# Live Overview, plus Last9 AI knowledge and custom MCPs

> The home page shows firing alerts, degraded services, unhealthy workloads, and exceptions. Add knowledge and MCP servers to Last9 AI from AI Settings.

Source: https://last9.io/changelog/live-overview-plus-last9-ai-knowledge-and-custom-mcps/

### A live Overview

To see what needs attention, you opened the Alert Monitor, Discover Services, Discover Kubernetes, and Discover Exceptions one by one. The home page is now an **Overview** of what is firing, what got worse, and what changed, in one place. One **1 h**, **6 h**, or **24 h** window applies to every block; **6 h** is the default.

- **Counters**: **Firing Alerts**, **Degraded Services**, **Unhealthy Workloads**, and **Spiking Exceptions**, each with a sparkline over the window and a link to the list it counts.
- **Firing Alerts**: Rules firing now in the Alert Monitor, named by alert group and rule. **Ack** an alert from the row. With no alert rules, the block offers **Create Alert Rule**.
- **Degraded Services**: Services whose Apdex dropped or P95 latency rose by 10% or more against the previous window of the same length, from [Discover Services](/docs/discover-services/). Each row shows the service's latest change event and its top exception.
- **Unhealthy Workloads**: Deployments, StatefulSets, and pods that [Discover Kubernetes](/docs/discover-kubernetes/) rates critical or warning, with the reason and when each became unhealthy. The block shows only when Last9 receives Kubernetes data.
- **What Changed**: Recent [change events](/docs/change-events/), linked to their service when they have one.
- **Exceptions**: Exceptions from [Discover Exceptions](/docs/discover-exceptions/), with ignored exceptions hidden.
- **Ask Last9 AI**: A question typed in the header opens a new AI Assistant chat.

Pinned dashboards stay below the Overview.

### Knowledge and MCP Servers in AI Settings

To give Last9 AI your runbooks or connect it to your own MCP servers, you had to call the API. **AI Settings** now has **Knowledge** and **MCP Servers** tabs.

- **Knowledge**: Add a topic, then add documents to it as text or as an uploaded file. Turn on **Use in incident triage** for a topic, and Last9 AI uses it when it investigates an incident. Under **MCP tool bindings**, pick a server and its tools for the topic, and set each binding to **Require approval** or **Auto-run**. See [Add knowledge to Last9 AI](/docs/ai/knowledge/).
- **MCP Servers**: **Add MCP Server** takes a name, a URL, the transport (**Streamable HTTP** or **SSE**), and optional headers for authentication. **Test Connection** checks that Last9 can reach the server and list its tools. See [Add remote MCP servers to Last9 AI](/docs/ai/remote-mcp-servers/).

### Improvements

- **Discover Services**: Added an **Apdex** column to the services list.
- **Logs**: Added **TraceId** under **Primary Fields** in the logs filter dropdown. It filters on the log's trace ID column and is case-sensitive.
- **Alert Monitor**: [Acknowledging alerts](/docs/acknowledge-alerts/) sends an email on each ownership change: to you on **Acknowledge**, to the new owner on **Reassign**, and to the previous owner on **Unacknowledge**. The email names the alert rule or alert group.
- **Agent Observability**: The **Overview** loads each card on its own, so one slow or failed query no longer blanks the page.
- **Time range picker**: A preset such as **Last 15 minutes** stays a sliding window after a page refresh. A custom range or a chart zoom stays fixed. Applies to Metrics, Traces, Logs, Discover, Exceptions, RUM, Profiles, Service Map, Alert Monitor, Synthetics, and Audit Trail. Links you already shared with fixed `from` and `to` stay fixed. See [Time range picker](/docs/time-range-picker/).

### Fixes

- **Agent Observability**: The **Conversations**, **Avg messages/conversation**, and **Conversation Cost** cards over-counted long conversations, for example 45 conversations shown over two days where the **Conversations** tab listed 11. They now count from traces, and cost uses the cost each model reports, with a token-based estimate when a model reports none. See [Agents Monitoring](/docs/agents-monitoring/).
- **Control Plane**: Typing a trailing space in a drop rule value hid the **Create** action, so exact-match values with spaces could not be entered. Drop rules now keep leading and trailing spaces, including values made only of spaces. See [Drop rules](/docs/control-plane-drop/).