This document is written for FastGPT v4.17.0.
What This Category’s Data Looks Like
Typical users include SEM optimizers, agency operations specialists, and brand marketing professionals. Daily processed documents include keyword lists, negative keyword tables, bid records, and search term reports. Data is primarily sourced from search ad backends, search term reports, bidding ranking data, and landing page data. Update frequency is mostly daily or weekly.
Document structure is fixed, with fields including keyword text, match type, bid amount, search volume, and more. Units include CNY, requests, characters, and more. Unlike other ad categories, this category focuses on user-initiated search terms. Core workflow time is concentrated on keyword expansion, negative keyword filtering, and bid matching adjustments.
What Constraints Do These Characteristics Impose on Model Integration and Configuration
Multi-source, fixed-structure data for this category requires models to adapt to multiple document formats during integration, to avoid field mismatch errors. Daily or weekly update frequency means model context must support bulk historical data import. Context length should not be set too short.
High time consumption in core workflows requires model calls to include automated tool capabilities, to reduce manual intervention. Errors directly impact ad delivery effectiveness. A complete log collection and failure retry mechanism must be configured. Model output must strictly match ad delivery compliance requirements, to avoid generating prohibited keywords or bid suggestions.
How to Set Configurations
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
maxContext | 8000–12000 token | Adapts to bulk context input from multiple documents including keyword lists, negative keyword tables, and search term reports |
maxTokens | 2000–3000 token | Meets long-text output requirements for negative keyword filtering, bid suggestions, and keyword expansion |
temperature | 0.2–0.4 | Ensures output consistency for keyword judgment and negative keyword filtering, avoids generating irrelevant or prohibited content |
toolChoice | auto | Automatically calls tools to complete tasks such as keyword matching and data querying, improves processing efficiency |
AIPROXY_API_ENDPOINT | Enter the root address of the AI Proxy service, for example http://aiproxy:3000. Do not append /v1 or model service address paths | Required environment variable for the App/Admin server, specifies the AI Proxy service access address |
AIPROXY_API_TOKEN | Enter the administrator token matching the AI Proxy server configuration. Cannot be empty | Required environment variable for the App/Admin server, used for AI Proxy service identity verification |
Three Common Configuration Mistakes
- Phenomenon: The agent execution details page shows that the model response completion reason includes "connection disconnected". Troubleshooting: Check if the App/Admin server has correctly set
AIPROXY_API_ENDPOINT(root address of the AI Proxy service) andAIPROXY_API_TOKEN(AI Proxy administrator token). Missing or incorrect configuration causes service startup verification failure, model call link interruption, or permission verification failure. - Phenomenon: The model provider management page does not display call logs. Cause: The interface log collection switch is not enabled, or the log storage configuration is not correctly bound to the current deployment environment.
- Phenomenon: A custom-deployed model cannot run normally in the agent link, but knowledge base search tests work normally. Cause: The tool call permission for this model is not enabled in the agent configuration, or the model input format does not match the keyword data structure of this category.
How to Confirm Proper Configuration
- Run a single keyword expansion test, verify that the model output keywords match the search term characteristics of the current category and the fields of the input keyword list.
- View the call log module on the model provider management page, confirm that the latest interface call records are generated.
- Simulate a negative keyword filtering task, verify that the model output results match the field rules of the local negative keyword table.
- Check the response status code on the agent execution details page, confirm that a normal status code is returned with no abnormal error prompts.
The values provided 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 2026-10-19.