This document applies to FastGPT v4.17.0, and all configurations below are based on the interface and API rules of this version.
What the data for this category looks like
The core documents for this category include training courseware, phishing email templates, drill plans, and drill result reports. Data sources typically include employee contact lists, corporate email gateways, internal training platforms, and user click statistics. Data update frequency is mostly quarterly or annual. Document structures include crowd tags, email content templates, drill workflows, and result statistics fields. There is no unified mandatory format, and configurations must adapt to the enterprise’s existing security operation ledger.
Constraints on Multi-turn Dialogue and Prompt Engineering
Dispersed data sources and inconsistent formats require multi-turn dialogue to support field extraction across data sources, and prompts to adapt to structural differences across different documents. Low update frequency requires prompts to support reuse and incremental updates of historical drill data. Time-consuming steps such as drill crowd screening and template design rely on parameter passing in multi-turn dialogue, so variable passing rules in the dialogue workflow must be clearly defined. The result review process needs to associate click statistics data, so prompts must support aggregated display of multi-dimensional results.
How to Configure Settings
| Configuration Item | Recommended Approach | Rationale |
|---|---|---|
chatId | Bind the unique identifier of the drill scenario | Distinguish dialogue contexts for different phishing drills, avoid cross-scenario data confusion |
messages | Split dialogue turns by drill phase | Adapt to segmented call requirements for training courseware, templates, and reports, align with multi-turn Q&A workflow logic |
variables | Pass employee department and drill batch parameters | Achieve precise matching for crowd screening, adapt to security training needs for different departments |
| Model Selection | Use general embedding models for text understanding, large language models for dialogue | Adapt to structured and unstructured processing requirements for category documents, improve Q&A accuracy |
| Single Recall Limit | Top 3 entries | Control the volume of returned historical drill data, avoid redundant information interfering with current Q&A |
| Timeout Threshold | 60 seconds | Adapt to latency issues across data source calls, ensure response efficiency for pre-sales and technical support |
| Maximum Context Length | 800–1200 characters | Accommodate core content of a single drill plan and template, avoid context overflow |
Three Common Configuration Mistakes
- Phenomenon: Calling the knowledge base association interface returns empty results, with no drill-related documents matched. Cause: The exclusive knowledge base ID of the drill scenario was not passed via
variables, so the dialogue cannot locate document data for this category. - Phenomenon: The model confuses content from different batches of phishing email templates during multi-turn dialogue. Cause:
chatIdwas not bound to distinguish scenario contexts, leading to mutual interference of dialogue data from different drills. - Phenomenon: Calling the agent creation API returns an undefined interface error. Cause: The official FastGPT agent creation interface was not used, and general dialogue interface parameters were misused.
How to Verify Successful Configuration
- Initiate a test dialogue, input "View the email template for this phishing drill", verify whether the returned content includes the exclusive template content for the currently bound scenario.
- Pass test department parameters via
variables, initiate a query such as "Filter the training crowd for the corresponding department", verify whether the returned results only include employee information for the corresponding department. - View the dialogue log, confirm that
chatIdmatches the unique identifier of the current drill scenario, with no cross-scenario context residue. - Simulate a delayed request scenario, confirm that the system returns a response or a clear timeout prompt within 60 seconds.
The values provided here are common starting points, and should be evaluated against the samples of the relevant use case.
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.