What the documents look like
Convenience-store fresh-food ordering runs on short shelf lives and frequent store-level replenishment. The working materials are fresh-food order forms, damage write-off forms, near-expiry handling sheets, and temperature records. Sales data arrives from POS, order data from the head-office ordering system, and stock data from store inventory. These materials update daily or by shift, and ordering decisions depend on the sellable quantity for the day and the remaining time before expiry.
Product information and selection questions sit inside this material set. A question about specification, shelf life, or ordering conditions should be answered from the uploaded documents rather than from general knowledge. That constraint shapes how models are connected and configured, because the answer must stay traceable to the order form, the write-off form, the near-expiry sheet, or the temperature record.
Constraints at the model connections step
FastGPT reaches model providers through AI Proxy. The App and Admin services must both set AIPROXY_API_ENDPOINT and a non-empty AIPROXY_API_TOKEN; a missing value in either one causes startup validation to fail. AIPROXY_API_ENDPOINT holds the AI Proxy service root address and does not carry a /v1 suffix. The system needs at least one language model and one indexing model before it works normally.
Model ID is globally unique, and model mapping translates the model requested by FastGPT into the model actually sent upstream. Model capability is checked against the input the application needs; multimodal input and tool calling are verified separately. Application parameters are set in the application or node configuration, and same-named fields on a compatible chat interface are not assumed to override orchestration settings.
Hypothetical paper exercise: a store's fresh-food order form dated 23 September 2026 records "rice ball category: orderable before 14:00 same day, arrives at store 06:00 next day, shelf life 48 hours." The temperature record shows the refrigerated cabinet at 4 degrees Celsius at 08:00 that day. A proposed sentence reads: "After 14:00 that day, the store can still add rice ball orders." This sentence conflicts with the cut-off time in the order form and remains a pending item.
Configuration and implementation checks
| Configuration item | Hypothetical starting point for this exercise | Reason and comparison method |
|---|---|---|
AIPROXY_API_ENDPOINT | Enter the AI Proxy service root address | Matches the official requirement of no /v1 suffix; check whether a path suffix is present |
AIPROXY_API_TOKEN | Enter a non-empty token consistent with the server side | Compare the App value with the server-side value; an empty value fails validation |
| Language model | Enable one language model that has been tested | Compare against the model test result and confirm the dialogue request returns success |
| Indexing model | Enable one indexing model that has been tested | Compare against the model test result and confirm the indexing request returns success |
| Model mapping | Enter the mapping per channel | Compare the FastGPT request model ID with the upstream actual model ID |
| Probe interval | Set to 30 minutes for this exercise | Official range is an integer from 5 to 60 minutes; compare with the alert-response need |
Three common errors
- Writing
AIPROXY_API_ENDPOINTas a full model request address with a/v1suffix. Input: the environment variable value. Comparison: the official requirement of a root address without/v1. Observable result: a value containing/v1or/chat/completionsdoes not match. - Using an application API key in place of
AIPROXY_API_TOKEN. Input: the source of the token field. Comparison: the AI Proxy administrator token. Observable result: a source that is an application key does not match. - Treating the proposed sentence "After 14:00 that day, the store can still add rice ball orders" as an answer. Input: the order form text and the proposed sentence. Comparison: the 14:00 cut-off in the source text. Observable result: the proposed sentence and the source cut-off do not agree, so the sentence is marked pending.
How to verify the setup
- Input the hypothetical order form text and the proposed sentence. Compare the source phrase "orderable before 14:00" with the proposed phrase "after 14:00 can still add." If the two time points are mutually exclusive on the same subject, same date, and same field, the proposed sentence is not supported; record the evidence gap.
- Input the actual
AIPROXY_API_ENDPOINTvalue. Compare whether it carries/v1or/chat/completions. If a suffix is present, the value does not match; keep the original value pending correction. - Input the probe interval value. Compare whether it falls inside the 5 to 60 minute integer range. If it falls outside, the value does not match; record the actual value.
Retrieval and relevance scores do not grant permission or isolate business entities. A shared vector model requirement for one library does not extend to unrelated external systems. A pending review does not establish a lack of operating permission, and an unsigned termination is not a completed termination. Where the source does not state a field, the value remains unknown rather than contradicted.