Indication-Based Drug Use Q&A with Tool Calling and Plugins

Data Characteristics for This Category

Biopharmaceutical indication data primarily originates from drug inserts, pharmacopoeias, clinical guidelines, and various medical literature. This data is typically structured or semi-structured, describing diseases or symptoms for which specific drugs are approved. The data update frequency is relatively stable, with updates occurring when new drugs are launched or existing drugs receive new indications. However, changes are less frequent than for drug interactions or adverse reactions. In terms of document structure, indication information usually includes fields such as disease name, drug name, dosage and administration, and precautions for special populations. Disease names may involve ICD codes or SNOMED CT terms, while drug names often use generic or brand names, potentially with ATC classification codes.

Constraints Imposed by These Characteristics on "Tool Calling and Plugins"

The relatively stable update frequency of indication data allows for a more relaxed caching strategy for tool calls, reducing unnecessary external API calls. Its high degree of structure helps define tool input and output precisely using methods like JSON Schema, improving parsing efficiency and accuracy. The presence of disease and drug codes requires plugins to support mapping and conversion across various coding systems when performing queries, ensuring seamless integration with underlying knowledge bases. Furthermore, indication descriptions may contain complex medical terminology and conditional statements, demanding strong text comprehension and conditional judgment logic from plugins to avoid erroneous recommendations due to semantic misunderstandings. For numerical information like dosage and administration, plugins must identify and convert units to ensure data consistency.

Configuration Settings

Configuration Item	Recommended Value	Rationale
`tool_timeout`	`60 seconds`	Most indication queries complete quickly, avoiding long waits.
`max_tokens`	`2000`	Indication descriptions can be long; ensures complete return.
`function_call_strictness`	`high`	Precisely matches indication query intent, preventing false triggers.
`schema_validation_level`	`strict`	Ensures tool input and output conform to predefined JSON Schema structure.
`cache_ttl`	`24 hours`	Indication data updates infrequently; cache period can be extended.
`max_retries`	`3`	Provides fault tolerance for network fluctuations or temporary external service unavailability.

Three Common Mistakes

Symptom: When a user queries for "hypertension" drug indications, the results are empty or irrelevant. Reason: The plugin incorrectly identifies or maps the disease name, leading to a mismatch between query parameters and knowledge base codes.
Symptom: The plugin calls an external API and receives a "format error." Reason: The JSON data structure returned by the tool does not conform to FastGPT's expected text or api_call node definitions, for example, missing required fields.
Symptom: After multiple indication entries are returned, some critical information, such as dosage and administration, is missing. Reason: The max_tokens setting is too small, causing the tool's output to be truncated and incomplete indication descriptions to be retrieved.

How to Confirm Correct Configuration

Simulate different phrasing for typical indication queries. Verify if the plugin accurately identifies the intent and triggers the correct tool call.
Inspect tool call logs. Confirm that the field names and data types of input parameters and returned results match expectations, paying special attention to disease and drug code mapping.
Use FastGPT's debugging interface to step through the tool calling process. Verify that configurations like tool_timeout and max_retries function as expected under abnormal conditions.
Test with indication cases containing long text descriptions. Ensure the max_tokens configuration is sufficient to accommodate complete information without truncation.

The values provided are common starting points and should be measured 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 2026-09-21.