Agents

Abilities

Keeping things organised with labelling

Task labelling allows your agent to categorize and organize its work.

In the “Abilities section” of your edit agent interface, go to “Label tasks” and add a tag.

Name the tag and give clear descriptions for how the agent should use it.

For example, a blog writing agent might label the first blog it produces ‘First draft’.

Planning ahead with scheduled messages

You can give your agent the ability to plan and send messages at specified times in the future.

Enable “Scheduled messages” in the “Abilities” settings, then tell your agent when you want it to send a message in it’s core instructions.

Reaching out with ‘Escalate to humans’

When your agent encounters situations beyond its capabilities or is unsure how to act, it can get in touch with a human member of your team and ask for instructions, giving your agent a direct line to your team's expertise.

This feature acts as a safety net, ensuring that complex or sensitive tasks are handled appropriately, and maintaining the quality of your agent's output.

Setting up an escalation channels

In the Abilities settings of your agent, you can configure various communication channels for escalation. These include:

  • Email: the agent can send an email to a designated address or team.
  • Slack: integration with Slack allows the agent to post messages to specific channels or individuals.

Defining escalation criteria

You have two options for determining when your agent should escalate:

  1. Explicit Instructions: In your agent's core instructions, you can define specific scenarios or conditions that warrant escalation to a human. For example, you might instruct the agent to escalate when dealing with high-value transactions or sensitive customer information.
  2. Agent Discretion: Alternatively, you can allow the agent to use its judgment to decide when to escalate. This approach relies on the agent's ability to recognize its limitations and seek help when needed.

When escalation occurs, the human team member receives a notification through the chosen channel. This notification typically includes context about the task, the reason for escalation, and any relevant information the agent has gathered. The human can then provide guidance, make decisions, or take over the task as needed.

Subagents

Sometimes one agent isn't enough to handle complex tasks, and that's where subagents come in. Subagents are specialized agents that work under a main 'manager' agent, creating a hierarchical structure that allows for more sophisticated and diverse task handling. This feature in Relevance enables you to build advanced multi-agent systems, mimicking the structure of human teams.

Why use subagents?

  1. Specialization: Just as human teams have members with different expertise, subagents can be designed to excel in specific domains or tasks. For example, you might have one subagent specialized in data analysis, another in customer communication, and a third in scheduling.
  2. Scalability: Complex workflows can be broken down into manageable parts, each handled by a different subagent. This modular approach allows you to tackle intricate processes more efficiently and makes it easier to scale your AI operations as needs grow.
  3. Flexibility: By combining different subagents, you can create versatile AI systems capable of handling a wide range of tasks without the need to build a single, overly complex agent.

How to set up subagents:

Navigate to the "Subagents" section in your main agent's settings. This is where you'll manage your AI team.

Click "Add Subagent" to select from existing agents. You can choose agents you've already created or build new ones specifically for this purpose.

For each subagent, you have three options for how your manager agent interacts with them:

  • Autorun: The subagent is automatically activated when needed, without requiring approval.
  • Approval Required: The manager agent must get permission before activating the subagent.
  • Let Agent Decide: The manager agent uses its judgment to determine whether to autorun or seek approval.

These options allow you to balance autonomy and control in your multi-agent system.

When adding subagents, consider the following best practices:

  • Clearly define the role and responsibilities of each subagent.
  • Ensure efficient communication protocols between the manager agent and subagents.
  • Regularly review and optimize your agent hierarchy to improve overall performance.

Advanced settings

Configure template

The "Configure template" feature in Relevance AI is essentially a way to create customizable settings for your agent, similar to how you set up user inputs in the tool builder. This feature makes agent creation and sharing more efficient and user-friendly.

Firstly, it's a time-saver for repetitive information. If you're building an agent that frequently needs to reference certain data - like your company's name or description - you can create template settings for these details. Instead of copying and pasting this information into various places such as the flow builder or core instructions, you can simply reference the template setting. This approach not only saves time but also ensures consistency across your agent's configuration.

Our template feature shines when you want to share your agent with others. By setting up template configurations, you make it easy for other users to customize the agent to their needs. They can simply adjust the template settings to make the agent relevant to their specific use case, without having to dive deep into the agent's core setup.

These template settings are versatile and can be used throughout the agent's configuration - in core instructions, the Flow Builder, and even in tool settings. This flexibility allows for a high degree of customization while maintaining the core functionality of the agent.

Bulk scheduling

Bulk Scheduling is a feature available to Team Plus users on the Relevance AI platform. This functionality allows you to set up and schedule multiple tasks for your agent to run automatically.

