What data for this category looks like
Core data for this category comes from operation and maintenance scheduling systems, fault ticket databases, monitoring alert platforms, and organizational contact directories. Daily documents include fixed-cycle duty schedules, on-duty handover records, escalation contact lists, and daily duty reports. Data updates occur daily or weekly. Fields include duty time slots, on-duty personnel, open fault ticket numbers, handover items, emergency contact methods, and more. Field naming varies across different sources, with no unified standardized format.
What constraints these characteristics impose on the source citation and traceability link
Scattered data sources and inconsistent field naming require precise matching of field correspondences across different systems for citation traceability, to avoid confusing records from different duty cycles. High data update frequency requires traceability information to link to the latest scheduling and handover content in real time, to prevent use of outdated data. The number of documents associated in a single query is limited, so the recall range must be controlled to avoid information overload that impairs troubleshooting efficiency. Handover records and duty schedules have a strong correlation, so accurate binding of associated documents from the same cycle is required to prevent incorrect traceability across cycles.
How to configure the settings
These configurations apply to FastGPT v4.17.0 and above.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
quoteMaxToken | 800–1200 token | Matches the average length of duty handover documents, avoiding truncation of critical open items and contact information |
responseData | Return only associated fields | Focuses on core fields of duty schedules and handover records, reducing interference from redundant data |
detail | Enabled | Supplements the original format and contextual details of traceable documents, making it easier to verify the completeness of handover items |
| Maximum context citation count | Top 3 entries | Matches the number of documents associated with a single duty shift, avoiding information overload that impairs troubleshooting |
| Citation matching threshold | 0.75–0.85 | Calibrated through actual testing, balancing recall accuracy and recall range |
Three common misconfigurations
- Only field names are displayed without actual content in context citations: This occurs when the
responseDataparameter is not properly configured, and specific fields associated with duty schedules and handover records are not specified. - Historical duty handover records cannot be recalled: This occurs when the citation matching threshold is not set appropriately. A threshold that is too high filters valid content, while a threshold that is too low introduces irrelevant data.
- Quoted document formats cannot be rendered as Markdown: This occurs when the
detailparameter is not enabled, so original format information of documents is not loaded, and only plain text content is returned.
How to confirm the configuration is correct
- Upload a standard duty schedule and handover record, initiate a query that includes specific duty time slots, and check whether returned results include correct fields and corresponding content.
- Check the App/Admin service startup logs to confirm there are no validation errors for missing
AIPROXY_API_ENDPOINT(root address of the AI Proxy service, for example http://aiproxy:3000, do not append /v1 or /chat/completions) orAIPROXY_API_TOKEN(AI Proxy administrator token, consistent with the AI Proxy server configuration and cannot be empty). - Initiate a query that includes an open fault ticket number, and check whether returned citation sources accurately correspond to the corresponding fault tickets and handover records.
- Test two consecutive associated queries, confirm that traceability information from context citations is correctly carried over to the second request, and no traceability drift occurs.
The values given are common starting points and should be measured against the reader's own samples.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-10-19.