What the documents look like
Fruit-store procurement copy drafting draws on purchase orders, acceptance records, pesticide-residue test reports and loss records. The underlying inputs include wholesale market prices, supplier quotations, store sales and test reports, refreshed daily or per batch. Copy that cites a grade, origin, test conclusion or batch must stay consistent with the batch and status recorded on the corresponding document.
Two industry facts shape this page. Fruit purchasing relies on wholesale market conditions and grade-based acceptance rather than a headquarters standard meal specification, so grade and origin wording depends on the specific acceptance record. Grade inspection and loss control consume substantial time, and copy that references a test conclusion must match the batch and report identifier on the source document.
Constraints at the model connections step
Model connections depend on AI Proxy. From V4.17.0, AI Proxy is a required dependency. 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 item causes startup validation to fail. Legacy OpenAI environment variables or a single-model custom request address cannot substitute for these two settings.
Model capability is checked against the input the application needs. Multimodal, tool-calling and reasoning settings are verified separately. Application parameters are set in the application or node configuration; a same-named field on a compatible chat interface is not assumed to override orchestration.
Hypothetical paper sample: a store fresh-produce supervisor submits an acceptance-record excerpt. The source text reads: "12 March 2026, batch A-0312, Red Fuji apple, grade one, pesticide-residue test report number JC-0312, test conclusion pass, accepted quantity 200 kilograms." A draft sentence prepared the same day reads: "This batch of Red Fuji apples has passed pesticide-residue testing, grade one, welcome to purchase." The scope of judgment is that batch and that day's acceptance record.
Configuration and implementation checks
The table lists documented settings or implementation checks relevant to this feature and primary task. The values are hypothetical starting points for this paper exercise, chosen so they can be compared against documented ranges or field requirements. They are not official defaults.
| Configuration item | Hypothetical starting point for this exercise | Basis for comparison |
|---|---|---|
AIPROXY_API_ENDPOINT | Fill the AI Proxy service root address, without appending /v1 | Documentation requires the root address; a missing value fails startup validation |
AIPROXY_API_TOKEN | Fill a non-empty administrator token consistent with the server configuration | Documentation requires a non-empty token that matches the server |
| Language model and index model | Enable at least one of each | Documentation states the system needs at least one language model and one index model |
Model capability vision | Check against whether the copy includes image input | Documentation states multimodal models support image recognition |
| Model status probe | After enabling, schedule at the configured interval | Documentation states probing covers only enabled system models, with an interval that is an integer from 5 to 60 minutes |
| Model call-log retention | Check against LLM_REQUEST_TRACKING_RETENTION_HOURS | Documentation states this variable controls retention duration |
Three common errors
- Treating the draft sentence "has passed pesticide-residue testing" as a general conclusion covering all batches. The input is the sample source text and the draft sentence; the comparison target is batch A-0312 and test report number JC-0312. If the draft sentence does not limit itself to that batch or report number, it exceeds what the material supports.
- Using an application API key instead of the administrator token for
AIPROXY_API_TOKEN. The input is the token value actually entered; the comparison target is the server configuration. If the two differ or the value is empty, startup validation fails. - Reading a probe "error" status as a business conclusion about the copy. The input is the
eventandstatusreturned by the probe; the comparison target is the draft sentence content. A probe result reflects model availability only and does not determine a test conclusion or grade.
How to verify the setup
- Input the sample source text and the draft sentence, then compare whether the draft sentence limits itself to batch A-0312 and test report number JC-0312. If the draft sentence says "all batches" or omits the batch, the support condition is not met.
- Input the actually entered
AIPROXY_API_ENDPOINTandAIPROXY_API_TOKEN, then compare whether the endpoint is a root address and whether the token is non-empty and consistent with the server. If the endpoint has/v1appended or the token is empty, the configuration condition is not met. - Input the probe interval from the model probe configuration, then compare whether it is an integer between 5 and 60 minutes. If it falls outside that range, the documented configuration condition is not met.
All three checks are paper comparisons. No probe or startup validation was actually executed.