Reference only; verify against the target FastGPT release and deployment documentation.
What the data for this category looks like
This category’s data comes from internal archived system documents of independent management units, including departments and subsidiaries. These documents cover departmental regulations, operating procedures, data ledgers, and more. The update rhythm is determined by each unit’s internal system revision process, with no unified cycle. Most documents are formatted PDFs, transcribed Word texts, or structured online documents. They contain fixed fields including issuing unit, document number, effective date, and main body content. These fields belong to each internal management unit.
What constraints do these characteristics impose on deployment and upgrade
Data is split and stored by independent management units. Deployments must configure multi-tenant isolated storage and indexing rules to prevent cross-unit data leaks. Different units have different system update cycles. Upgrade processes must use incremental synchronization instead of full re-import to reduce impact on online services. Documents have diverse formats and structured fields. A multi-format parsing pipeline must be preset before deployment to ensure consistent field extraction. Upgrades must also be compatible with each unit’s existing custom field mapping rules to avoid breaking existing retrieval logic.
How to set the configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
DB_MAX_LINK | 100–300 seconds | Adapt to the timeout of batch retrieval links for multi-unit system documents, avoid service blocking from large document queues of a single unit |
MONGODB_URI | Cluster connection address sharded by management unit | Store system metadata and retrieval indexes for each unit; sharding by unit reduces cross-unit data interference |
REDIS_URL | Exclusive cluster address with password authentication, memory quota no less than 2048 MB | Cache retrieval sessions and temporary parsing results for each unit, reduce high-frequency read/write pressure on the database |
STORAGE_S3_ENDPOINT | Object storage endpoint with buckets divided by unit | Isolate original system documents uploaded by each unit, prevent unauthorized cross-unit file access |
FILE_TOKEN_KEY | Randomly generated 32-character encrypted string | Encrypt access tokens for files uploaded by each unit, ensure access security of document resources |
Three common misconfigurations
- Symptom: After offline private deployment, the workflow editing interface displays client error messages. Cause:
FE_DOMAINis not configured to the actual access domain name of the current deployment, causing cross-domain loading failure of front-end static resources. - Symptom: Compatibility errors appear when accessing the platform via browser, and the retrieval interface fails to load normally. Cause: The browser version used is not supported by FastGPT, or local browser cache has not been cleared.
- Symptom: An error indicating missing rehype-raw module is prompted when building the deployment image, even after running the dependency installation command. Cause: The build process did not fully pull all dependency packages, or local cached dependency packages are missing.
How to confirm the configuration is complete
- Run the
docker pscommand, confirm all service containers are in running state with no abnormal exit records. - Upload a test system document, bind it to a specified management unit, run a keyword search, confirm returned results only include documents from that unit.
- Access the deployment address corresponding to
FE_DOMAIN, open the workflow editing interface, confirm no client error pop-ups appear. - Check the storage bucket corresponding to
STORAGE_S3_ENDPOINT, confirm uploaded test documents are stored classified by management unit.
The values provided here are common starting points, and should be evaluated against the user’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 2026-09-28.