Creating Agents

Follow these steps to create custom AI agents

Name and description

Give your agent a name, icon, and short description. The description helps others quickly understand what the agent does.

Triggers

A Trigger is an event that automatically starts an Agent, transforming it from an on-demand tool into a proactive agent running in the background. You can set up Triggers to run Agents based on a schedule, specific events in your other software, or incoming webhooks.

A Trigger is not required if you only plan to run an Agent manually.

Based on a schedule

This trigger runs an Agent on a recurring, time-based interval. You can set it to run hourly, daily, weekly, or monthly.

Based on events

This trigger runs an Agent automatically when a specific event occurs in one of your connected systems. This allows you to automate workflows in response to real-time business activities.

Based on a webhook

This trigger runs an Agent when it receives an HTTP request from an external system. This is a powerful way to integrate with internal tools or any third-party platform that can send webhooks, even if it doesn’t have a direct integration with Realm.

For detailed step-by-step instructions on setting up each trigger type, see Creating triggers.

Agent instructions

Instructions are the most important part of an agent. They define how the agent behaves, what it should do, and how it should approach the task. You can write them manually or use Write with AI to generate a starting point.

Specificity improves results. Clear goals, step-by-step guidance, relevant context, and explicit references to tools or documents all help the agent perform more reliably.

The key concepts in agent instructions are:

• Purpose: Start by clearly describing what the agent is meant to do. For example: “Your task is to help me prepare for discovery calls. Gather relevant background information and produce a concise summary that helps me understand who the customer is, what they do, and how we might fit.”
• Step-by-step guidance: Break the workflow into numbered steps. This makes the agent’s process easier to follow and improves consistency.
• References to the right context: You can point the agent to specific documents, integrations, tools, or even other agents. For example, you might ask it to review a specific document in one step, check HubSpot in another, or call another agent as part of the workflow. To insert these references, type @ in the instructions field or click the insert icon.
• Variables for reusable workflows: Use {{ }} to define variables the user fills in when running the agent, such as {{Customer}} or {{Deal}}. This is useful for reusable workflows like a Customer Health agent, where the agent can run against any company based on the input provided.

Tools

Tools allow agents to take actions in external systems, such as sending emails, updating CRM records, creating tickets, or posting messages. Realm includes built-in tools and also supports integrations via MCP (Model Context Protocol).

In the agent editor, you can control which tools an agent can use and configure how each action behaves.

For each action you can choose:

Setting	Description
On	The agent can run the action automatically
Ask	The agent must request user confirmation before running the action
Off	The agent cannot use the action

You can also configure parameters for each action:

• Preset parameters: Lock a parameter to a fixed value (for example, always post messages to a specific Slack channel).
• Hidden parameters: Hide parameters from the model so they cannot be modified or accessed by the agent.

For more details on managing tools across your organization, see Admin guide: Tools.

Knowledge

The Knowledge section defines what indexed context the agent can access when running.

When you add knowledge to an agent, it becomes searchable. The agent can use it when it’s relevant, but it won’t read everything by default.

If you need the agent to always consider a specific source, tag it. Tagged content is included every time the agent runs. Untagged content is only used when the agent decides it’s relevant.

Allow all

The agent can access all connected knowledge sources. Existing document permissions are respected, so users only see information they already have access to.

Limited

Restrict the agent to specific folders, channels, or files. Selecting a folder includes all subfolders and content within it. Use this when the agent should only work with a defined set of sources.

No knowledge

The agent runs without access to internal company data and behaves like a general-purpose AI.

You can also control whether the agent can use Web Search, or leave it for the user to enable when running the agent.

Additionally, you can upload files directly from your computer to include them in the agent’s knowledge.

Advanced settings

Advanced settings allow you to control how the agent reasons, performs research, and processes information.

• Thinking mode: Controls how deeply the agent reasons about a task. Fast prioritizes speed, Research performs deeper multi-step reasoning for complex tasks, and Auto lets Realm decide the best mode.
• Research agents: Allows the agent to delegate sub-tasks to other specialized agents. This is useful when you want different agents to contribute specific expertise during a workflow.
• Data analysis: Enables the agent to analyze structured data. When enabled, the agent can calculate metrics, work with spreadsheets, and generate charts.
• Image generation: Allows the agent to generate or edit images using AI. Enable this if the agent needs to create visuals as part of its workflow.
• Skip search: Allows the agent to answer questions directly from the selected knowledge without running a search step. This setting is only available when the agent’s knowledge base is small and clearly defined, typically less than 20 documents.
• Search filters: Defines default filters used when the agent searches for information. Filters can restrict results by attributes such as author, date, project, or other metadata.
• Override model: By default, Realm selects the best language model automatically. You can enable Override model to choose a specific model (for example, Claude 4.6 or Gemini 3.0) for the agent’s final answer.
• Temperature: Controls how predictable or creative the agent’s responses are. Lower values make the agent more consistent. Higher values make the agent more creative and varied in its wording. For most business workflows, a lower temperature (around 0.1 to 0.3) typically produces the most reliable results.

Testing and iterating

The agent editor provides a side-by-side chat view, allowing you to test your agent with different queries and see the results in real time. Refine the instructions, adjust knowledge sources, and tweak the configuration until the agent performs exactly as needed.

Sharing agents with your team

Share an agent with your team so others can use it directly or customize their own copy. By default, team members with access can run the agent from the shared link or the agent page, and you can also give individual users access with User or Editor permissions.

You can control who can access the agent from the sharing menu:

• Restricted: Only people who already have access can use it through the link.
• Anyone in Realm: Anyone in your Realm workspace can use the agent.
• Specific people: Add individual teammates and choose whether they are a User or Editor.

Use Copy Link to share the agent quickly, and update access at any time from the sharing settings.