What data for this category looks like
Data for general comprehensive financial report analysis comes from publicly disclosed annual, semi-annual, and quarterly financial report files of listed companies, with update cycles aligned with financial report disclosure schedules. Most documents are full PDF content or standardized structured XBRL format files. Fields include various detailed item amounts, opening balances, ending balances, and change reason descriptions for other comprehensive income. Units are mostly Renminbi yuan or ten thousand yuan.
What constraints these characteristics impose on tool calling and plugins
Data is stored in PDF or XBRL format, so tool calling must integrate PDF text parsing plugins and XBRL structured data parsing plugins to extract valid fields. Update cycles fluctuate with financial report disclosure dates, so tool calling must support scheduled triggering of data pulling and updates to avoid using expired historical data. Fields include multiple detailed items and have varying units, so tool calling must support specifying detailed field filters and configuring unit unified conversion rules to ensure consistency in calculations and displays. Single financial report documents are lengthy, so the tool calling context window must support long text processing to avoid truncation of critical information.
How to set configurations
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
PARSE_FILE_TIMEOUT_SECONDS | 300 seconds | Aligns with the time required for long financial report parsing and data pulling, to avoid premature interruption |
maxContext | 8192-16384 tokens | Covers the context length of long financial report documents to prevent key fields from being truncated |
UPLOAD_FILE_MAX_SIZE | 200 MB | Adapts to the maximum size limit for single financial report PDF or XBRL files |
recallChunkSize | 800-1200 characters | Segment length when splitting financial report text, balances information completeness and retrieval efficiency |
similarityThreshold | 0.75-0.85 | Similarity threshold for filtering paragraphs related to other comprehensive income in financial reports, filters irrelevant content |
xbrlParseEnabled | true | Other comprehensive financial reports include structured XBRL format data, enabling this configuration allows direct extraction of standardized fields |
The parameter values provided on this page are common recommended starting points for configuration. Actual values are affected by material form, data volume, and business rules. Specific issues require individual analysis, and it is recommended to test on your own samples before finalizing settings.
Three common mistakes
- Symptom: Tool call returns
Tool call Parser nerror message and fails to parse model output normally. Cause: The inference model wraps thinking content in think tags and places it in the content field, which does not meet the standard parameter format requirements for FastGPT tool calls. - Symptom: Using an external fetch call to the
/api/v1/chat/completionsinterface returns a cross-origin error. Cause: The corresponding calling domain name is not added to the FastGPT cross-origin configuration whitelist, causing the browser same-origin policy to block the request. - Symptom: Tool calls do not respond after a new version is deployed, and checking logs reveals an MCP server connection failure. Cause: The MCP server address and port are not configured in the FastGPT plugin service management interface, or the server firewall blocks traffic on the relevant port.
How to confirm configurations are correct
- Upload a test general comprehensive financial report file, check if the parsed text fragments include detailed fields related to other comprehensive income, to confirm that the parsing configuration takes effect.
- Trigger a tool call, check if the units in the returned results are unified, to confirm that the unit conversion rule configuration is correct.
- View the FastGPT plugin management interface, confirm that all associated PDF parsing and XBRL parsing plugins are enabled, and that the configuration parameters match the preset values.
- Simulate a cross-origin request, confirm that there are no same-origin policy errors in the browser console, to confirm that the cross-origin whitelist configuration takes effect.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-09-14.