What the data for this category looks like
The data sources for meeting minutes primarily include on-site meeting recording transcripts, scanned handwritten meeting notes, and draft files submitted by attendees. Updates follow a per-meeting cadence. One independent minutes file is generated for each meeting. Typical document structures contain fixed modules: meeting topic, start time, attendees, discussion topics, resolved items, and follow-up items. Some files include quantified fields: number of attendees (unit: people) and meeting duration (unit: minutes). Most content consists of natural language paragraphs, with colloquial expressions and temporary supplementary content mixed in.
What constraints these characteristics impose on model integration and configuration
The fixed modular structure of meeting minutes requires models to accurately identify and extract specific fields such as topics and resolved items. Model output must follow clear format rules. Colloquial expressions and temporary supplementary content in raw text require models to have text standardization and error correction capabilities. This prevents generation of redundant information that deviates from original content. Each meeting generates a single independent file, so no incremental update logic is needed. Only parsing and organizing work for individual files is required. Meeting minute lengths vary widely. Long files may exceed the base context window. Adaptable context capacity parameters are therefore necessary.
Configuration Recommendations
| Configuration Item | Recommended Value Range | Rationale |
|---|---|---|
maxContext | 8000–12000 token | Adapts to the total length of most meeting minutes, accommodates raw text and parsing instructions |
maxTokens | 2000–3000 token | Reserves sufficient space for structured output, ensures complete output of content such as resolved items and follow-up items |
temperature | 0.2–0.4 | Maintains accuracy of output content, reduces the probability of generating colloquial redundant content |
AIPROXY_API_ENDPOINT | Interface address deployed by local MinerU, such as http://localhost:3001/api/v1/parse | Meets parsing requirements for PDF and recording transcript files of meeting minutes, ensures raw files can be read normally |
AIPROXY_API_TOKEN | Authentication token for the corresponding interface | Ensures call security of the parsing interface, prevents unauthorized access |
toolChoice | auto | Allows the model to independently determine whether to call the parsing tool to process raw meeting files, adapts to input data of different formats |
Three Common Mistakes
- Model icons do not display in workflows, and file uploads prompt network errors. The cause is incorrect configuration of
AIPROXY_API_ENDPOINTorAIPROXY_API_TOKEN, leading to failed parsing interface calls. - Generated meeting minutes lack core fields such as resolved items and follow-up items. The cause is an excessively low
maxTokensvalue, which cannot accommodate complete structured output content. - Meeting minute content generated by the model deviates from original meeting discussion content, with large amounts of irrelevant content appearing. The cause is an excessively high
temperaturevalue, leading the model to generate redundant information beyond the original content.
How to Confirm Successful Configuration
- Use a curl command to call the interface corresponding to
AIPROXY_API_ENDPOINT, upload a test meeting minute file, and verify that the interface returns valid parsed data. - Upload a test meeting minute text in the FastGPT v4.17.0 workflow, check whether the model loads normally and triggers the parsing process.
- Adjust the
temperatureparameter to the 0.2–0.4 range, compare output content across different values, and confirm that output meets minute standardization requirements. - Check the values of
maxContextandmaxTokens, ensure they cover the total length and output requirements of the test minutes.
The values listed here are common starting points, and should be evaluated against the reader'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.