Example workflows.
From a question to the evidence.

Run a weekly review

Every Monday, before the team meeting.

1. AskGive me a liability summary of all my agents for the last 7 days: conversations checked, violations by severity, and the three rules broken most often. Compare it with the week before.
   Usesquery_deploymentsquery_metrics

   Totals, a severity breakdown and the top rules, set against the previous week.

Tip Ask for it formatted for Slack or email, and paste it straight into your team’s channel.

Investigate a violation

When a violation shows up and you need to know why.

1. AskWhat high-severity violations did the appointment booking agent have this month?
   Usesquery_deploymentsquery_violations

   Each violation, with the rule, where it comes from and how many conversations it appeared in.

2. AskWhat do the conversations behind the top one have in common?
   Usesquery_violations

   The patterns Ollive found, such as the agent greeting callers with a human name.

3. AskOpen the most recent one and quote the turn where it went wrong.
   Usesquery_violationsquery_transcripts

   The agent’s exact words, next to the rule they broke.

Explain one conversation

When a complaint or an escalation points to a single call or chat.

1. AskWhy did conversation call-2026-10-14-0912 fail? Explain each failed rule in plain words, with the evidence.
   Usesquery_verdictsquery_transcripts

   Each failed rule, the reasoning behind it, and the words that decided it.

Check a change before it ships

While you work on your agent’s prompt or code, in Cursor, Claude Code or Codex.

1. AskI changed the agent’s greeting. Send these three test conversations to Ollive for the appointment booking agent.
   Usesquery_deploymentscreate_sessions

   The conversations are queued to be checked, apart from your production results.

2. AskAre the results in? Did any rule fail?
   Usesquery_sessionsquery_verdicts

   Each test conversation’s verdicts, once checking finishes.

Tip Checking takes a little while. If the results aren’t in yet, ask again in a few minutes.

Track a trend

After a release, to see whether things got better.

1. AskShow violations per day for the appointment booking agent over the last 30 days. Did anything change after October 10?
   Usesquery_metrics

   A day-by-day count and what changed. Many assistants can draw it as a chart.

Get better answers

• Name the agent the way it appears in Ollive.
• Give a period of time, like “the last 7 days” or “since October 1”. Without one, most answers cover the last 30 days.
• Ask for the evidence: quotes, clauses and conversation IDs.
• Follow up instead of starting over. Your assistant keeps the IDs from earlier answers, so it can go straight to the detail.