What the documents look like
On-demand retail fulfillment handles online orders on a minute-by-minute basis. The working documents are instant orders, picking lists, out-of-stock reports, and delivery exception forms. These materials are generated per order, and each order record carries product names, specifications, stock status, and substitution notes that a copy-drafting workflow may draw on. A copy check therefore compares a proposed sentence against one specific order record, not against a periodic inventory count. This page addresses one combination only: on-demand retail store fulfillment, product and marketing copy drafting, and the model connection and configuration step. Whether a particular promotional sentence is supported by the source material is the question here; corporate identity verification is a different task.
Constraints at the model connections step
Model access depends on AI Proxy. From V4.17.0, both the App and an Admin service that reuses App/Service configuration must set AIPROXYAPIENDPOINT and a non-empty AIPROXYAPITOKEN. Missing either one causes startup validation to fail. AIPROXYAPIENDPOINT holds the service root address without an appended /v1. The system needs at least one language model and one index model to operate normally. A model ID is the value of the model field in the interface request body and is globally unique; the model name is for display and can be customized.
Hypothetical paper exercise sample. Source material sentence one: "Item for this order: a brand of pure milk, 250 ml × 12; store stock status is in stock; the picker has completed picking." Source material sentence two: "Substitution note: when the original item is out of stock, substitute an item of the same specification and brand; substitution requires confirmation by the store duty manager." Proposed draft sentence: "The milk for this order has been picked, stock is sufficient, order with confidence." The question is whether the proposed draft sentence is supported by the material.
Configuration centers on this sample. The language model generates copy; the index model retrieves material. Model capabilities are checked against the input the application needs, with multimodal input, tool calling, and thinking settings verified separately. Application parameters are set in the application or node configuration; same-named fields of a compatible chat interface cannot be assumed to override the orchestration. The retrieval library and the vector model identifier are checked first, and libraries built on different vector models cannot be associated together.
Configuration and implementation checks
| Configuration item | Hypothetical trial starting value | Reason and comparison object |
|---|---|---|
| AIPROXYAPIENDPOINT | Hypothetically set to http://aiproxy:3000 | The documented example is a service root address without /v1; the comparison object is whether a path is appended |
| AIPROXYAPITOKEN | Hypothetically set to a non-empty administrator token | The document requires consistency with the server side and a non-empty value; the comparison object is whether it is empty |
| Language model | Hypothetically one language model enabled | The document requires at least one language model; the comparison object is whether one is enabled |
| Index model | Hypothetically one index model enabled | The document requires at least one index model; the comparison object is whether one is enabled |
| Model ID | Hypothetically matching the model in the channel | The document states the model ID is globally unique; the comparison object is the model name in the channel configuration |
| Model probe | Hypothetically automatic probing on, interval set to 30 minutes | The document supports integers from 5 to 60 minutes; the comparison object is whether the value falls in that range |
Every value in the table is a hypothetical trial starting point for this page, not an official default. Model probing covers only enabled models; after a first failure the system retries up to 3 times, and only when all attempts fail is the model judged abnormal.
Three common errors
- Filling AIPROXYAPIENDPOINT with a full request address that appends /v1. Check: take the variable value as input, compare whether it contains only protocol, host, and port, and observe whether a path is appended.
- Using an application API key in place of AIPROXYAPITOKEN. Check: take the entered token as input, compare whether it matches the server-side administrator token, and observe whether it is empty or of the wrong kind.
- Treating "stock is sufficient" in the proposed draft sentence as already stated in the material. Check: take the source material text and the proposed draft sentence as input, compare whether the same field carries the same value in both, and observe that the material records only "in stock" and does not record "sufficient."
How to verify the setup
- Input the source material text and the proposed draft sentence from the hypothetical sample. Compare whether the draft sentence uses only the product name, specification, and status that appear in the material. Observe whether "stock is sufficient" has a corresponding field in the material. When the material does not state it, record the sentence as unsupported.
- Input the configured language model and index model identifiers. Compare whether both are enabled. Observe whether the requirement of at least one of each is met. When either model is missing, record the configuration as incomplete.
- Input the interval value of the model probe configuration. Compare whether it is an integer between 5 and 60. Observe whether it falls inside the range the document describes. When it falls outside, record it as needing adjustment.
All judgments above are paper checks; no test, deployment, or approval was actually executed. When a retrieval result is empty, a downstream design can be checked against the explicit paper input []; this does not claim that a fictional rule retrieval necessarily returns an empty array. A relevance score is not a category, entity, or permission filter, and retrieval does not grant authorization. Forms collect values; business verification, approval, database writes, and external synchronization require their stated downstream implementation.