What the documents look like
Member private-domain operations keeps its selection questions in four recurring document families: member tag tables, outreach plans, coupon rules and privacy authorization records. The material behind a selection answer arrives from the CRM system, WeCom, POS purchase records and mini-program behavior data, refreshed daily or weekly. For product questions, three field groups carry the decision: the member tag, the purchase history in POS records, and the outreach status held in the privacy authorization record.
A selection answer is only as complete as those three groups. The tag table describes the member segment, the POS record describes what was bought and when, and the authorization record describes whether outreach is permitted and through which date. When any one of the three is absent, the selection conclusion can only be recorded as pending verification. This page does not describe batch synchronization mechanisms.
Constraints at the model connections step
Model connection and configuration determine whether a question-answering application can read the correct material. AI Proxy is a required dependency from V4.17.0. Both the App and any Admin service that reuses the App/Service configuration must set AIPROXY_API_ENDPOINT and a non-empty AIPROXY_API_TOKEN; missing either one causes startup validation to fail. AIPROXY_API_ENDPOINT takes the service root address without /v1 appended. Legacy OpenAI environment variables and single-model custom request addresses do not substitute for these two settings.
Model capability is checked against the inputs the application actually needs. Multimodal input, tool choice and reasoning configuration are verified separately. Parameters belong in the application or node configuration; a same-named field on a chat-compatible endpoint is not assumed to override orchestration. Retrieval returns material and does not perform authorization. A similarity score is not an entity or permission filter, and a retrieval hit is not outreach consent.
Hypothetical paper exercise sample: a brand CRM team reviews one member tag table on paper. The source text reads: "Member ID M-2041, tag high-repurchase infant and maternal segment, most recent POS purchase date 2026-03-14, privacy authorization record shows marketing outreach status authorized, authorization valid through 2026-12-31." The draft sentence under review is: "M-2041 can currently receive a WeCom push for the new diaper product." The comparison covers three fields — tag, purchase date and authorization status — and the decision boundary is whether the authorization validity period covers the proposed push date.
Configuration and implementation checks
| Configuration item | Hypothetical starting value for this exercise | Reason and comparison object |
|---|---|---|
AIPROXY_API_ENDPOINT | Assume http://aiproxy:3000 | Documentation requires the root address without /v1; compare against a value that appends /v1 |
AIPROXY_API_TOKEN | Assume a non-empty administrator token | Documentation requires a non-empty token consistent with the server; compare against an application API key |
| Model capability check | Assume one language model and one index model are each verified | Documentation requires at least one language model and one index model for normal operation |
| Retrieval scope | Assume the request is limited to this brand's member tag library | Compare against a cross-library request and confirm the library identifier matches |
| Authorization status field | Assume the outreach status is read from the privacy authorization record | Compare against the tag table field and confirm the same subject and the same field |
Three common errors
- Entering
AIPROXY_API_ENDPOINTas a full request address with/v1appended. The documentation specifies the root address; an appended path does not match the documented form and startup validation or calls will not behave as documented. - Using an application API key in place of
AIPROXY_API_TOKEN. The documentation requires an administrator token consistent with the server side; the two credentials have different origins and are not interchangeable. - Treating a retrieval hit as outreach authorization. Retrieval returns material only; the authorization status must be read from the privacy authorization record, and the two cannot substitute for each other.
How to verify the setup
- Input the tag table source text and the draft sentence from the hypothetical sample, then compare the proposed push date against the authorization validity end date of 2026-12-31. When the validity period covers the push date, the draft sentence is supported; when it does not cover the date, or the authorization field is missing, the conclusion is pending verification.
- Input the assumed
AIPROXY_API_ENDPOINTvalue and the documented requirement, then compare whether/v1is appended. When no path is appended andAIPROXY_API_TOKENis non-empty, the connection configuration matches the documented conditions; when a path is appended or the token is empty, record it as non-conforming. - Input a retrieval request for member ID M-2041 from the hypothetical sample and observe whether the returned results come from this brand's member tag library. When the library identifier matches, record the material as retrieved; when the return is empty or the library identifier differs, record the gap and do not infer authorization status from it.