Community Support

Purpose

Use Katara to automate answering community questions on native platforms based on your curated data collection(s) - this can include technical docs, website copy, blogs, code repos, and more.

Katara uses confidence scores and thresholds (see Q&A agent) to route AI generated responses—you are in full control of what gets sent to your users.

Prerequisites

  • An agent connected to one of your community platforms, with access to the necessary channels. Currently, only Discord is supported (but Telegram and Slack are coming soon!)

  • Website or GitHub agent(s) connected to source(s) that contain information the AI agent will reference to provide responses.

    • You must manually synchronize your collection once this data is collected by clicking the refresh (↻) icon on your collection.

  • A Q&A agent configured to use the information collected by these agents.

How it Works

This workflow consists of the following steps:

Inputs

  • Files: Files that contain FAQs, documentation, or other relevant information are collected from sources as set up by the user, and are synced into a collection.

  • Platform Integration: An agent must be connected to a community platform where questions will be asked.

Processing

  • The agent receives user questions from the community platform.

  • It processes these queries based on the files provided.

  • A confidence score is generated for each response. If the score meets or exceeds the set threshold, the response is posted directly in the same channel.

  • If the confidence score is below the threshold, the response is not sent back. You can choose whether to inform the user that the question could not be answered.

Outputs

  • Direct Response: If the confidence threshold is met, the AI posts its answer in the Discord/Slack channel.

  • Manual Review (coming soon!): If not, the response goes to the devrel channel for approval, denial, or modification. The manual response can replace the AI-generated one, further training the agent.

Example

  • Input: A user asks, "How do I reset my password?" in Discord.

  • Processing: The agent reviews the markdown documentation and generates a response.

  • Output: The response is posted directly in the channel.

    Community Support Workflow

Advanced Features

  • Adjusting Confidence Threshold: Users can fine-tune the confidence level for when a response should be manually reviewed versus automatically posted.

  • Custom Workflow Combinations: The workflow can be combined with other workflows like document parsing or task automation to streamline operations.

Troubleshooting

  • Agent Not Responding: Ensure the files are properly uploaded and the correct channels are configured.

  • Low Confidence Threshold: If many responses are not meeting the confidence threshold, consider adjusting the agent's training data or lowering the confidence threshold.

  • Log Issues: Logs for workflow activities can be accessed from the admin dashboard to track issues or performance problems.

Last updated