Analytics

Analytics brings the organization’s operational metrics together at /admin/analytics. Access to Overview, Agents, Users, Models, Credits, and Support requires the analytics.view permission. Logs uses the independent audit_log.view permission and can also be opened directly at /admin/analytics/logs.

The Analytics item may disappear from the sidebar while the organization has no messages. This does not change route permissions: the menu is only a navigation hint, while direct access remains protected by the permissions above.

Understand the time ranges before comparing numbers

The tabs do not share one global filter:

• the four Overview cards, the Agents and Users rankings, and Models data are cumulative, with no 30-day cutoff;
• on Overview, the Consumption Mix card and the Daily Rhythm chart cover the last 30 days;
• Credits has its own agent, channel, and date filters;
• Support has its own 7-, 30-, or 90-day selector plus a custom range;
• Logs starts at 7 days and offers 24 hours, 7 days, or 30 days.

Do not compare two cards only because they appear on the same page. First confirm the period and population represented by each one.

The seven tabs

• Overview: cumulative totals, the consumption mix for the last 30 days, and a chart that toggles between conversations and credits over that same period.
• Agents: up to 10 non-archived agents, ordered by cumulative conversation count.
• Users: up to 10 non-removed members, ordered by cumulative conversation count.
• Models: AI responses, tokens, and credits grouped by the recorded model.
• Credits: consumption for the period broken down by usage type — agent replies, meeting transcription, site indexing, automation, and the rest.
• Support: queue now, volume, resolution, first response, satisfaction, inboxes, attendants, topics, and contact tags.
• Logs: audit trail for actions involving agents and knowledge bases.

Overview

The cards show:

• Total Conversations: every conversation in the organization; the caption says how many currently have active status.
• Total Users: organization memberships that have not been removed. The current count also includes members without an active plan seat; confirm seats under Settings → Users.
• Credits Consumed: the cumulative sum of credits_used on conversations, not only the current month.
• Active Agents: agents with at least one conversation among the 10 agents returned by the ranking. This card can undercount organizations with more than 10 used agents and should not be treated as a complete inventory.

Two blocks on this tab use the last 30 days rather than cumulative data:

• Consumption Mix names the three entities that spent the most in the period and groups the rest under Other types, with each share as a percentage. The See detail in Credits link opens the Credits tab.
• Daily Rhythm carries the Conversations / Credits toggle. The period is the same for both; only the unit being measured changes. Under Conversations, the chart counts by UTC creation date and fills days without new conversations with zero.

Agents

The list shows up to 10 non-archived agents, ordered by cumulative conversation count. If the organization has fewer than 10 agents, agents with zero conversations may appear. Each row’s bar is proportional to the first-ranked agent.

Select Details to open /admin/analytics/agents/:agentId. The detail page starts at 30 days and lets you select 7, 30, or 90 days or a custom range. It includes:

• conversation, message, credit, and tool-execution cards;
• daily activity;
• tool usage and errors;
• model and channel distributions;
• satisfaction average, coverage, distribution, and low-satisfaction conversations.

An archived agent can still be opened by URL and receives an Archived badge on the detail page, although it is absent from the main ranking.

Users

This tab shows up to 10 non-removed members, ordered by cumulative conversation count. Each row includes name, email, conversations, and credits associated with conversations whose user_id belongs to that member. If there are fewer than 10 members, people with no conversations may also appear.

Because the current query does not exclude plan_restricted_at, a member without an active plan seat may remain in this list. Use Settings → Users to verify who actually occupies an active seat.

Models

This tab considers only assistant messages with a recorded model. Its data is cumulative and does not follow the Overview chart’s 30-day window.

It contains:

• Total Messages, Models Used, and Most Used Model;
• Distribution by Model, showing each model’s percentage;
• Usage by Model, showing response counts;
• a Model, Messages, Tokens (In/Out), Credits, and Credits per Message table.

Messages without a model identifier are excluded from this distribution.

The tab opens with the notice This tab covers only what goes through a chat model. Meeting transcription, site indexing, automation, and organization research consume credits without going through a chat model, so they never appear here. When the period contains that kind of consumption, a footer states how much was left out and links to the Credits tab.

