What this category of data looks like
The service inquiry data for municipal water utilities mainly comes from the internal management systems of municipal water operations. It covers guide content for official service items including water supply connection, drainage repair, water quality query, payment processing, and other related matters. The data follows a layered sync update rhythm: water quality monitoring data updates every hour, while service work orders and frequently asked questions update daily. Documents present a mix of structured fields and unstructured text. Structured fields include item name, processing time limit (unit: working days), required material list, inquiry phone number, and more. Unstructured content consists of detailed procedures for item processing and frequently asked question answers.
What constraints these characteristics impose on tool calling and plugins
Since the data includes both structured fields and unstructured text, tool calling must support both structured parameter queries and unstructured content retrieval to avoid returning incomplete service information. The high-frequency update feature requires plugins to be configured with scheduled sync tasks to ensure that the latest water utility service data is used for calls. Fields have dedicated units. For example, water quality index units are mg/L, and processing time limit units are working days. Tool calling must automatically adapt to unit formats to prevent returning value expressions that do not meet user expectations. Additionally, inquiries in this scenario focus on specific items. The context for tool calling must be limited to the relevant item range to avoid interference from redundant information.
Reference only; verify against the target FastGPT release and deployment documentation.
How to configure the settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Chunk length | 800–1200 characters | Adapts to the length of individual water utility service guide content, avoids information fragmentation within chunks |
| Maximum number of single tool calls | Top 3 entries | The number of water utility service items is concentrated, excessive entries will cause context redundancy |
| Tool call timeout | 60 seconds | Matches the typical response duration of municipal water utility system APIs |
| Reranking strategy | From highest to lowest relevance | Prioritizes the most frequently consulted service items |
toolChoice | auto | Allows the model to independently decide whether to call tools for water utility service inquiries |
functionCall | Bound to custom water utility service functions | Ensures the data source for tool calling matches the water utility data in the knowledge base |
Three common errors
- Tool call results are not linked to knowledge base retrieval content. This occurs because the water utility service data from the knowledge base is not bound to the tool call data source, preventing the tool from retrieving detailed processing procedures for corresponding items.
- Tool call response timeout. This happens when the tool call timeout setting is too short, and does not match the typical response duration of municipal water utility system APIs.
- Tool call returns empty results. This happens when no scheduled sync task for water utility service data is configured, so the knowledge base has not loaded the latest service items and question-and-answer content.
How to confirm the configuration is correct
- Submit a test query such as "How to apply for water meter installation", check if the tool call returns the corresponding service item content.
- View the tool call logs to confirm the response duration meets the set timeout requirement.
- Check the data source sync status to confirm that the latest sync update of water utility service data has been completed.
- Test a query involving unit expressions, such as "Is the current water quality up to standard", confirm that the indicator units returned by the tool match official expressions.
The configuration values provided are common starting points. They should be evaluated against relevant sample data and operational requirements.
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.