> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.revvue.ai/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.revvue.ai/_mcp/server.

# Jonna (AI agent)

Jonna is RevVue's AI agent. It reads incoming guest messages in the [Inbox](/inbox), works out what's being asked, pulls the facts from the [Knowledge hub](/knowledge-hub), and replies in your voice — handling routine requests end-to-end and escalating to a human the moment something falls outside what it can safely do. The [Jonna API](/api-reference/jonna) exposes configuration, conversations, and statistics.

## Assistants

An **assistant** is a configured instance of the agent for your tenant: which tools it may use (replying to guests, reading the wiki, tagging tickets, …), how autonomous each tool is, and when it must hand over to a human. Guardrails are per-tool: a tool call can be allowed to run autonomously, judged by an LLM first, held for human approval, offered to a human as a draft to send themselves, or denied outright.

Manage assistants with the `jonnaAssistant` query and mutation, and remove one with `jonnaDeleteAssistant`. The design principle baked into the defaults: **a guest is never answered by an agent that can't actually do what they asked** — requests the assistant has no tool for are routed to your team instead.

## Conversations

Every run of the agent happens on a thread. `jonnaThreads` lists them and `jonnaThreadMessages` returns the full exchange — the guest's messages, the agent's replies, and the tool calls in between — which is what you'd render for a "show me what the agent did here" view.

## Statistics

Each guest request the agent engages becomes a **case**, and the statistics answer the question that matters: how much is Jonna actually handling?

* `agentCases` — the individual cases, with outcomes.
* `agentCaseFunnel` — the funnel from engaged → resolved autonomously → escalated to a human.
* `agentTagFacets` / `agentTagActionStats` — volume and outcomes broken down by [super tag](/inbox#super-tags), so you can see *what kinds* of requests the agent resolves versus hands over.
* `agentEscalationCategories` — why runs get escalated.
* `queryAgentAnalyticsBatch` — batch analytics for dashboards, same shape as the inbox's analytics batch.
* `jonnaCurrentMonthCredits` — this month's usage.

## What the agent knows

Jonna's answers are only as good as the [Knowledge hub](/knowledge-hub) — the per-tenant wiki of pages, scoped per location and brand, that the agent reads before it commits to an answer. If the agent keeps escalating a certain question, the fix is usually a wiki page, not a config change.

Full operation reference: [Jonna API](/api-reference/jonna).