What Data for This Category Looks Like
The core data of vulnerability scanning services comes from vulnerability scanners, asset inventories, CVE databases, and patch bulletins. Daily output documents include scanning plans, vulnerability lists, rectification work orders, and rescan reports. Fields cover asset IP or domain name, CVE number, vulnerability severity, repair status, rescan time, and more. Data updates run weekly or monthly. A single scan result set includes multiple types of structured entries. It also includes some unstructured text fragments associated with vulnerability details. Standardized fields form the core support for the overall data set.
How Data Characteristics Constrain Model Access and Configuration
The data characteristics of vulnerability scanning services directly constrain model access configuration design. Multi-source data requires configurations to support mixed parsing of structured fields and unstructured text. Configurations must adapt to the formats of different documents such as scanning plans and vulnerability lists. Weekly or monthly update cycles require model call caching policies to match the data update period. This avoids invalid repeated pulls. Single-task multi-entry data structures require context-related parameters such as maxContext to reserve sufficient length. This fully accommodates all information from a single batch scan. The time-consuming false positive elimination process requires configurations to balance analysis accuracy and generation efficiency. Overadjusting randomness parameters can cause result deviation, so this must be avoided. The FastGPT v4.17.0 configuration interface includes optimized default parameters for multi-source adaptation for these scenarios.
How to Set Configurations
| Configuration Item | Recommended Value Range | Rationale |
|---|---|---|
maxContext | 8000–16000 token | Adapts to the full vulnerability details and context requirements of a single batch scan task |
maxTokens | 2000–4000 token | Meets the length requirements for generating defect analysis conclusions and rectification suggestions |
quoteMaxToken | 500–1000 token | Limits the length of referenced vulnerability detail fragments to avoid context overflow |
temperature | 0.2–0.4 | Reduces generation randomness to improve the accuracy of defect classification and false positive elimination |
AIPROXY_API_ENDPOINT | Fill in the root address of the AI Proxy service, for example http://aiproxy:3000 | Access root address for the AI Proxy service. Do not add path suffixes such as /v1 or /chat/completions. Do not point to a single model service address |
AIPROXY_API_TOKEN | Fill in the non-empty administrator token consistent with the AI Proxy server configuration | Identity verification credential for the AI Proxy service. Do not use application API keys, model vendor keys, or other business keys |
Three Common Configuration Errors
- Testing model access returns a 404 status code. Troubleshooting steps include:
AIPROXY_API_ENDPOINTis not correctly set to the AI Proxy service root address, orAIPROXY_API_TOKENis not correctly configured or is empty. - A batch scan task completes, and the model continues to occupy video memory. The cause is missing model auto-unload logic, or
maxContextis set too large, pushing video memory usage beyond server thresholds. - Multiple API keys cannot be managed using business groupings. Troubleshooting steps include: identity credential grouping binding is not completed via model channel configuration, or access parameters for the corresponding model channel are not correctly set.
How to Confirm Correct Configuration
- Run a model analysis task for a single vulnerability list entry. Check if the returned defect analysis results include the severity level and rectification suggestions for the corresponding CVE.
- Check the access logs of the AI Proxy service. Confirm that the requested
AIPROXY_API_ENDPOINTandAIPROXY_API_TOKENmatch the configured values. - Simulate a model call for a batch scan plan. Observe server video memory usage. Confirm no abnormal continuous occupation occurs.
- Configure two sets of different model channel identity credentials. Verify that grouped calls using different credentials function correctly. Confirm the grouping logic behaves as expected.
The values listed here are common starting points. They should be evaluated against the specific samples of the deployed environment.
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.