Analytics

Summary KPIs and the five charts

The Analytics page (sidebar → Analytics) provides detailed performance data for your AI assistant. Each chart has its own independent time range selector — you can compare different periods side by side.

Summary cards (top of page)

Card	What it shows
Conversations	Total sessions in the active chart period
Messages	Total messages in the active chart period
Feedback	Total feedback items plus the positive percentage
Ø Response time	Average LLM response time in seconds across all requests

These numbers are derived from the already-loaded chart data — no extra network requests needed.

The five charts

Chart	Description
Feedback	Daily positive and negative feedback counts
Messages	Daily conversation and message volume
Response time (Antwortzeit)	Average LLM latency per day in seconds
Token consumption (Token-Verbrauch)	Input and output tokens consumed per day
Latency breakdown (Latenz-Breakdown)	Average duration per span type (e.g. retrieval, generation). Clicking a bar in "per chat" granularity opens that specific session.

Time range selector

Each chart has a small selector in its top-right corner. You can choose:

• Presets: last 7 days, 30 days, 90 days, 6 months, 1 year, all time
• Custom range: pick any start and end date

Your selection is saved to local storage and restored the next time you open the page.

Exporting analytics data to Excel

You can download a complete analytics report as an .xlsx file.

Steps

1. Open the Analytics page.
2. Click the Excel exportieren (Export to Excel) button in the top-right corner of the page header.
3. A dropdown menu appears — choose the time period you want to export:
   • Last 7 days
   • Last 30 days
   • Last 90 days
   • Last 6 months
   • Last year
   • All time
4. The file downloads automatically to your browser's default downloads folder.

The button shows "Exportiert…" while the file is being generated. If the export fails an alert is displayed — try again in a moment.

Latency breakdown and drilling into sessions

The Latency Breakdown (Latenz-Breakdown) chart at the bottom of the Analytics page visualises where time is spent during each AI response, broken down by span type (e.g. retrieval, LLM generation, embedding, tool calls).

Granularity options

The time range selector for this chart includes an extra granularity dropdown:

Granularity	What each bar represents
Hour	Average span duration per hour
Day	Average span duration per day
Week	Average span duration per week
Chat (per session)	One bar per individual chat session

Drilling into a session

When the granularity is set to Chat, clicking a bar navigates directly to that specific chat session (/chat/<session_id>). This lets you investigate a single slow response in detail.

Use the Chat granularity to identify outlier sessions that had unusually high latency, then review the conversation to understand why.