What the documents look like
Retail checkout and loyalty records center on price and membership entitlements. The main sources are the POS system, the member CRM, the payment platform and the price-tag system. Everyday documents include promotional price tags, membership points rules, the cashier daily report and exception refund slips. A promotional price tag carries a validity period, applicable stores and member tiers, and those fields determine whether a product-data question can produce a checkable answer. The accumulation and redemption conditions inside a membership points rule are the basis for entitlement judgments. The cashier daily report and exception refund slips record shift-level amount differences. These materials update daily or per shift, and that update frequency does not change how this page configures anything; model connection and configuration only determine whether question answering can read and cite these materials.
Constraints at the model connections step
Product-data and selection question answering requires the model to answer from promotional price tags, points rules, cashier daily reports and refund slips, so model connection must complete the AI Proxy dependency configuration. Starting from V4.17.0, AI Proxy is a required dependency. Both the App and any Admin service that reuses App/Service configuration must set AIPROXY_API_ENDPOINT and a non-empty AIPROXY_API_TOKEN; missing either item causes startup validation to fail. AIPROXY_API_ENDPOINT holds the AI Proxy service root address and does not append /v1. Model capabilities are checked against the inputs the application needs, and multimodal, tool-calling and thinking settings are verified separately. Application parameters are set in the application or node configuration, and it cannot be assumed that same-named fields on a compatible chat interface override the orchestration. Retrieval and relevance scoring do not carry permission or entity isolation, and the same-vector-model requirement for a retrieval library does not extend to external systems.
Hypothetical paper exercise sample: the text of a promotional price tag for a store dated March 1, 2026 reads "Product A promotional price 9.9 yuan, valid March 1 to March 3, 2026, this store only, same price for members and non-members." The text of the membership points rule for the same day reads "Members earn 1 point per 10 yuan spent, and points redemption is limited to the member's own account." The draft sentence to be judged is "Product A still sells at 9.9 yuan on March 5, and non-members may also redeem using another person's points." This request is pending verification and has not been authorized for execution.
Configuration and implementation checks
The table below lists only officially documented settings or implementation check items related to this page's documents. The values are starting points for this hypothetical exercise, used for comparison against the sample source text, and are not official default values.
| Configuration item | Hypothetical exercise starting point | Check method |
|---|---|---|
AIPROXY_API_ENDPOINT | Hypothetically filled as http://aiproxy:3000 | Check the root address has no appended /v1 and matches the AI Proxy service address |
AIPROXY_API_TOKEN | Hypothetically filled with a non-empty admin token | Compare with the server-side configuration; confirm non-empty and consistent |
| Model capability | Hypothetically a language model supporting text question answering | Check against the inputs the application needs; verify multimodal and tool calling separately |
| Application parameters | Hypothetically prompts and quote limits set in the application node | Compare with node configuration; confirm no reliance on same-named chat-interface fields |
| Retrieval library identifier | Hypothetically the knowledge base uses one index model | Check the library identifier against the vector model identifier; exclude mismatches |
| Alert Webhook | Hypothetically a POST callback address is configured | Check request headers and the 5-second timeout; no Token is carried when none is configured |
Three common errors
- Filling
AIPROXY_API_ENDPOINTwith a complete request address and appending/v1. The input is the environment variable value, the comparison object is the AI Proxy service root address, and the observable result is whether startup validation passes. - Using an application API Key in place of
AIPROXY_API_TOKEN. The input is the token value, the comparison object is the server-side admin token, and the observable result is whether the call is rejected. - Treating a retrieval relevance score as permission or entity filtering. The input is the query and library scope, the comparison object is the store and membership boundaries in the materials, and the observable result is whether returned content exceeds the sample's authorized scope.
How to verify the setup
- For the sample draft sentence "Product A still sells at 9.9 yuan on March 5," input the price tag source text, compare the validity-period field March 1 to March 3, 2026, and observe that the draft sentence exceeds the validity period and is unsupported.
- For the sample draft sentence "non-members may also redeem using another person's points," input the points rule source text, compare the field "limited to the member's own account," and observe that the draft sentence contradicts the source text.
- For empty-result handling, input the paper empty array
[]and check the downstream design; observe whether downstream treats it as an empty result. This does not claim that a search for a fictional rule necessarily returns an empty array. Cross-model records cannot be jointly associated; first check the library identifier against the vector model identifier, exclude non-matching ones and record the actual observation.