Using Ollive
Tools.
What your assistant can reach.
When you ask a question, your assistant chooses the Ollive tools it needs and reads what they return. You never call a tool yourself, but knowing what each one does helps you ask for exactly what you want.
Words you’ll see
| Deployment | An agent Ollive monitors. |
|---|---|
| Session | One conversation between your agent and a person, such as a call or a chat. |
| Rule | Something your agent must or must not do, drawn from a law, a regulation or a contract. |
| Verdict | The result of checking one rule in one conversation: pass, fail or indeterminate. |
| Violation | One agent breaking one rule, across one or more conversations. |
| Pattern | What the conversations behind a violation have in common. |
All tools
query_deploymentsReadsList deployments
The agents Ollive monitors in your workspace, active or paused. Your assistant usually starts here, to find the agent you mean.
Try asking
- “Which agents does Ollive monitor?”
- “Is the billing agent still being monitored?”
query_metricsReadsCount sessions or violations
Totals, rankings and trends over a period of time: how many conversations were checked, how many violations there were, and which agents, rules or severities account for most of them.
Try asking
- “How many conversations did Ollive check this week?”
- “Which five rules were broken most often in the last 30 days?”
- “Show high-severity violations per day for the last month.”
query_violationsReadsFind violations
Where agents broke rules, with each violation’s severity, the clauses it comes from and how many conversations it appeared in. From a violation, your assistant can drill into the patterns behind it and the conversations themselves.
Try asking
- “What critical or high-severity violations does the booking agent have?”
- “What do the conversations behind that violation have in common?”
query_sessionsReadsList sessions
Recent conversations, newest first, with how many rules each one passed and failed, and whether checking has finished.
Try asking
- “Which conversations failed in the last 24 hours?”
- “Has Ollive finished checking the conversations I sent?”
query_verdictsReadsGet a session’s verdicts
Every rule checked in one conversation and its result: pass, fail or indeterminate. A failure comes with the reasoning, the facts that decided it and the clause it comes from.
Try asking
- “Why did conversation call-2026-10-14-0912 fail?”
- “Which rules did that conversation pass?”
query_transcriptsReadsGet a session’s transcript
What was said in one conversation, turn by turn, and the steps the agent took, such as looking up a record.
Try asking
- “Show me the transcript of call-2026-10-14-0912.”
- “Quote the turn where the agent introduced itself.”
create_sessionsSendsSend sessions to be judged
Sends conversations you give it to Ollive, up to 10 at a time, to be checked against an agent’s rules. Their results show up in the other tools once checking finishes.
Try asking
- “Send these three test conversations to Ollive for the booking agent.”
Your assistant sends only the conversations you give it, never ones it writes itself. A conversation can be sent once, and can’t be changed afterwards. Conversations sent this way are labelled MCP, and kept apart from the ones Ollive collects from your agent.
Same results as Send sessions in the APIGood to know
- Answers cover a period of time. Say which one, like “the last 7 days” or “since October 1”. If you don’t, most tools look at the last 30 days.
- Counts and violations cover the conversations Ollive collects from your agents. To ask about conversations you sent yourself, say so.
- The six reading tools are marked read-only, so you can let your app run them without asking. Sending conversations is marked as a change, so most apps ask you first.
- If nothing matches, your assistant gets an empty answer, not an error. Try a longer period of time or another agent.