Workspace admins can create and configure AI Agents from the Admin → Agents page. Each agent has its own name, personality, tools, and permissions.
Creating a New Agent
Agent Settings
Name
A descriptive name for the agent (max 256 characters). This is displayed in the chat interface, ticket list, and Slack messages.
Avatar
Upload a profile picture (PNG, JPEG, or WebP, max 10 MB). The avatar appears in chat messages and Slack replies.
Role / System Prompt
The system prompt defines the agent's personality, expertise, and behavior guidelines (max 10,000 characters). This is the most important configuration — it tells the agent how to approach tasks, what domain knowledge to apply, and any rules to follow.
Example system prompt:
You are a data analyst assistant for the marketing team. When asked about
campaign performance, always query the `analytics.campaigns` table first.
Present results as charts when possible. Keep explanations concise and
business-focused.
Status
Default Agent
Toggle to make this the workspace's default agent. The default agent is automatically selected when users start a new chat session. Only one agent can be the default.
Permissions
Permissions control which tools and capabilities the agent can use. They are organized into groups:
Assigning Tools
Agents can execute blocks (SQL, Email, Slack, Chart, Tableau, Google Slides) that have been shared with them. To give an agent access to a block:
- Open the block (notebook task) you want to share
- Go to Sharing & Permissions
- Add the agent as a viewer or editor
The agent will then be able to run that block as a tool during conversations. The agent list page shows icons for each tool type assigned to the agent.
Sharing & Access
Agent access is controlled through the standard PushMetrics sharing model:
- The creator gets edit permission by default
- You can share agents with specific users or roles
- The default agent is shared with the Everyone role so all workspace users can chat with it
Agent Runs
The Runs tab on an agent's detail page shows the last 50 execution runs linked to that agent, including:
- Run status (success, failed, running)
- Session/ticket title
- Start time and duration
Use this to audit agent behavior and troubleshoot failed sessions.