Brand-owned store operations: model connections for copy drafting

What the documents look like

Brand-owned stores are staffed and accounted for directly by the brand, without franchise contracts governing the relationship. Day-to-day records include daily sales reports, shift schedules, display photographs and stock transfer orders, refreshed daily or weekly. When product and marketing copy is drafted, the prices, stock states and campaign scopes quoted in that copy need to correspond item by item with these records. Store inspection notes and member CRM data may supply display and customer-group information, but this page addresses only the model connection and configuration step, not copy distribution or approval.

A practical consequence is that a drafted sentence can be checked against a specific record. The record supplies the object, the effective range and the stated value; the drafted sentence either matches those fields or it does not. A model connection that produces fluent text does not make the underlying claim supported.

Constraints at the model connections step

Copy drafting depends on a language model to generate text. FastGPT reaches model providers through AI Proxy. From V4.17.0, AI Proxy is a required dependency, and both the App and Admin services must set AIPROXYAPIENDPOINT and a non-empty AIPROXYAPITOKEN. A missing value in either causes startup validation to fail. Legacy OpenAI environment variables or a single-model custom request address do not replace these two settings.

Model capability is checked against the input the application actually needs. Multimodal input, tool calling and reasoning configuration are verified separately. Parameters are set in the application or node configuration, and it cannot be assumed that a same-named field on a compatible chat interface overrides the orchestration. The same-vector-model requirement for retrieval libraries is limited to the vector retrieval association and does not extend to business registration sites or other external systems.

Hypothetical paper exercise: a brand-owned store's daily sales report for 1 June 2025 to 30 June 2025 records "summer short-sleeve T-shirt, list price 199 yuan, in-store pickup price 149 yuan from 15 June to 30 June, stock 320 pieces." A display photograph shows the T-shirt on the second display stand at the store entrance. The proposed sentence reads: "This store's summer short-sleeve T-shirt was originally 199 yuan, now 149 yuan, at the same price across all channels." The question is whether that proposed sentence is supported by the records.

Configuration and implementation checks

The following entries relate to this page's copy drafting task and rest on settings or implementation checks the official documents describe. The values are starting points for this hypothetical exercise, not official defaults.

Three common errors

1. Entering AIPROXYAPIENDPOINT as a complete model request address with /v1 or /chat/completions appended. The document requires a service root address. The check compares the entered value against the documented example format and confirms no path suffix is present.
2. Using an application API key in place of AIPROXYAPITOKEN. The document requires this token to be the AI Proxy administrator token, consistent with the server side and non-empty. The check compares the entered token against the AI Proxy server configuration and confirms it matches and is non-empty.
3. Extending the same-vector-model requirement for retrieval libraries to external systems. The document describes only the vector retrieval association condition. The check compares retrieval library identifiers, excludes those that do not match, and records the actual observation without running a cross-model retrieval to compare missed recalls.

How to verify the setup

1. Input the daily sales report text from the hypothetical exercise and the proposed sentence. Compare the price, date range and channel wording in the proposed sentence against the recorded text. The record states "in-store pickup price 149 yuan," while the proposed sentence says "at the same price across all channels." The record does not state same price across all channels, so that wording is unsupported. The record states "15 June to 30 June," while the proposed sentence carries no campaign period, which is a missing piece of information. Expected result: the price figure matches, the campaign period is absent, and the channel scope is unsupported.
2. Input the display photograph description from the hypothetical exercise and the proposed sentence. Compare whether the proposed sentence includes the display position shown in the photograph. The photograph shows "the second display stand at the entrance," and the proposed sentence does not mention display position. Expected result: the proposed sentence does not cite the display position, and that information is not carried into the copy.
3. Input the stock transfer order from the hypothetical exercise, which is not provided, and the proposed sentence. Compare whether the proposed sentence addresses stock state. The transfer order is not provided, so the stock state field is unknown. Expected result: comparison is not possible, and the evidence gap is recorded.

Empty-result handling can use the paper input [] to check downstream design, without claiming that a search of invented rules necessarily returns []. Retrieval checks state observations and failure branches, without guaranteeing a first hit, full recall or an empty result for an unrelated query. A pending review does not demonstrate that an operating permission is absent. A quotation validity period does not indicate a supply period. A statement that is not recorded, or that has only evidence from a related stage, is described as unsupported or pending verification.

Related reading

• self-host/config/env
• self-host/config/model/intro
• FastGPT model connections