’s own checkout, collects the shopper’s approval in chat and holds a confirmed order number before the first minute is up, all through four calls to the same small API. | Step | Call | Elapsed | | --- | --- | --- | | Connect: the shopper opens your one link and connects their stores | `POST /v3/api/agent/connect-links` → `{ url }` | 0:08 (+8 s) | | Quote: Nook returns the store’s real total, line by line | `POST /v3/api/agent/quotes` → 202, then `GET /v3/api/agent/quotes/q_…` → `ready` | 0:27 (+19 s) | | Approve: the shopper confirms the total in their own words | `POST /v3/api/agent/purchases` with `{ quote_id, confirmation }` | 0:39 (+12 s) | | Order confirmed: the store’s order number comes back on the receipt | `GET /v3/api/agent/purchases/p_…` → `confirmed`, `receipt.order_id` | 0:53 (+14 s) | These are illustrative timings, with the quote step set to the 19 s median shown on this page rather than measured end to end. ## Agents never write code against your accounts. Nook runs every store action as one of its own fixed, guarded steps and never runs browser code that a model wrote, which is a deliberate design choice that keeps both your orders and your data out of a model’s reach. - **Guarded actions only**: every store action runs as one of Nook’s fixed steps, and the step that presses a store’s final order button only does so after the shopper has approved that exact quote. - **No model-written code**: an agent can ask Nook for a quote or a purchase, but it can never hand Nook code or requests of its own to run, so nothing a model writes can place an order or read your session cookies on the side. - **Tokens stay with Nook**: raw store responses are full of personal data and session tokens, so they stay inside Nook and your agent receives clean quote data with the items, the charge lines and the total. ## One small API from quote to order - **Authorized connections**: Easily create authorized connections, so people rarely revisit passwords and never deal with browsers-in-browsers. - **Keys**: your agent generates its own partner key (`nk_` + 43 random characters); Nook stores only its hash. `POST /v3/api/agent/register` → 201; resending the same key and id returns the same record. - **Idempotency**: every quote and purchase takes an `Idempotency-Key`. Same key and body → the first result, never a duplicate. Same key, different body → `409 idempotency_key_reused`. - **Quotes**: `POST /v3/api/agent/quotes` → 202 `preparing`; poll `GET /v3/api/agent/quotes/` until `ready` (about a minute on Amazon, Instacart, Walmart and Chewy; two to three on Uber Eats). The quote's `lines` are the store's own charge rows and sum exactly to `total_cents`; `destination` and `payment` report whether they matched; `stages` reports each stage's milliseconds; after quoting, Nook restores the store's cart (`cleanup`), and anything unrestorable arrives as a `warnings` entry. - **Purchases**: `POST /v3/api/agent/purchases` with `{quote_id, confirmation: ""}` buys at once. Nook re-reads the checkout just before its one click and buys at the quote's total or less; a higher total (`price_increased`) or a changed item, address or card stops it, buying nothing. States: `awaiting_approval`, `approved`, `submitting`, `confirmed` (with `receipt.order_id`), `blocked`, `uncertain` (never buy again; check the store's order history). - **Errors**: `{ "error": "", "details": { … } }`; request bodies are JSON, at most 4096 characters. - **Surfaces**: the same protocol is exposed as a REST API, an MCP server and ChatGPT plugins — any model that can call a tool can finish a checkout. Business accounts hold partner keys per agent, product or team. - **Quote success by store**: the page shows a clearly labeled hypothetical example of the 30-day quote success view for JD Sports, Best Buy, OpenTable, Sweetgreen, Fanatics, Shein and Costco; those figures are for illustration, not measured. - **Training environment**: fine-tune a model on your own store's flows — real sign-ins, carts and checkouts as training signal. ## Delivered from the stores you already order from Nook quotes delivery the way each store's own checkout does, with the real fee, the delivery window and your saved address, so groceries, takeout, pet food and household restocks (Amazon, Uber Eats, Instacart, Walmart, Target, Chewy, Best Buy, Costco, Home Depot, CVS, Safeway, Albertsons, Wayfair, 7-Eleven, Wawa and more) arrive when the quote said they would.