What the documents look like
Fruit-store procurement records revolve around purchase orders, acceptance sheets, pesticide-residue test reports and loss records. Their sources include wholesale-market quotations, supplier offers, store sales and test reports. Compared with convenience-store fresh-food ordering, fruit purchasing depends more on wholesale-market conditions and grade-based acceptance, and it lacks a head-office standard meal as a single baseline. In cross-border material and multilingual retrieval work, the purchase order and acceptance sheet for one batch may be completed in different languages, and a pesticide-residue test report may come from an overseas institution. Records update daily or per batch.
These differences mean model connections must be checked by record type. Retrieval returns candidate text; it does not decide whether a purchase or an acceptance step is authorized. A retrieved acceptance sheet is not a completed acceptance, and a form that collects a value is not business verification, approval, a database write or external synchronization.
Constraints at the model connections step
The system needs at least one language model and one index model to work normally. AI Proxy is a required dependency, and 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 one causes startup validation to fail. AIPROXY_API_ENDPOINT takes the AI Proxy service root address, for example http://aiproxy:3000, without a trailing /v1. Legacy OpenAI environment variables and single-model custom request addresses cannot replace these two settings.
Model capability is checked against the input the application actually needs. Image input and tool calling are verified separately, and a similarly named field in a compatible chat interface is not assumed to override orchestration. Application parameters are set in the application or node configuration. Retrieval is a recall process and carries no authorization or entity-isolation duty. A relevance score is not a category, entity or permission filter, and a similarity setting does not exclude unrelated businesses or unauthorized records.
Hypothetical paper exercise sample: the following is paper review material, not real customer data.
Source text one: "Purchase order PO-2025-0417; buyer is a fruit store; supplier is overseas orchard A; item is table grapes; quantity 200 boxes; delivery date 20 April 2025; acceptance standard is grade-based acceptance."
Source text two: "Acceptance sheet RC-2025-0420; corresponds to purchase order PO-2025-0417; acceptance date 20 April 2025; received 200 boxes; grade qualified; acceptance-signature field blank."
Draft sentence to be judged: "This purchase has completed acceptance, and the loss record has been archived."
Configuration and implementation checks
| Configuration item | Hypothetical starting value for this exercise | Comparison method and reason |
|---|---|---|
AIPROXY_API_ENDPOINT | AI Proxy root address, for example http://aiproxy:3000 | Compare against the documented requirement that the root address carry no /v1; a missing value fails startup validation. |
AIPROXY_API_TOKEN | Administrator token matching the server side | Confirm the value is non-empty and matches the server side; an application API key does not substitute for it. |
| Language model and index model | At least one of each enabled | The system needs at least one language model and one index model to work normally. |
| Model capability check | Checked against the text input of the purchase order and acceptance sheet | Verify image input and tool calling separately; do not assume a similarly named field overrides orchestration. |
| Retrieval scope check | Purchase order, acceptance sheet and test report identified separately | Libraries on different vector models cannot be associated together; check the identifier first and exclude those that do not match. |
| PDF enhanced parsing | Optional custom PDF parsing service, SoMark, TextIn or Doc2x | Choose one service; when several are configured, the call priority is custom PDF parsing, SoMark, TextIn, Doc2x. |
Three common errors
- Treating a retrieval hit as acceptance authorization. Source text two shows the acceptance-signature field blank, so a retrieved acceptance sheet does not mean acceptance is complete. The blank signature field is an unsupported evidence gap.
- Entering a full request address in
AIPROXY_API_ENDPOINT. The documented requirement is the service root address without/v1; an appended path does not match the requirement, and startup validation may fail. - Replacing AI Proxy configuration with legacy OpenAI environment variables. The documents state that legacy variables cannot substitute for
AIPROXY_API_ENDPOINTandAIPROXY_API_TOKEN.
How to verify the setup
- Input source text one and source text two, then compare whether the purchase order number and the acceptance sheet number match. If the two numbers match and the acceptance-signature field is blank, the phrase "has completed acceptance" in the draft sentence is unsupported and should be recorded as pending verification.
- Input the configured value of
AIPROXY_API_ENDPOINTand compare whether/v1or/chat/completionshas been appended. If a path is appended, the value does not match the documented requirement, and startup validation may fail. - Input the draft sentence "This purchase has completed acceptance, and the loss record has been archived" and compare whether the source text states a loss record. If the material does not state a loss record, the archiving part of the sentence is unsupported, and the evidence gap remains open.
These checks test the named primary task of model connection and configuration for multilingual procurement records. They do not confirm that any test, deployment, approval or refund took place. Where a field is not provided in the source text, the value remains unknown.