Credits

The Agent, Usage type, Channel, Start Date, and End Date filters update the tab immediately. The initial range runs from the date 30 days ago through today, inclusive. There is no Apply button.

Usage type separates what generated the credit: Agent replies, Agent-desk messages, Meeting transcription, Meeting minutes, Site indexing, Email, Organization research, Automation, Satisfaction analysis, Model test, and Other. All types keeps everything in the result.

The channel selector offers Web, Telegram, WhatsApp, and Public. All channels keeps other channels in the result, but the current selector cannot isolate them. Entries without a conversation are grouped as Web by the query.

The tab shows:

• Total Credits: the absolute sum of credit_usage entries in the filter;
• Billed Messages: agent replies billed in the period. It counts distinct messages, not ledger entries — transcription, indexing, and automation stay out of this number;
• Average per Day: total credits divided by the calendar days in the range, including days with no consumption;
• Biggest Consumer: the usage type that spent the most in the period, with its total;
• Where Credits Went: one bar per usage type, from largest to smallest;
• Usage per Day: credits by date. Hover a day to see that day’s composition by type;
• Top Agents by Consumption: up to 10 agents associated with entries, including messages, credits, and main channel. Main channel means the channel with the most entries, not necessarily the highest cost.

The agent table ends with a No agent assigned row whenever the period has ownerless consumption — meeting transcription, site indexing, and model tests belong to no agent. That row is what makes the credits column add up to the Total Credits card.

Support

The Support tab shows how customer support went in the selected period and how the queue looks right now. Only customer conversations count; the team’s internal conversations are left out. A conversation counts when the customer sent a message within the period.

• Queue now ignores the selected period and counts conversations still waiting for a person’s first response. At risk, Overdue, and Unassigned are subsets of Total waiting. An item becomes at risk when up to 25% of the deadline remains, with a minimum 5-minute window. With permission to view conversations, each card opens the conversation list already filtered to that subset.
• Volume: Conversations, With AI (an agent replied in the period), With a person (a team member sent a message in the period; assigning the conversation without replying does not count), and AI only. The same conversation can count in both With AI and With a person.
• Resolution: conversations Resolved in the period and how many were Resolved by AI.
• First response: Within SLA, Response time (p50), Response time (p95), and Violations, over first human responses completed in the period. p50 is the time within which half of the responses were sent; p95, the time within which 95% were sent. Times count only the inbox schedule’s service hours.
• Satisfaction: how many conversations were rated in the period and the average score, from 1 to 5.
• By inbox: conversations, with AI, with a person, SLA compliance, and p50 for each inbox.
• Performance by attendant sorts people with the most violations first. The table shows responded, on target, violations, compliance, and time (p50); it also flags former members and responses without a recognized attendant (Unidentified).
• Topics lists the 15 most frequent topics in the period’s conversations. Contact tags shows tags given in the period and the total contacts with each tag.

Use 7, 30, 90 days, or Custom, up to 400 days. A custom range only applies after you choose both dates and click Apply. The same numbers are available through the API in Support metrics.

Logs

Logs requires audit_log.view, independently of analytics.view. A person who only has this permission opens /admin/analytics/logs directly and sees only the Logs tab.

Available filters are:

• Period: last 24 hours, 7 days, or 30 days;
• User: the action’s author;
• Action: creation, update, removal, and relationship changes for agents, tools, categories, and bases;
• Search by name: searches only the entity name, not the complete change text.

The table shows When, User, Action, and Entity in batches of 50 records. Select a row to view its author, timestamp, before and after values, and available metadata. Load more currently replaces the visible batch with the next one instead of appending it; record any details you need before moving forward.

How to investigate a discrepancy

1. Identify whether the question concerns cumulative data, the Credits range, the Support range, or Overview’s last 30 days.
2. For costs, start with the Biggest Consumer card and the Where Credits Went block: they name the usage type that drove the period. Then filter by Usage type, date, and agent.
3. For one agent’s performance, open Agents → Details and select the appropriate range.
4. For human-response delays, compare Queue now with the First response section of the Support tab; changing the period does not change the queue.
5. To discover who changed an agent or a base, use Logs and open the row details.