Skip to main content

Overview

Beak AI Runtime is the hosting layer for deployed AI agents. It lets your team run, interact with, and manage CopilotKit-based agents from a single web workspace. The Dashboard at / is the first screen administrators see after login. It combines runtime-level KPI tiles, a conversation usage trend chart, a recent-activity feed, and quick-action shortcuts into one continuously refreshed view. Data refreshes automatically every 30 seconds so the page can be kept open as a live ops board.

The sidebar groups all navigation into four sections:

Every entry is gated on its own permission rather than on a blanket admin flag, and a section appears exactly when the signed-in user can reach something inside it — so two operators can see different sidebars. The sidebar collapses to icon-only mode and expands again on hover in the Linear layout.

Runtime Hero​

The hero banner at the top of the dashboard identifies the module as AI Agents · Overview and shows the title "Run and chat with live agents." An Open chat button navigates directly to the /agents grid so administrators can start a conversation in one click.

KPI Row​

Five stat cards sit below the hero and summarise the runtime at a glance. All counts are live-queried and refresh every 30 seconds alongside the page data.

  • Active Agents — count of published, active agents returned by the agent API; badged Live
  • Conversations — total thread count aggregated across every active agent
  • Knowledge — distinct datasource count deduplicated across all agents
  • Skills — distinct skill count deduplicated across all agents
  • Credentials — number of stored credentials available for skill and datasource connections

Usage Trend Chart​

An area chart plots conversation volume over time. The time range is toggled with the 7d, 30d, and 90d buttons. Ranges beyond 60 days are bucketed into weekly totals to keep the chart readable. Hovering a data point shows the date label and exact conversation count for that period.

Recent Activity​

A card alongside the trend chart lists the five most recently updated conversation threads across all agents. Each row shows the thread name, the agent it belongs to, and a human-readable timestamp (e.g. 2h ago). Clicking any row opens that specific thread in the chat view at /chat/:agentId/:threadId.

Recent Agents​

A table below the trend row lists the first four active agents ordered by the API response. Each row shows the agent icon, name, type badge, and description. Clicking a row navigates to /chat/:agentId. A View all link in the header navigates to the full /agents grid.

Quick Actions​

Three shortcut buttons provide fast navigation without using the sidebar:

  • Start New Chat — navigates to /agents to pick an agent
  • View All Agents — navigates to /agents
  • Configure Agent — navigates to /configure/:agentId for the first active agent, or falls back to /agents when no agents exist

Best Practices​

  • Pin the Dashboard in a dedicated browser tab during business hours — the 30-second refresh keeps conversation and agent counts current without manual reloads.
  • Use the View all link from the Recent Agents panel rather than the sidebar when you need to open an agent quickly; it loads faster than navigating away from the dashboard.
  • If the Active Agents tile reads zero, verify that agents are published and marked is_active in the agent builder before checking network connectivity to the Runtime backend.
  • Watch the Usage Trend chart after releasing a new agent — a flat line in the first 30-day window indicates users have not discovered it yet; check channel configuration on the Configure page.
  • Credential count should match the number of skill and datasource connections that require authentication; a discrepancy signals missing credentials in Connections.
  • Agents — browse the agent grid and start conversations
  • Configure — manage skills, knowledge, MCP servers, channels, and notifications per agent
  • Suites — install curated bundles of agents and supporting resources
  • Approvals — review and act on pending human-in-the-loop approval requests
  • Connections — manage stored credentials used by skills and datasources
  • Providers — configure LLM provider credentials
  • Settings — tenant-level settings and user management