Monitoring

Once a project is live, these screens show what it's doing and whether it's healthy.

Analytics

How much the agent is being used and what it costs. Headline totals (AI calls, tokens, estimated cost), cost by model, usage by intent and mode, a daily cost trend, monthly plan-limit usage, and voice-minute usage. You can set a custom price for any model that's missing one so cost estimates stay accurate.

Threads

An audit view of every conversation the project's agent has had — a table of conversations with the end user, message count, tokens, and last activity. Open any conversation for its full transcript: the messages in order, the recognised intent, tokens, and any tools called, with problematic turns highlighted. A conversation can be deleted from the history.

Tool Health

An operational dashboard for the reliability of the tools the agent calls: active alerts, reliability indicators, recent "bad turns" (conversations where a tool call went wrong), and per-tool health. Choose the time window; the whole dashboard follows the selection, and alerts can be resolved once handled.

Model Pricing

Defines the per-token rates used to estimate what the project's AI usage costs, so the figures shown across Analytics and Overview are accurate for the models the project uses. Set custom rates for specific provider/model combinations on top of the curated defaults; cost resolves as override → default → unpriced. The same "set price" action is available from the Analytics cost breakdown for any model missing a rate.

Compliance Incidents

Content that the compliance scanner flagged as potentially containing regulated or sensitive data, drawn from knowledge uploads, tool definitions, and sampled live traffic. Only redacted snippets are shown. Reviewers filter by status and triage each incident (clear or escalate), and can jump to the source conversation for context.