Build AI Agents

Create New AI Agent

  • From the left navigation bar on the ClearFeed WebApp, go to AI > AI Agents.

  • Click on “New AI Agent”.

Configure your Agent

Configure Knowledge Sources and Response Settings

Under Knowledge Sources:

  • Select Add All to allow the AI Agent to use all indexed knowledge sources.

  • Or choose Pick Specific to limit the AI Agent to selected sources, and use tags to specify which ones to include.

Advanced Settings:

  • Display AI-Generated Answers: Choose whether to show or hide AI-generated responses.

  • Share Search Results: Enable the AI Agent to display relevant search results along with its answer, and configure how many results to show.

  • Enable AI Assistance Beyond Knowledge Base: Allow the AI Agent to respond to queries even when they go beyond the information available in your connected knowledge sources.

Add Integrations

  • Under Integrations, connect any external tools (like JIRA, HubSpot, Zendesk) that the agent will use for executing actions.

  • Under Advanced Settings, you can control the level of access granted to the AI Agent by specifying which actions it is allowed to perform for each integration. You can choose to:

    • Grant specific permissions manually

    • Allow only current permissions

    • Enable all current and future permissions

    By default, when an integration is connected, all current and future permissions are enabled — this can be customized anytime.

Provide Instructions for the AI Agent

You can tailor the AI Agent’s behavior by customizing its instructions. Instructions define the Agent’s tone, style, and response rules—ensuring it aligns with your support guidelines and organizational policies.

Key Instruction Areas

  1. Restricted Queries Specify a list of queries that the AI Agent must never answer. This ensures sensitive or non-permissible requests are ignored.

    • Example restrictions: system credentials, admin access, salary information.

    • When a restricted query is detected, the Agent will skip answering.

  2. Response Tone and Length Define how the Agent should respond. Keep responses clear, concise, and in line with your brand’s voice.

    • Example: “Answer in 2–3 short sentences, using professional and simple language.”

  3. Response Language Control which language the Agent uses. By default, it replies in the same language as the query.

    • Example: Always respond in English.

  4. Knowledge Lookup & Fallback When connected to Knowledge Sources, the Agent attempts to find relevant information to answer queries.

    • If information is found, generate the response.

    • If no information is found, the Agent uses a fallback message.

    Default fallback:

    “We could not generate a response for your query.”

    Custom fallback example:

    “We could not find any relevant answer for your question. An HR representative will follow up with you, or you can ask a different question to our HR bot in the thread by tagging @HRBot.”

Using Placeholders in Instructions

You can include placeholders to give the AI Agent relevant context:

  • Tool placeholders: You can also include placeholders from tools like Jira, GitHub, or Zendesk to provide the agent with relevant context for generating accurate responses.

  • Conversation analysis: Use {{chat_analysis.process_chat}} to let the Agent analyze, summarize, or process conversation history before responding.

Best Practices

  • Keep instructions specific and unambiguous.

  • Update the restricted queries list regularly.

  • Match tone and language rules with your brand style.

  • Test fallback and placeholders in real scenarios to validate outcomes.

Customize Look and Feel

  • Customize the Name and Profile Photo – these will be shown in Slack responses posted by the AI Agent.

    • Use the Edit icon to update the AI Agent's Display name.

    • Paste the image URL into the Profile Photo field under General Information.

  • Add a brief description of the agent to help your team understand its role.

Test your Agent

Use the built-in Test Interface to preview how your agent will respond to real messages using your current instructions.

What you can do:

  • Simulate conversations: Send test messages and review how the agent responds — before deploying changes live.

  • See the Execution Plan: Each response now includes a detailed execution plan showing:

    • Tools selected, along with inputs

    • Multiple tools used in a step (stacked for clarity)

    • References and sequential steps

    • Messages indicating when no tool was used

This ensures your instructions are working as expected before going live by allowing you to quickly edit, test, and iterate — all from within the same interface. The execution plan view lets you review the tools, steps, and references used, giving you full confidence that your AI Agent behaves as intended.

Assign AI Agents to Collections

Create your AI Agents and assign them to a Collection to start using them.

  • Go to your Collection > Settings

  • Open AI Agent settings

  • Toggle on the AI Agent

  • Select either the Virtual Agent or the Agent Assistant based on your needs.

    1. Learn more about them here.

    2. Note that you cannot enable both for a single collection.

  • Choose the preferred AI Agent for the Collection

  • (Only for Virtual Agent) Trigger mode:

    • Manual: Respond with a generated answer only when an emoji (🔍) is applied to the thread. (Note: Manual mode does not work with automatic ticketing).

    • Automatic: Generate an answer for all requests on a thread.

  • Bot Interactions: Lets you tag ClearFeed (@clearfeed) to invoke the AI Agents and take actions or generate answers

    • Learn more about Bot Interactions here.

    • This is enabled by default for Agent Assistant.

Last updated

Was this helpful?