What the documents look like
Creator sales campaigns run on external creator fulfillment and commission settlement. The working papers are creator cooperation agreements, commission setup forms, sample shipment forms, and livestream performance reports. Cross-border materials and multilingual retrieval must handle differences in language, version, and counterparty across those papers. One campaign can carry a Chinese agreement alongside a foreign-language commission confirmation, and creator historical data and store transaction details come from separate sources. Retrieval results support human review and do not constitute authorization. Materials update per campaign and per week; this page does not derive batch synchronization or transaction triggers from that cadence.
Constraints at the model connections step
Model connections depend on AI Proxy. Both the App and an Admin service that reuses App/Service configuration must set AIPROXYAPIENDPOINT and a non-empty AIPROXYAPITOKEN; a missing item causes startup validation to fail. AIPROXYAPIENDPOINT takes the service root address without /v1 appended. The system needs at least one language model and one index model. The model ID is globally unique, and the model name can be customized. Semantic recall and relevance scores do not carry permission or entity isolation. Library vector model identifiers must be checked first, and libraries on different vector models cannot be associated together.
Hypothetical drill sample: A brand business team retains a paper excerpt titled "Creator Campaign Cooperation Confirmation," reading: "Party A: Xingye Trading Co., Ltd.; Party B: Creator Lin; campaign session: September 20, 2026 special; commission rate: per the settlement sheet; sample shipment: sent, receipt status not returned." The same batch retains a commission setup sheet excerpt: "Session September 20, 2026, commission rate 15%, status is pending review." The draft sentence to assess is: "The commission rate for this campaign is confirmed and can be used for settlement." Materials needed for the check must sit inside an accessible target retrieval scope or explicit manual input; when missing, record the gap.
Configuration and implementation checks
| Configuration item | Assumed starting point for this page | Check method |
|---|---|---|
| AIPROXYAPIENDPOINT | Assumed filled as http://aiproxy:3000, chosen because the official example uses a root address form | Compare with the deployed service root address and confirm /v1 is not appended |
| AIPROXYAPITOKEN | Assumed filled with a non-empty administrator token consistent with the server | Compare with the AI Proxy server configuration and confirm non-empty and consistent |
| Language model | Assumed one language model enabled after passing a channel test | Compare with the enabled state after the model channel test |
| Index model | Assumed one index model enabled after passing a channel test | Compare with the requirement of at least one index model |
| Model ID | Assumed to match the model field in the channel | Compare character by character with the channel's model field |
| Retrieval library vector model identifier | Assumed recorded as the same identifier | Compare with the target library identifier; exclude mismatches first |
These values are assumed trial starting points for this page, not official defaults.
Three common errors
- Treating "pending review" on the commission setup sheet as confirmed. Input is the two excerpts; the comparison field is the commission rate status; the observable result is that the setup sheet status remains pending review.
- Substituting an application API key for the administrator token in AIPROXYAPITOKEN. Input is the environment variable value; the comparison object is the server administrator token; the observable result is that the two are inconsistent.
- Treating a semantic recall score as a permission filter. Input is one retrieval request; the comparison object is the record's owning entity; the observable result is that score level does not change the visibility of unauthorized records.
How to verify the setup
- Input the two excerpts above and the draft sentence; compare the draft sentence "the commission rate is confirmed" with the setup sheet status "pending review"; the observable result is that the sentence is unsupported and recorded as pending verification.
- Input the filled AIPROXYAPIENDPOINT value; compare whether /v1 or /chat/completions is appended; the observable result is that the condition is met only when neither is appended.
- Input the vector model identifiers of two retrieval libraries; compare whether they match; the observable result is that a mismatch excludes joint association, recording only the actual observation without inferring missed recall or redundant results.