BeforeQuery closes the loop: connect sources → answer questions → measure → improve content → better answers. Everything below lives in the dashboard under Project → Analytics and Workspace → Analytics.

​
Watch conversation volume and feedback

Open Project → Analytics to see conversations, messages, and thumbs up/down feedback over time (up to 90 days). Use it to spot spikes after launches, quiet surfaces that need promotion, and the overall helpful/unhelpful ratio.

​
Find and fix knowledge gaps

The Knowledge Gaps view lists questions the assistant couldn’t answer (no citations found) or answered badly (negative feedback) — sorted so it reads like a prioritized writing backlog, with token/cost usage alongside.
1

Review the gap list

Open Project → Analytics → Gaps. Each entry shows the unanswered question and how often it was asked.
2

Generate doc drafts automatically (optional)

Run a doc-proposal scan from the gaps view — an LLM drafts documents for unanswered questions.
3

Accept or dismiss proposals

Review each draft: accept to ingest it straight into the knowledge base, or dismiss it. Accepted drafts start answering immediately.

​
See which sources earn their keep

Project → Analytics → Sources shows, per source: document count, chunk count, last synced time, and — most usefully — citation count: how often that source actually powered an answer.
Sources that never get cited are candidates for pruning — they add index noise and sync cost without improving answers.

​
Workspace-wide overview

Workspace → Analytics rolls everything up: questions per week (12-week trend), top sources, top user languages, totals, and a gaps summary across all projects — the view to share with leadership.

​
Trust the answers: LLM-judge evals

Sampled answers are automatically scored, no setup required. Find scores under Project → Analytics → Evals: Watch the daily series after big content changes — a groundedness dip or a rise in hallucination flags means retrieval quality regressed and a source needs attention.

​
Track billable outcomes

Workspace → Usage meters the outcomes that matter: resolved conversations and deflected tickets, with daily series. Use it for ROI reporting and to anticipate plan limits — see Billing & Plans.

​
Trace every AI interaction

Every chat, search, MCP call, A2A exchange, ticket deflection, and agent run is recorded as a trace. Open Project → Traces, filter by kind, and drill into any trace to see exactly what was retrieved and answered — invaluable when debugging a bad answer someone reported.
Self-hosted and enterprise deployments can additionally export OTLP-shaped spans to an OpenTelemetry endpoint (otel.enabled + otel.endpoint in server config).