Planview Anvi agents

What are agents?

You can create and save agents, which are collections of up to five Anvi Chat prompts that you can run manually or on a schedule. You can also automate the steps (prompts) within an agent so that they do not require your direct intervention to run.

           

NOTE

Agents were called Workflows in earlier versions.

           

For scheduled agents, you can also send or receive emailed reports on the chat history, the last message sent, or errors.

           

Creating an agent

You can also create an agent by using a predefined agent in the Anvi Chat template library. See Planview Anvi Prompt and Agent Library for more information.

To create an agent:

  1. In the bottom left corner of the Anvi Chat pane, click Manage Favorites icon > gear icon Manage favorites.
  2. On the Manage favorites screen, click Agents > Create an agent.
  3. If you want the prompts in the agent to run without your direct intervention, enable the Automate Steps switch.
  4. Enter a Name for the agent.
  5. In the first Prompt cell, enter the first prompt you want to run in the agent.
  6. Click Add step and add another prompt. Repeat this step until you have added all the prompts you want the agent to run (up to five prompts).

               

    TIPS

    • Click and drag remove icon in a Step cell to rearrange steps.
    • Click remove icon next to a prompt to remove that step from the agent.

               

  7. Click Save.
  8. If you want to run the agent on a schedule, take the actions in the following procedure.

          

           

Scheduling an agent to run automatically

You can set up an agent to run automatically once at a specific date and time or to run repeatedly based on a schedule. You can also configure Anvi to email reports about the scheduled agent runs.

To run an agent automatically on a schedule and enable reporting:

  1. In the bottom left corner of the Anvi Chat pane, click Manage Favorites icon > gear icon Manage favorites.
  2. On the Manage favorites screen, click Agents and select the agent you want to schedule or click Create an agent and follow the steps in the previous procedure. 
  3. Enable the Run agent on a schedule switch.
  4. Click First run and select the first date on which you want the agent to run.
  5. If you want the agent to run more than once, click Repeat and select an option such as Daily or Biweekly.
  6. Click Time and select the time at which you want the agent to run.
  7. If you want Anvi to email reports about the scheduled agent runs, take the following steps:
    1. Enter an Email address to send reports to.
    2. Click the Report content list and select the type or report you want: Chat history, Last message, or Related errors.
  8. Click Save.

          

           

Viewing the history of agent runs

To view the history of agent runs:

  1. In the bottom left corner of the Anvi Chat pane, click Manage Favorites icon > gear icon Manage favorites.
  2. On the Manage favorites screen, click Agents and select the agent you want.
  3. Scroll down to the History section to see the status of all past agent runs.
  4. To view details of a specific agent run, click Manage Favorites icon for that run.

          

           

Best practices when using agents

Follow these best practices, tips, and tricks to maximize your experience when using Anvi Chat agents.

Best practice Explanation

Ask Anvi to return counts instead of lists

Scheduled agents run automatically and unattended. However, large sets of returned results, such as those in lists, can cut off silently. To avoid this, use prompts that ask for counts and aggregations instead of lists; counts and aggregations are reliably returned with small or large results. 

Chain simpler prompts together instead of overcomplicating prompts

Use a series of simple prompts, as in the following example, instead of overcomplicating a single prompt with multiple tasks:

  • Prompt 1: Specify the date and time context.
  • Prompt 2: Retrieve reference data (using the Knowledge Library).
  • Prompt 3: Ask for a specific analysis using the previous responses.

Specify the dates and times; avoid words like "today"

Agents don't inherently know concepts such as "today." Instead, start an agent with a date prompt. Then reference that date explicitly in subsequent prompts; never assume that the time context will carry over to future prompts. 

Match the data model's terms and vocabulary

Use the exact field names, entity types, and hierarchy terms from the product you are asking about. Mismatched terminology can cause unpredictable results.

Avoid ambiguous terms and conditions

The more specific the conditions in a prompt are, the more repeatable the results will be, as in the following examples:

  • Asking for "recent" results is ambiguous.
  • Asking for results "where the date is within the last 7 days" is specific and repeatable.

Test the agent before scheduling it

 Run the agent manually two or three times, testing all prompts in the agent, before scheduling the agent for automatic runs. If the results vary among the manual tests, refine the prompts until the results are stable. Inconsistent prompts become inconsistent agents.