Skip to main content
After creating, training, and testing your AI agent, you can activate it to start serving customers automatically. You can also deactivate it at any time — for maintenance, updates, or improvements — without losing any of your training or configuration.

​
When to activate an agent

Activate your agent when:
  • You have completed the essential training
  • You have configured the necessary skills
  • Testing has produced satisfactory results
  • You have validated the agent’s behavior and tone

​
When to deactivate an agent

Deactivate your agent when:
  • You need to make adjustments or improvements
  • You are performing maintenance or updates
  • You have identified issues in the agent’s live behavior
  • You want to run temporary A/B tests with different configurations

​
How to activate an agent

1

Open AI Agents

In the left sidebar, click AI Agents.
2

Select your agent

Click on the agent you want to activate.
3

Enable the Active toggle

On the agent detail page, find the Active toggle and turn it on.
Make sure the agent has been thoroughly trained and tested before activating it. A poorly configured agent can negatively affect the customer experience.

​
How to deactivate an agent

1

Open AI Agents

In the left sidebar, click AI Agents.
2

Select your agent

Click on the agent you want to deactivate.
3

Disable the Active toggle

On the agent detail page, find the Active toggle and turn it off.
4

Confirm deactivation

The agent will immediately stop taking new conversations.
When you deactivate an agent:
  • The agent stops handling new conversations immediately.
  • Conversations already in progress with the agent are not affected.
  • The agent can be reactivated at any time.
  • All training and configuration is preserved.

​
How an active agent handles conversations

​
Automatic assignment

When an agent is active, new conversations can be automatically routed to it based on your team’s routing configuration. The agent starts responding immediately once assigned.

​
Manual handoff

A human agent can transfer any conversation to the AI agent at any time. This is useful when the agent identifies that the AI can resolve the customer’s request more efficiently.

​
Visual indicators

When an AI agent is actively handling a conversation, the interface shows:
  • AI icon — Visible in the conversation header
  • Agent name — Identifies which agent is responding
  • Active status — Confirms the agent is live in that conversation

​
Managing multiple agents

If you have more than one agent, you can:
  • Activate multiple agents at once — Each handles its own specialty.
  • Deactivate temporarily — Pause an agent without losing its configuration.
  • Switch between agents — Activate and deactivate as your operational needs change.
  • Test in parallel — Compare the performance of different agents before committing to one.

​
Best practices

  1. Activate gradually — Start with one agent at a time. Monitor its performance before activating others.
  2. Monitor after activation — Follow the first conversations closely, verify responses are accurate, and be ready to intervene if needed.
  3. Keep human backup — Always have human agents available. Configure skill-based handoffs for complex cases and avoid relying entirely on AI in the early stages.
  4. Deactivate for maintenance — When making significant changes, deactivate the agent first, test your changes, and reactivate only after validating them.
  5. Document changes — Keep a record of when the agent was activated or deactivated, and note the reasons for any changes.

​
Important notes

  • Inactive agents do not appear as options when initiating a handoff.
  • You can have multiple agents active simultaneously.
  • Activation and deactivation take effect instantly.
  • There is no limit on how many times you can activate or deactivate an agent.
  • Deactivated agents retain all training, skills, and settings.
Consider activating your agent initially only during off-peak hours or for specific conversation types. This lets you validate its performance in a controlled way before a full rollout.