Skip to main content

Application Settings

The Application settings in influence various aspects of the Application's behavior and performance across different components and user interfaces. These settings ensure a consistent and optimized experience for users interacting with the Application, whether it's through the knowledge base, AI models, or navigation metrics.

General​

  • Application Name: Assign a name to your Application for easy identification.

  • Website Location: Enter the exact URLs where the User Interfaces will be embedded on your website. Multiple URLs can be specified if the User Interfaces are to appear on various pages.

  • Knowledge Base Language: Select the language and country code for your Knowledge Base content. For a multi-language Knowledge Base, add combinations of two-digit ISO language and country codes (e.g., en_US for English/United States). This setting ensures that users see content tailored to their language and locale, leveraging branchly's extensive multi-language database support.

  • Status: Choose between Development and Production modes. Switch to Production when your user interfaces are ready to go live.

  • Dynamic Navigation: Configure dynamic navigation based on usage data within a specific timeframe.

    • Primary Metric: This metric is the primary method for edge weighting. The initial recommendation is to stick with weight_manual and weight as default options.

      tip

      Tip: Starting with weight_manual and weight is advisable. You can later transition to a dynamic edge weighting method based on accumulated data.

    • Secondary Metric: Used when the primary metric does not yield a distinct edge weight for a node or when node values are identical. This metric helps determine the weighting for these nodes.

    • Edge Weighting Methods:

      • weight_manual: Ideal for ensuring certain nodes consistently appear prominently. For instance, highlighting a specific contact node prominently.
      • weight: A customizable value set during data upload to the database.
      • weight_rolling_30: Calculates average weights from the past 30 days of usage.
      • weight_rolling_60: Averages weights from the past 60 days of usage.
      • weight_rolling_alltime: Averages weights from all usage data since the Application's deployment.

  • Data Retention Policy: Defines the days after which Request data from users is deleted. This means insights/analytics will not show data for this time period, with the exception of number of active sessions.

Search​

  • Dense Embedding Model: Proprietary AI model that supports major languages and excels in semantic search tasks.
  • Sparse Embedding Model: Proprietary Embedding model that supports major languages and excels in information retrieval tasks, especially advanced keyword search. We currently offer a custom bm25 implementation.
  • Search Mode: How to search for information in the knowledge base. Applies to all interfaces. Options:
    • Dense: Default semantic search.
    • Hybrid (recommended): Hybrid search for improved keyword and semantic retrieval using sparse and dense Embedding Models. Results from both searches are combined using fusion algorithms.
    • Sparse: Sparse embedding model only.
  • Search Result Highlighting: This setting allows you to customize which text is highlighted in the search results. Choose Chunk for the best-matching text snippets or Page Metadata Description to use the description from the result’s meta description.
  • Rerank Settings: These settings help you fine-tune the order of search results. Turn on the sections you need and adjust them to guide which answers appear first. This is an advanced feature. You can differentiate for which interactions (search or chat) these settings apply.
    • Datetime reranking
      • Pushes newer content higher so fresh articles show up sooner than older ones.
      • Lets you decide whether to look at the published_date or the modified_date when judging what counts as "new". Both fields are part of the Page Metadata.
    • Record source reranking: Use this to give a higher score to results based on the title or the text fields.
    • Custom boosting: Uses the boost_score field in the custom_metadata to increase the rank accordingly. Use this in combination with mode and modifier build a powerful and custom search function specific to your use case.

Chat​

  • Chat Strategy: You can choose between Agent and Standard RAG. This setting activates tools (see AI Actions). Can only be activated, if at least one knowledge_base tool is activated.
  • Reply in User language: Activated by default. Here, we infer the language of the user input and the chat replies in the user language. In case the user language is not clear, the chat defaults to using the specified locale.
  • Routing Model (deactivated): The current model in use for handling user requests is from Azure OpenAI service.
  • Generation Model (deactivated): Response generation in the Chat and Search Interface utilize  gpt-5 models from the Azure OpenAI Service.
tip

Note on Alternative Models: Other model providers (aside from OpenAI) available are Mistral, Cohere, DeepSeek or Qwen.