How to access bulk scheduling

  1. Navigate to your agent's interface.
  2. Look for the dropdown menu next to the "+ New Task" button.
  3. Select "Schedule in bulk" from the options.

Setting up a bulk schedule

When setting up a bulk schedule, you'll need to:

  1. Select the knowledge table that contains the data you want your agent to process.
  2. Name your batch job.

Using the agent message template

The Agent Message template is where you define how your agent should interact with each entry in your knowledge table. This is done using a templating system that allows you to dynamically insert values from your table into the agent's instructions.

Key points about the agent message template:

  1. Double Brace Syntax: to insert values from your table, use double braces around the column name. For example: {{column_name}}
  2. Column Names: you can use any column from your selected knowledge table.
  3. Example Usage: if you have a column named "company" in your table, you could create a template like this: "Please analyze the performance of {{company}} based on the provided data."

Using the cadence contract

A key feature of bulk scheduling is the cadence contract, which allows you to fine-tune how and when tasks are distributed. This ensures that your automated workflows align perfectly with your team's schedule and work patterns.

Key components of the cadence contract include:

  1. Timezone:
    Select your preferred timezone to ensure tasks are scheduled according to your local time, for example: Australia - Sydney.
  2. Work Days:
    Specify which days of the week tasks should be scheduled.
  3. Work Hours:
    Set the time range during which tasks should be executed. You also have the option to select "All day" for 24/7 task execution.
  4. Task Distribution:
    Define the number of tasks to be sent per interval. For example, set a specific number of tasks per day. The actual number of tasks your agent can execute may vary due to factors like conflicting contract conditions and rate limiting.
  5. Override Mode:
    This give you the option to run all created tasks in override mode, which Tcan be useful for testing or when you need to maintain specific inputs/outputs across all tasks.

Best practices for cadence contract

Align with team availability

Set work days and hours that match your team's actual working schedule to ensure timely handling of task outputs.

Consider task complexity

Adjust the number of tasks per interval based on the complexity of each task and your team's capacity.

Start conservative

Begin with a lower number of tasks per interval and gradually increase as you assess your team's capacity and the agent's performance.

In summary

Once you’ve configured the cadence contract, hit ‘Schedule’.

You can check which bulk schedules you have set up by navigating to the three dots in the top right corner of your agent chat interface and selecting “Bulk schedules”.

Task template settings

When you need to modify an agent's behavior for a single task or a specific scenario, you can change the task template settings. This lets you make temporary adjustments without altering the agent's core configuration.

Let's say your agent has a template setting for a ' friendly personality,' which generally defines how the agent interacts. This setting probably works for most tasks, but there might be situations where you want to temporarily adjust it. If you need the agent to adopt a more serious tone for a particular task, you can override the 'personality' setting just for that interaction.

This override capability allows you to fine-tune your agent's behavior on the fly without the need to save changes to the overall agent settings. It's like giving your agent a temporary persona or skill set, tailored to the specific requirements of a single task.

Once the task is complete, your agent reverts to its default settings automatically. This means you can experiment with different configurations or respond to unique situations without worrying about permanently altering your agent.

Override Mode

When developing or refining an AI agent, it's important to test specific scenarios and responses. Override Mode lets you:

  • Hard-code inputs: You can manually set specific inputs for your agent, simulating particular situations or user queries.
  • Define outputs: Predetermined outputs can be set at various stages of the agent's task, allowing you to test how the agent handles specific information or decisions.

This level of control is really valuable when you're troubleshooting issues, verifying the agent's logic, or exploring how it handles edge cases.

Beyond testing, Override Mode offers a practical way to conserve resources:

  • Lock in inputs and outputs: by predetermining certain inputs and outputs, you prevent the agent from regenerating them repeatedly.
  • Save credits: this approach can significantly reduce the number of API calls or computations required, thereby saving on usage credits.

This is particularly useful in scenarios where parts of the agent's task remain constant across multiple runs, or when you're working with a limited budget.

How to use Override Mode

Navigate to the three dots in the top right corner of the agent chat interface and toggle on “Override Mode”.

Select whether you want to clone the task into override mode for testing, or go directly to Override Mode.

Click the gear icon to bring up the edit agent interface, and navigate to “Tools”. If a tool has been added to your agent, you can now toggle on and configure “Override input” or “Override output”.

Specify the inputs or outputs you want to override, then run the task with these overrides in place.

Remember, while in Override Mode, your agent will use the specified inputs and outputs instead of generating them dynamically, and this can affect the agent's normal decision-making process.

Learn Relevance

XXXXXXX To be determined
0:00