AI Agents
Capacity AI Agents allow Administrators to manage custom AI Agents across the Capacity platform. AI Agents also come with a default full screen Chat Page experience by defaut.
AI Agent Settings
Capacity AI Agents are the default place to build and manage agents in Capacity. Administrators can create and manage agents from Settings → AI Agents, which provides a table of AI Agents built in the Organization.
Clicking the row for an Agent takes the Administrator to the Agent Settings page for that specific AI Agent.
The Agent Settings page includes the following sections:
General Settings
Model Options
Data Sources
Dev Platform
Chat Page Settings
General settings
The general settings include fields to define the name, description, and most importantly the prompt which defines how the agent should behave and answer. The prompt instructions are the most important part of any agentic experience.
Model options
The Model Options section allows the AI Agent builder to choose the model and temperature for the AI Agent. Below are some basic definitions for the fields.
Model
The LLM used by the agent, for example GPT-4.1, Gemini, Claude Sonnet, or other models.
Temperature
A value set between 0 and 1. It determines how “creative” the agent is allowed to be.
0 — No creativity; minimal hallucination risk.
1 or 2 — Very creative; higher hallucination risk.
Data sources
The Data Sources section allows the builder to choose which search “tools” the agent may use, including the Knowledge Base vector search, searching third party sources via Answer Engine, and searching external web sources.
Using the “Allow user to select sources…” option, Administrators can allow Chat Page users to limit which sources the agent can search. The “Source Collection Name” field controls the name of that collection of selectable sources in the right sidebar of the Chat Page.
To learn more about how to properly reference Data Source “tools” in your prompt, read the AI Agent Tools section below.
Dev Platform
The Dev Platform section allows the builder to give the agent access to Dev Platform App Actions as tools which the agent can call.
Agents will have access to the App Action information shown as context, including the App name, App Action name, inputs, and outputs, as well as which inputs are required.
To learn more about how to properly reference Dev Platform “tools” in your prompt, read the AI Agent Tools section below.
Chat page settings
The Chat Page Settings section lets the builder customize the default full screen chat experience, including
New Chat button text,
Message input placeholder,
Whether the AI Agent selector dropdown includes this agent
Whether this agent is the default agent for the Organization at agent.capacity.com
Full Screen Chat Page
Capacity AI Agents each have a default full screen Chat Page located at agent.capacity.com. The default full screen Chat page can be opened from the AI Agents table or individual AI Agent settings page. For anyone with experience with the most common Agentic experiences, the interface should feel very familiar.
Chat Page Access
At this time, AI Agent Chat Pages are available to any authenticated user, including Chat role users. If your Organization has Chat role users who are not employees, please be aware that any authenticated user can access AI Agent chat pages if they have the URL to the agent. Do not share agents externally
Searches and tools however run as the signed-in user, so people only see knowledge and results they are allowed to access.
Public unauthenticated Chat Pages are not available at this time.
Agent Selector
The Agent Selector dropdown at the top right allows authenticated users to select different agents to interact with. Administrators can limit which agents are available to select in this dropdown from the individual AI Agent settings page.
Chat Thread
User messages are sent from the message input at the bottom of the center section. The placeholder text for the input can be customized in the AI Agent settings page.
The message input is a multi-line field. Hitting Enter sends; Shift+Enter / Alt+Enter adds a new line. Sending shows a loading state until the reply finishes
The chat thread shows messages sent by the user, and responses from the AI Agent. The interface automatically transforms markdown responses into a human friendly format, so we recommend giving prompt instructions for AI Agents to reply in markdown.
Chat History
Individual user chat history is stored in the left sidebar. Users can pick back up a conversations as long as the history is still stored. Users should be aware that the AI Agents have limited context windows and will only be aware of the more recent messages in the conversation up to that limit.
Source Sidebar
If the AI Agent builder has enabled the option “Allow user to select sources…”, and provided a Source Collection name, the Sources sidebar will show on the right side of the Chat Page. The user chatting with the AI Agent will then have the option to “limit” which sources the AI Agent can search by selecting or deselecting source options.
These manual selections by the user will override any defaults or prompt instructions regarding the Search tools specifically. The following rules apply:
Only selected sources will be used by the search tools provided to the AI Agent
If all sources (i.e. folders, etc.) are deselected then the tool will no longer be used by the agent.
If the user deselects all options no search tools will be used.
When the “Allow user to select sources…” option is not selected, and the AI Agent uses the tool, all sources selected in the AI Agent settings will be used by default.
AI Agent Tools
AI Agents use “tools”, sometimes referred to as “functions”, to perform external actions. Capacity AI Agents leverage Search tools and Dev Platform tools to perform those external actions, which can be any App Action from any App.
Search Tools
Search tools connect the AI Agent to your Knowledge Base, Answer Engine sources, or external Web sources. What the agent can retrieve matches the current user’s permissions: the agent uses the user’s access when running searches, so sensitive content stays restricted.
When building an agent, you can choose which folders, sources or urls to include. If you allow user-selected sources in chat, end users can narrow which folders or items apply from the sources selected in the settings for that conversation
Dev Platform Tools
Dev Platform App Actions can be added as tools so the agent can trigger third party tools and requests. The agent receives context about each app action as defined in the dev platform.
Searches and tools run as the signed-in user, so people only see knowledge and results they are allowed to access.
How to Reference Tools in Prompts
Capacity AI Agents let Admins have full control over the prompt and behavior of their AI Agents. While this is useful in customizing the behavior of AI Agents, it also means the builder must provide detailed instructions to the AI Agent in order for it to behave correctly.
For AI Agents to use Search or Dev Platform tools correctly, the AI Agent Prompt instructions must reference them in a way the AI Agent understands.
Referencing Search Tools
Search tools can be referenced by the following names
Answer Engine search tool (answer_engine_search)
Web search tool (tavily_search)
However, the “Knowledge Base” vector search tool is a special situation. The Knowledge Base search tool does NOT need to be referenced in the prompt. To make Agent responses as fast as possible, if you have added a Knowledge Base vector search tool in the “Data Sources” section, it will always run the Knowledge Base search. This allows Agents to get results and context faster, which in turn speeds up their response time.
The one exception is if the user is allowed to manually deselect Knowledge Base sources, and chooses to do so. In that situation, the Knowledge Base vector search will not run.
Referencing Dev Platform Tools (App Actions)
To reference App Action tools correctly, use the name of the Tool in the section. For example, the Helpdesk “Create ticket” app action would be referenced as the “Create Ticket’ Helpdesk app tool.
Sometimes it can also be useful to provide an example in the AI Agent prompt instructions detailing which data to use for which “parameters” (i.e. inputs) for the App Action tool, as well as the data type for the parameter.
Use the Create Ticket Helpdesk app tool with the following inputs
- title: [User’s question] (string)
- description: [Description of user’s issue] (string)
- reporter_email: [collected email] (string)