Analytics
The Analytics dashboard provides a comprehensive overview of assistant usage, costs, and API activity. Navigate to Analytics in the sidebar to access it.
Filters
All charts and metrics respond to three filters at the top of the page:
| Filter | Options | Description |
|---|---|---|
| Assistant | Dropdown of active assistants | Scope every metric to a single assistant |
| Model | Dropdown of models that have been used | Scope LLM-related metrics (costs, tokens, requests) to a specific model |
| Date range | 7, 30, or 90 days | Control the time window for all data |
Filters combine — selecting both an assistant and a model shows only activity for that assistant using that model.
Summary Cards
Six cards across the top give an at-a-glance overview:
- Conversations — Total conversation count in the period
- Messages — Total messages exchanged
- API Requests — Number of LLM API calls made
- Tokens — Total tokens consumed (input + output)
- Content Items — Number of generated content pieces (blog posts, summaries, etc.)
- Total Cost — Aggregate LLM spend in USD
Cost Analysis
Cost per Assistant
A horizontal bar chart ranking assistants by total spend. Hover for exact cost and token counts.
Top Spenders
A table showing the 15 highest-cost people — both contacts (external users interacting via channels) and staff users (internal team members). Each row shows the person’s name, type, conversation count, and total cost.
Cost Over Time
A stacked bar chart showing daily cost broken down by assistant. Useful for spotting cost spikes and understanding which assistants drive spend on a given day.
API Usage
Daily API Requests
Bar chart of LLM API call volume per day.
Daily Tokens
Bar chart of total token consumption per day.
Usage by Model
A table breaking down usage per model with columns for requests, input tokens, output tokens, total tokens, and total cost. Helpful for comparing model efficiency and cost.
Sandbox Containers
Shows the number of currently active sandbox containers. Each conversation that uses code execution gets its own container, which is automatically cleaned up after 10 minutes of inactivity.
Conversation Insights
Conversations Over Time
A stacked bar chart showing daily conversation volume broken down by assistant. Helps identify usage trends and peak activity periods.
Average Conversation Length
A horizontal bar chart showing the average number of messages per conversation for each assistant. Longer conversations may indicate more complex tasks or opportunities for prompt improvement.
Channel Breakdown
A doughnut chart showing how conversations are distributed across channels:
- Web — Direct browser-based conversations
- Slack — Conversations via the Slack integration
- Email — Conversations via the Postmark email integration
Tool & Content Metrics
Most Used Tools
A horizontal bar chart showing which tools are called most frequently, grouped by assistant. Includes built-in tools like search_knowledge and browser_action as well as any custom tools.
Content Output
A stacked bar chart showing the volume of generated content over time, broken down by content type (blog posts, summaries, etc.).
Guardrail Metrics
When guardrails are enabled on any assistant, the analytics API includes additional metrics:
Pass Rates by Type
Shows the total evaluations and pass rate for each guardrail type (faithfulness, instruction adherence) across the selected time period. Helps identify which assistants are producing grounded, on-brand responses and which need prompt tuning.
Fire Rate Over Time
A time series showing how often each guardrail type fires per day. A rising trend in knowledge grounding nudges may indicate the assistant’s personality prompt needs stronger instructions. A falling faithfulness pass rate may indicate knowledge base gaps.
Evaluation Cost
Total cost of guardrail evaluation LLM calls in the period. At default sample rates with Claude Haiku, this is typically a small fraction of overall LLM spend.