Skip to Content
AI & AgentsAutonomous agents

Autonomous agents

BiVelio’s autonomous agents operate continuously, subscribed to system events. When an event occurs that matches their activation conditions, the agent executes automatically without human intervention.

How they work

System event (message received, case created, etc.) → Event engine evaluates subscriptions → Autonomous agent activated → Agent processes context (event data + history) → Agent executes actions (respond, classify, escalate, etc.) → Result recorded in the activity log

Lifecycle

  1. Configuration — the system prompt, subscribed events and permitted actions are defined
  2. Activation — the agent is marked as active and begins listening for events
  3. Execution — each time a subscribed event occurs, the agent executes
  4. Logging — each execution is recorded with input, output and credits consumed
  5. Deactivation — the agent can be paused or deactivated at any time

Available autonomous agents

Support Autoresponder

Events: Message received in ChannelHub Function: Generates contextual responses to support messages, taking into account the conversation history and contact information.

  • Responds automatically outside of office hours
  • Escalates to an operator if urgency or high complexity is detected
  • Maintains a professional and empathetic tone configured by the workspace

Knowledge Base RAG

Events: Message received with an identifiable question Function: Queries the workspace knowledge base to answer frequently asked questions using Retrieval-Augmented Generation (RAG).

  • Searches through documents, FAQs and workspace articles
  • Generates natural responses based on verified sources
  • Cites sources so the user can find further information

Multichannel Triage

Events: Message received on any channel Function: Classifies incoming messages by topic, urgency and channel, and assigns them to the most suitable operator or team.

  • Analyzes message content to determine the topic
  • Assigns priority based on keywords and context
  • Routes to the operator with the appropriate specialization
  • Creates cases automatically if needed

Gmail Ops

Events: Email received via Gmail integration Function: Manages email operations: classifies messages, prepares responses and manages pending follow-ups.

  • Classifies emails into categories (support, commercial, administrative, spam)
  • Generates response drafts for review
  • Detects emails requiring follow-up and creates reminders

Customer Followup

Events: Case with no activity for X days (configurable) Function: Detects customers with no recent activity and generates proactive follow-up actions.

  • Identifies cases or contacts with no recent interaction
  • Generates personalized follow-up messages
  • Suggests actions to the assigned operator

Quality Compliance

Events: Message sent by an operator Function: Reviews outgoing communications to ensure compliance with workspace quality standards.

  • Evaluates tone, professionalism and completeness
  • Detects spelling errors or incorrect information
  • Generates alerts if the communication does not meet standards

SLA Risk

Events: Case reaches 75% of the SLA time limit Function: Monitors cases at risk of breaching SLA and takes preventive action.

  • Notifies the assigned operator and supervisor
  • Suggests actions to resolve the case before the deadline
  • Can automatically reassign if the operator does not respond

Draft Reply

Events: Message received in ChannelHub Function: Generates response drafts based on the conversation context for the operator to review and send.

  • Analyzes the full conversation history
  • Takes into account contact data and associated cases
  • The draft remains pending human approval before being sent

Each autonomous agent consumes AI credits with each execution. Typical consumption is 1-5 credits per execution, depending on the complexity of the processing.

Configuring an autonomous agent

To activate an autonomous agent:

  1. Go to AI Agents from the sidebar
  2. Select a template from the marketplace or create a custom one
  3. Define the system prompt (or use the AI prompt generator)
  4. Select the events to subscribe to
  5. Configure the activation conditions (optional)
  6. Define the permitted actions (respond, classify, escalate, etc.)
  7. Activate the agent

The agent will start operating immediately, processing the events that match its configuration.

Last updated on