Invoke agents in Opal Chat

  • Updated

Agents automate a specific task in Opal Chat, such as drafting content, analyzing data, or auditing a page. Invoke an agent whenever you want it to run that task for you.

Invoke an agent in whichever way suits your task. Call an agent directly when you already know the one you want. Start from a chat pill or the Agent Library when you want a ready-made prompt or need to browse.

  • Chat pills – A selection of recommended actions that Optimizely features on the home page of a new chat, organized by task category. Optimizely updates them periodically. A chat pill can run an agent or send a plain prompt.
  • Agent Library – A browsable catalog of agents only. Use it to explore the available agents and read what each one does before you run it.

Both chat pills and the Agent Library pre-populate the chat input when you select them. Edit the prompt before you submit it.

Invoke an agent manually

To invoke an agent directly, complete the following steps:

  1. Open Opal Chat. See Access Opal Chat.
  2. Click Mention Agent, enter @AGENT_ID, or click View Agents.
  3. Select an agent from the list, or begin entering the agent's ID.
  4. Add any required variables, such as URLs or files.
  5. Press Enter or click Submit.

Opal runs the agent and displays each step as it works.

Screenshot of Opal as it's running an agent and working.

After the agent finishes, Opal displays the response. Depending on the output, the result displays in a canvas or as a separate agent output. Click the canvas or agent output to view the full response.

Screenshot of Opal Chat showing an agent response displayed as a canvas element

Use a chat pill

Chat pills are pre-populated prompts that help you get started. They display as buttons at the bottom of a new chat, organized by task category. Selecting a pill inserts a ready-to-use prompt into the chat input. Required parameters display highlighted for editing.

To use a chat pill, complete the following steps:

  1. Open Opal Chat. See Access Opal Chat.
  2. Click a category button (Brainstorm, Create, Repurpose, Test, Analyze, or Learn). A panel opens showing pre-populated prompts relevant to that category. Prompts that reference an agent display as @AGENT_NAME.

    Screenshot of the Create category panel in Opal Chat where pre-populated prompt options display for selection.
  3. Click a prompt. The prompt populates the chat input. Any parameters you need to fill in display highlighted in brackets, such as [topic or launch] or [audience].
  4. Click a highlighted parameter or press Tab to move between parameters.
  5. Enter your value to replace the parameter text.
  6. Press Enter or click Submit. Opal runs the prompt, invoking any referenced agents as needed, and displays the response.
If you delete or deactivate an agent associated with a chat pill, the chat pill prompt no longer displays. If you re-add or reactivate the agent, the pill prompts update accordingly.

Agent Library

The Agent Library is a browsable catalog of agents on the Opal Chat page. Open an agent to see what it does, then click Use agent to populate the chat input. See Agent Library overview.

Opal populates the library automatically for all users and displays a curated subset of the agents in the Agent Directory. To remove an agent from your library, delete it from Agents > Your Agents. See Delete a specialized agent.

Multi-turn agent sessions

When you invoke a multi-turn agent, Opal Chat enters an active session with that agent. Opal Chat displays a Working with AGENT_NAME banner at the bottom of the chat, and the input field updates to Message AGENT_NAME. All prompts you send go directly to the agent until the session ends.

Screenshot of Opal Chat showing an active multi-turn agent session with a Working with agent name banner at the bottom of the chat and an agent-specific message input field

After each response, Answer completed displays to confirm the agent has finished its turn. Enter a follow-up prompt to continue the conversation.

The session stays active until one of the following occurs:

  • The agent completes its task.
  • You click Done in the session banner to end the session manually.
  • You start a new thread or clear the context.