Data for This Category
Data for this category comes from enterprise existing IT system inventories, official directories of domestic software and hardware, reports generated by independent innovation migration test environments, and adaptation materials provided by vendors. Daily output documents include software and hardware adaptation lists, compatibility test reports, migration plans, and issue ledgers. Fields included are software and hardware model, version number, compatibility status, test conclusion, and migration steps, among others. Data update frequency adjusts based on project progress or quarterly cycles. Adaptation data for different migration projects is independent, with no unified global update cycle.
Constraints for Knowledge Base Retrieval and Recall
Adaptation documents are structured content tied to specific models and versions. Retrieval must prioritize keyword matching, while also supporting semantic understanding of migration plans. Data is isolated per project. Retrieval scope must be limited to the knowledge base set for the corresponding project to avoid cross-project data interference. Individual documents are lengthy, and include non-text content such as test screenshots. The number of recalled entries must be controlled to prevent context overflow. Retrieval exceptions for non-text knowledge blocks must also be handled. Pre-sales Q&A often requires expanding user questions into compatible scenario descriptions. Retrieval logic must be optimized to improve recall coverage.
Configuration Settings
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Independent innovation adaptation documents rely on precise model matching. A threshold that is too low will introduce irrelevant compatibility results, while a threshold that is too high may miss existing system entries |
limit | Top 8–12 entries | Adaptation documents are lengthy individually. Too many entries will exceed the context window, while too few will fail to cover multiple model compatibility scenarios |
searchMode | semantic+keyword hybrid mode | Balances the need for model keyword matching and semantic understanding of migration plans |
usingReRank | Enabled | A large number of similar model entries exist in adaptation documents. Reranking improves precise matching accuracy |
datasetSearchUsingExtensionQuery | Enabled | Pre-sales Q&A often requires expanding user questions into compatible scenario descriptions, which improves recall range |
collectionIds | Bind exclusive knowledge base set ID per project | Adaptation data for different independent innovation migration projects is independent, avoiding cross-project interference |
Three Common Misconfigurations
- Failure to enable
datasetSearchUsingExtensionQuery, or setting thesimilaritythreshold too high. This may result in failure to recall knowledge blocks with specified keywords during semantic retrieval. - When using a strict question-answering template, failure to configure filtering logic for non-text knowledge blocks. This may result in a "no answer" prompt after retrieving image address knowledge blocks.
- Failure to use the
collectionIdsparameter to specify the correct project-specific knowledge base set ID, or passing a collection ID that does not belong to the current project. This may result in a permission or data range error after passing the knowledge base parameter during API calls.
How to Verify Correct Configuration
- Check FastGPT v4.17.0 App/Admin server environment variables. Confirm that
AIPROXY_API_ENDPOINT(root address of the AI Proxy service, for example http://aiproxy:3000; do not append paths such as /v1 or /chat/completions) andAIPROXY_API_TOKEN(AI Proxy administrator token, consistent with the AI Proxy server configuration and not empty; do not use this for application API call authentication) are correctly configured and not empty. Service startup verification will fail otherwise. - Perform knowledge base retrieval in the FastGPT backend, input typical adaptation questions, and confirm that the number and relevance of recalled results meet expectations.
- Call the application API while passing the
collectionIdsparameter, and confirm that the returned results only include adaptation documents for the specified project, with no cross-project data included. - Test knowledge blocks containing image addresses. Confirm that adaptation descriptions are returned normally when using a non-strict template, and non-text content is automatically filtered when using a strict template.
The values provided here are common starting points, and should be evaluated against the samples of the relevant users.
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.