Assistant

The Revvly Assistant.

An AI assistant built into your dashboard. Ask it about your subscription business in plain language, and have it set up clients, catalog, and subscriptions for you — you review and confirm irreversible or high-impact changes before they happen.

01What it is

The Assistant lives in the dashboard as a chat panel. It reads the same data you see — scoped entirely to your workspace, never another tenant’s — and it knows the product: pricing models, catalog structure, billing timing, renewals, and how metrics like MRR and churn are defined. Ask it a question and it fetches the live numbers; ask it to build something and it drafts the change for your approval. It answers with the takeaway plus cards — metric tiles, charts, and tables — beside the reply.

02What you can ask it

Read-only questions — answered from your live data or the product knowledge base. Nothing here changes anything.

Clients

Find the client called Acme. / Which clients are still on a trial?

Subscriptions

Show all active subscriptions for Acme. / What is renewing next month?

Invoices

List Acme's overdue invoices. / How many invoices are still pending?

Revenue analytics

What is our current MRR and how has it moved over 6 months? / What is our ARR by plan? / What is our monthly churn?

Catalog

List our bundles and which are published. / What metrics do we meter? / Which regions and currencies do we sell in?

Metered usage

Which usage quantities are missing in arrears? / How many seats is Acme using this period?

How Revvly works

How is MRR calculated? / What is the difference between volume-tiered and graduated pricing?

Pricing guidance

Which pricing model should I use for seat-based billing with an included allowance?

OKR hypotheses

Which Key Results are falling behind this cycle? / What have we already tried on New deals?

03What it can do

High-impact actions wait for a confirmation card you approve or decline. Reversible writes — notes, logged emails, creating a prospect or task — run immediately.

Create a client

Add a customer organization — with billing details, a trial window, and contacts.

Create a subscription

Subscribe a client to a bundle, modules, or variable components, priced from your catalog.

Create a bundle

Draft a new plan customers can buy, with its modules and per-region base prices.

Create a module

Draft a capability group, optionally with priced tiers.

Create a metric

Define a metered unit — seats, GB, transactions — to bill usage against.

Create a variable component

Wrap a metric in a usage-based pricing model (flat, tiered, overage, commission, …).

Publish a variable component

Freeze a draft component so it can be attached to a module or bundle.

Publish a module

Freeze a draft module (once fully priced) so it can be added to a bundle.

Publish a bundle

Freeze a draft bundle (once fully priced) so it can be sold on a subscription.

Attach a variable component

Add a published component to a draft module or bundle, with its inclusions and price.

Create a new draft version

Branch an editable draft from a published bundle, module, or component — the way to change a bundle or module, since published catalog entities are immutable. A branched component still needs its structure edited in the dashboard before the draft is published.

Enter a usage quantity

Seal a missing in-arrears quantity on a closed billing period.

Waive a usage period

Bill nothing for a closed period that has no usage, with a recorded reason.

Propose hypotheses for a Key Result

For a Key Result falling behind, up to three testable bets grounded in its numbers and past learnings, added to Backlog. It never claims a bet will hit the target, and avoids bets already tried.

04How a change happens

Reversible writes run as soon as the Assistant does them — no card. For a high-impact change, it looks up what it needs (say, your regions), then shows a confirmation card summarizing exactly what it will create — with names, not ids. Nothing on that card is written until you confirm. On confirm, the change runs through the same validation as the manual screens (name uniqueness, region and currency checks, catalog pricing), and the Assistant reports back with a link to the new record. Decline it and nothing happens. A proposal is valid for 15 minutes, after which it cannot be approved. Each mutation is its own proposal: approving one does not approve the next, and a stale proposal whose underlying record has moved since it was staged is refused rather than writing over someone else’s edit.

A high-impact change

You ask
For a high-impact change.
Confirmation card
Exactly what it will create, with names, not ids.
You confirm
Runs through the same validation as the manual screens.
or
You decline
Nothing happens.
or
After 15 minutes
The proposal cannot be approved.
or
The record has moved
Refused, rather than overwriting someone else’s edit.
Assistant reports back
With a link to the new record.

Nothing is written until you confirm, and a late or outdated confirmation is refused.

05Building catalog in the right order

Usage-based pricing is built in steps, and the Assistant walks them in order across a few confirmations: create a metric, wrap it in a variable component, publish the component, then attach it to a module or bundle with its price. A component must be published before it can be attached, a module must be published before it can join a bundle, and a bundle must be published before it can be sold on a subscription — published catalog items are immutable, so you attach into drafts. If you ask to attach, sell, or subscribe to something still in draft, the Assistant proposes publishing it first.

Asking to change a published bundle or module — attach a component to a live bundle, adjust a module’s billing periods — is not a dead end. The Assistant branches a new draft version from it first, carrying over its modules, tiers, attached components and prices, applies your change to that draft, and proposes publishing it once you approve. The version you had stays exactly as it was until the new one ships. A published variable component can be branched the same way, but chat has no tool to edit its structure — the Assistant will point you to the component’s page in the dashboard to finish that edit before the draft is published.

06Numbers you can trust

The Assistant never quotes figures from memory. Every number — MRR, ARR, churn, an invoice balance — comes from a live query against your data at the moment you ask, using the same analytics engine the dashboard does. Knowledge questions cite the documentation they draw from, so you can read further.

07What it remembers

Tell the Assistant something that should hold beyond one conversation, such as how you like answers or a rule of the business (“we never discount more than 20%”), and it offers to remember it on a card. Nothing is saved until you approve. Every later conversation starts with what was saved and with your Business profile (Org Settings, Business), so its examples and suggestions fit your business.

For you

Shapes only your own conversations. Nobody else sees it, admins included.

For everyone

Shapes every member’s conversations. Only workspace admins can save, change or remove one.

Limits

One sentence each, up to 300 characters. Up to 50 for the workspace and 30 per person. Saving the same words twice keeps one.

Ask “what do you remember?” to see it all, then ask the Assistant to reword or forget one; each change is a card you approve. The same list is under Org Settings, General, Agent memory, where you can also edit in place and choose Show what the agent sees to read exactly what the Assistant is given at the start of every turn. What is saved is treated as information from your team, not as instructions: it cannot change what the Assistant is allowed to do. When you ask for something a memory rules out, it does what you asked and says which memory it departs from. Claude and other MCP clients do not receive memories.

08Outside the dashboard

The same tools work from Claude or any MCP client — clients, subscriptions, invoices, live MRR, catalog, OKR hypotheses, and writes such as entering a missing in-arrears usage quantity or proposing hypotheses for a Key Result. Mutating tools return a proposal; the credential owner must approve it while signed in to Revvly. The MCP client cannot approve its own request. The full list of Assistant tools is in what the Assistant can do. See Connect Claude with MCP.