Agent Settings
Large Model Configuration
【Model】Select the large model that the agent will call
【Output Language Preference】Set the preferred output language of the agent. By default, it automatically matches the user’s language, but you can also fix it to a specific language
【Agent Prompt】Define the agent’s identity, skills, personality, and answer constraints. This is where you specify what you expect the agent to do; you can use the AI prompt optimization feature to generate it with one click
【Context Memory Rounds】Set how many rounds of historical messages are carried into each Q&A for the model. The higher the setting, the more points consumed, but the better the model performs
【】When enabled, user questions will be optimized based on conversation history. This consumes additional points and may slightly delay responses
Knowledge Configuration
Configure knowledge bases for the agent
Currently, the following knowledge base types are provided:
1. Online Document Knowledge Base: Can be used as a help center’s online documentation; only published sections and articles are retrievable by the agent. The downside is that corpus content cannot be precisely managed
2. Text Document Knowledge Base: Unstructured corpus data. Supports importing by uploading local documents, custom creation, or collecting web data. Data is not visible externally and can be managed by slicing the content
3. Product Database Knowledge Base: Structured data entered according to a fixed template. Allows the agent to return product card formats to guide users in placing orders
4. FAQ Document Knowledge Base: Structured data entered according to a fixed template. FAQ data can be directly returned via retrieval matching without calling the AI, reducing point consumption
【Show Knowledge Source】The agent will show the source of the answer in its response
【Find Similar Products】When enabled and a product database knowledge base is configured, users can search for similar products via the Salesmatly web plugin
【Knowledge Matching Score】The minimum matching score for the agent to match knowledge snippets (excluding FAQ documents, where the matching score is set separately within the document)
【No Matching Knowledge Response】When a user asks a question and no knowledge snippet meets the minimum matching score, you can choose whether the AI should generate a reply based on the settings or use a default reply
Preview and Debugging
Every configuration change is saved as a draft in real time and can be tested in the preview/debug window on the right
Publishing and Integration
Once the agent configuration is complete, you can click 【Publish】 to go live. The available usage channels can be viewed under 【Integration】
Publishing requires a version description. You can restore the agent configuration to any previously published version under 【Version】
Note: The API channel must be enabled for the agent to be used in Salesmatly