Monty — AI Assistant

Overview#

Monty is the AI assistant built into the WhatsApp/SMS messaging panel. It reads every incoming customer message, tags it by type, and — for planning messages — prepares a reply draft and, where relevant, a booking action for you to review. Monty is designed around one hard rule: it never sends a message and never changes a booking on its own. Every outgoing message goes through your send button; every booking change goes through an explicit Accept click.

Monty is optional and per-tenant: without an API key, the messaging panel works exactly as it always has.

Getting started#

Monty runs on your own Anthropic API key, so the usage costs are billed directly to your own Anthropic account — Quick Release adds no margin and has no access to your key.

1. Create an API key at console.anthropic.com (Settings → API keys) and make sure the account has credits.
2. In QR, open Settings → Monty and paste the key. It is validated immediately and stored securely — it is never shown again (replace-only).
3. That’s it. From this moment, new incoming customer messages are tagged automatically. There is nothing else to configure.

The status line shows where you stand:

When Monty is paused, nothing breaks: messages simply arrive without tags, and no errors appear in the panel.

Message tags#

Every incoming customer message in a booking conversation gets one colored letter, visible on the message bubble and on the conversation row in the inbox list:

A few things to know:

• The conversation row shows the tag of the newest incoming message; an orange dot means a Monty proposal is waiting for you.
• When Monty is not confident about a message, it stays untagged rather than guessing.
• Crew conversations are never tagged.
• An emoji reaction (👍 on your date proposal) is understood as a planning response.

Reply proposals#

For messages tagged P, Monty drafts a reply in the customer’s language (Dutch, French, English, German or Spanish), based on the booking, the flight, the location and your instructions (see below).

• When you open the conversation with an empty composer, the draft is pre-filled under a marked “proposal from Monty” strip. Edit freely — the send button sends exactly what is in the composer.
• If you were already typing, your text is never touched; a small “proposal available — insert” link appears instead.
• The ✕ on the strip discards the proposal; replying manually makes it disappear silently.
• There is always at most one open proposal per conversation — a newer planning message replaces the older proposal.

Booking actions#

When a planning message implies a booking change, an action card appears above the composer, for example: Confirm booking on flight Tue 21 Jul — evening (3 p.).

• Accept executes the change under your own user account, with the same capacity and status checks as a manual change. A confirmation line appears in the thread.
• Click the flight name to pick a different flight first — the card lists valid alternatives with free seats.
• If the flight filled up or was cancelled in the meantime, nothing changes and the card explains why, offering the alternatives.
• The ✕ dismisses the action without touching the booking. The reply draft and the action are independent — you can use either one, both, or neither.

Instructions per category#

Under Settings → Monty → Instructions you can write free-text instructions per category (P, Q, W, C): your tone, whether you address customers formally or informally, standard meeting-point details, refund or voucher policies, and so on. Monty consults these every time it drafts a proposal for that category.

Learn from history#

Instead of starting from a blank page, Monty can learn your answering style from your own message history:

1. Press Learn from history in the Monty settings.
2. Pick a period — 1, 3, 6 or 12 months. The dialog shows how many messages that period contains and an estimated cost in € before anything runs, with an explicit warning that the run uses your own Anthropic credits.
3. Confirm. The run works in the background — you can follow the progress (and cancel) from the settings card, and keep using the panel normally.

The run tags your historical messages (older conversations light up with tags too) and then distills, per category, how you typically answer — delivered as a clearly marked, editable draft under each category’s instructions, in your own UI language.

What to do with a learned draft: Monty already uses the draft together with your own instructions as soon as it appears — you don’t have to do anything. But a future learning run will replace the draft. So:

• Adopt into my instructions — appends the draft to your own instruction text, making it permanent. Recommended once you have reviewed and tweaked it.
• Leave it — it stays active, but the next learning run overwrites it.
• Discard — removes it; Monty then uses only your manual instructions.

Messages that already carry a tag are never classified again — re-running learning after an interruption (for example when credits ran out mid-run) continues where it left off instead of paying twice.

Usage & costs#

The settings section shows per month: the number of classifications, proposals and learning runs, and an estimated cost in €. The estimate is computed from current Anthropic list prices and marked ± — the actual invoice comes from your own Anthropic account. As a rule of thumb, classification plus proposals cost well under €0.05 per handled conversation; a learning run costs roughly €1 per 1,000 historical messages.

Privacy & safety#

• All processing happens per tenant: your messages, instructions, usage and audit records are never visible to or used for any other tenant.
• Monty never auto-sends and never auto-changes bookings — every proposal resolution (sent, edited, refused, dismissed, expired) is recorded in an audit trail with who did what, when.
• Removing your key stops Monty immediately; existing tags remain visible.