What the data for this category looks like
This category’s typical documents include requirement research notes, PRDs, user story lists, and requirement change forms. Data sources are internal team requirement management tools, meeting records, and ticket systems. Document update frequency is mostly weekly or per version release cycle. Structures typically include business objectives, acceptance criteria, function descriptions, boundary conditions, and change records. Fields include requirement ID, priority, affected modules, change time, and more. Priority mostly uses P0-P3 tiers. Some documents include functional test case tables.
What constraints do these characteristics impose on the document parsing and chunking link
The hierarchical structure of requirement documents requires that chunking must not destroy the integrity of chapters and logical units. Avoid splitting core information such as acceptance criteria or change records. Frequently updated documents require parsing processes to adapt to fast iteration rhythms, reducing unnecessary full parsing. Documents include tables and custom fields, so parsing tools must support structured extraction. This prevents loss of table content such as function lists and acceptance items. Internal documents have no public web attributes. Parsing only supports locally uploaded formats, no additional web scraping configuration is needed.
How to set the configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Matches the length of single logical units in requirement documents, such as single acceptance criteria or single user story. Avoids splitting core information |
chunkSplitter | Split by heading hierarchy | Requirement documents have clear chapter hierarchies, such as requirement overview, boundary conditions, change records. Splitting by hierarchy preserves context integrity |
customPdfParse | Enabled | Most requirement documents are in PDF format. Enabling this calls enhanced parsing capabilities to correctly extract tables and nested structures |
PARSE_FILE_TIMEOUT_SECONDS | 300–600 seconds | Large PRD collections or multiple change forms take longer to parse. Prevents parsing failure due to timeout |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Adapts to the scale of team bulk uploads of requirement document collections, meeting bulk processing needs |
chunkSettingMode | Manual configuration | Requirement document structures are relatively fixed. Manual configuration accurately matches the chunking logic for business documents |
Three Common Mistakes
- Symptom: After enabling enhanced PDF parsing, the knowledge base still uses the default engine for parsing, and table content is lost. Cause: The
customPdfParseparameter is not correctly configured to point to the local MinerU service address, or the MinerU service is not running properly. - Symptom: Parsed chunked content splits a single acceptance criterion. Cause: The
chunkSizesetting is too small. Hard splitting by character count destroys the logical integrity of acceptance criteria. - Symptom: Large PRD documents fail to parse, and the interface shows a timeout error. Cause: The
PARSE_FILE_TIMEOUT_SECONDSsetting is lower than the actual parsing time of the document, causing parsing to interrupt.
How to confirm the configuration is correct
- Upload a test PRD PDF document. Verify that parsed chunks retain chapter hierarchies and table structures. This document is written based on FastGPT v4.17.0.
- Navigate to the knowledge base settings page. Confirm that the
customPdfParseswitch is enabled, and the MinerU service port configuration matches local deployment parameters. - Upload a requirement change collection document of approximately 100 MB. Wait for parsing to complete, then check for timeout errors.
- Review chunked content. Confirm that single acceptance criteria or user stories are not split.
The values provided are common starting points and should be measured against local 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.