What Data for This Category Looks Like
The data for this category comes from CMDB, cloud resource inventories, auto-discovery tools, and procurement ledgers. Update frequency is daily or weekly. Daily produced documents include configuration item lists, service component relationship topologies, change records, and asset inventory sheets. Core fields include configuration item unique identifiers, asset classifications, upstream and downstream dependency relationships, last update times, and responsible persons. All units are standard IT operations and maintenance fields, with no custom non-standard units.
Constraints Imposed by These Characteristics on Context and Token Processing
Data for this category comes from multiple sources, with significantly different update rhythms. Context recall must prioritize matching the latest daily updated data to avoid introducing expired configuration item information. Dependency relationships between service components are core content. Token allocation must prioritize retaining upstream and downstream associated topology fields to prevent truncation of critical link information. Structured documents account for a high proportion of total content. Context recall must focus on core fields such as configuration item IDs and dependency relationships to reduce token usage by non-essential text. When the total number of configuration items is large, the total token amount for context recall must be limited to avoid exceeding model support limits.
Configuration Settings
These configurations are based on FastGPT v4.17.0.
| Configuration Item | Recommended Approach | Rationale |
|---|---|---|
maxContext | First 15–25 associated data entries | Adapts to the scale of configuration item quantities in CMDB, prevents context token from exceeding model limits |
quoteMaxToken | 6000–10000 token | Matches the average token length of structured documents for this category, covers core configuration and dependency information |
| Context time filtering threshold | Within 7 days | Matches the daily/weekly update frequency of this category, filters expired configuration item data |
| Recall field scope | Only include configuration item ID, asset type, dependency relationship | Reduces token usage by non-essential text, focuses on core business fields |
| Single search association limit | First 3 layers of topology relationships | Avoids excessive recall of redundant dependency links, controls total token consumption |
Three Common Misconfigurations
- Phenomenon: Context recall token consumption far exceeds expectations, and the interface shows single-call token exceeds the model limit. Cause: The number of recalled entries for
maxContextis not limited, and full CMDB data is included in the context. - Phenomenon: Search results cannot combine the dependency relationships of configuration items, and the returned content is irrelevant to the current configuration scenario. Cause: The recall field scope is not limited, and a large amount of non-core redundant text is recalled.
- Phenomenon: After private deployment, the custom context path when accessing the application does not take effect, and access via the specified path is not possible. Cause: The context path parameter is not correctly set in the deployment configuration, resulting in routing matching failure.
How to Confirm Successful Configuration
- Run a test script to generate requirement documents, check whether the returned content includes the dependency relationships and last update times of the current configuration items, and confirm that core fields are correctly recalled.
- Check the token consumption statistics, confirm that the single-call token value is within the preset range of
quoteMaxToken, with no abnormal overflow. - Check the context path parameter in the deployment configuration, verify that the specified path can access the application normally, and confirm that the routing configuration takes effect.
- Simulate an expired data scenario, confirm that configuration item data exceeding the set time is not included in the context, and verify that the time filtering rule takes effect.
The values provided 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.