Rutba
Start free

Rutba Assistant · across the suiteComing soon

Ask your business a question, from whichever app you are in.

One assistant across the whole suite. It answers from your own records, reads them as you — with your permissions and never beyond them — and shows you what it read.

Most business software has added a chat box that answers questions about the software. This one answers questions about your business, from your records, and it cannot see or do anything you could not see or do yourself.

A question asked from an app's top bar is narrowed to the API calls the person asking is allowed to make, sent through the same API the app's own screens use — where the organization's licence, the person's role and the endpoint itself are each checked — and answered with a list of the endpoints that were read.1. You ask2. Narrowed to you3. The same API4. Answered, with sourcesfrom the app you are inWhich branches arerunning low?or say it out loudIt acts as you:your sign-in, nothing more.before anything is sentyour rolesyour licence1,069 read calls exist.It is offered only yours.no private doorLicensed?your organization has itAllowed?your role may do itExists?the API serves itA refusal comes back inthe words of the check.checkable, not trustedRead from 3 endpointsGET /branchesGET /stock-levelsGET /stock-alertsOr ask for a report, andit lands in your Drive.

51

named agents — one for each of 41 apps and 10 areas

1,069

read-only API calls it can answer from

39

apps with the Ask button in their top bar

0

permissions of its own

Not released yet. Rutba Assistant is built into the suite and exercised against it running, but it is not switched on for customers and cannot be bought yet. This page describes it as it will ship; what is still open is listed at the end.

What asking looks like

One button in the top bar, and an answer you can check.

There is no separate app to open and no place to copy your question to. The assistant knows which app you are in, and it answers by reading the same API the screen in front of you reads.

01

Ask from the screen you are on

The Ask button sits in the top bar of 39 apps. Dock the panel to the side or the bottom and keep working — it is not a page you have to leave your work to reach.

02

Watch it read

The answer arrives as it is written, and while it works it names the part of your records it is reading. A question aimed at the wrong place shows itself in the first second, not after the answer.

03

Every answer lists its sources

Under each answer are the API calls it made. They are taken from what actually ran, not from what a model says it did, so an answer can be checked rather than trusted.

04

A report, when a report is the answer

Press Report instead of Ask and the answer lands in your own Drive as a spreadsheet, with a native chart and a Sources sheet. The model picks the layout; every figure is filled in from the recorded responses, so it never types a number.

05

Speak instead of typing

Up to two minutes of speech becomes text in the box, for you to read and correct before anything is sent. Nothing you said is acted on until you have seen it written down.

06

A “no” you can act on

“Your organization is not licensed for Payroll” is a purchasing question. “You do not have access” is one for an administrator. “Out of credit” says where to top up. They stay three different answers, because they send you to three different people.

Named agents

An agent for every app, and one for every area of the business.

Ask the whole suite, or address one part of it by name — the Payroll agent, or Finance & Accounting across its apps. Naming an agent keeps a question inside it; it never widens what you may see. There are 51, generated from the suite's own list of apps, so a new app has its agent the day it ships.

Area agent

Mail, Chat & Calls

and an agent for each app in it

Area agent

Marketing & Web

and an agent for each app in it

Area agent

Facilities & Assets

and an agent for each app in it

Area agent

Finance & Accounting

and an agent for each app in it

Area agent

Inventory & Purchasing

and an agent for each app in it

Area agent

Logistics & Fleet

and an agent for each app in it

Area agent

People & Payroll

and an agent for each app in it

Area agent

Manufacturing

and an agent for each app in it

Area agent

Sales & Customers

and an agent for each app in it

Area agent

Documents & Sign

and an agent for each app in it

An agent with nothing you can reach says so. Ask the Payroll agent a question while holding no role in Payroll and it tells you that, in those words, instead of answering from general knowledge — a confident salary figure with no source is the one answer worse than none.

It works as you

The assistant has no permissions of its own. That is the design.

Every limit you already rely on applies to it unchanged, because it goes through the same checks as you do rather than around them. Nothing here depends on a model choosing to behave.

No account of its own

Every call carries your own sign-in and meets the same three checks the screen would: is your organization licensed for this, may you do it, does it exist. A tool you could not use is never even offered to it.

A permission removed is removed at once

Your roles are read afresh on every call. A role taken away this morning is gone from the assistant this morning — including from anything you had asked it to do on a schedule.

Naming an agent only ever narrows

Asking the Payroll agent keeps a question inside Payroll. It never adds a right you do not hold, and an agent with nothing you can reach says so instead of answering from general knowledge.

Its changes are signed as its own

A change the assistant makes is recorded as the assistant’s, under whose grant, in which run. The mark is signed and checked, so “the assistant did it” cannot be claimed of something it did not do.

What it will never do, whoever asks

A few things stay out of its reach even for somebody who holds every role.

“The assistant did it” is not an answer a finance director accepts about a payroll run, however correct the permissions were. Your organization can add to this list. It cannot take anything off it — and neither can the assistant.

  • Run or approve payroll

    Paying people is the one change where a wrong number reaches somebody’s bank account and cannot be quietly corrected.

  • Post to the general ledger

    A posting is an accounting event. Reversing it is another one, and an auditor sees both.

  • Delete anything

    Deletion has no undo. The assistant may propose it; a person does it.

  • Change who may do what

    Every other rule rests on who is allowed what. That is never delegated.

  • Change its own limits, grants or balance

    An assistant that could widen its own authority or top up its own credit would have no limits at all.

  • Send anything to your customers

    What reaches a customer in your name is yours to send. A message people have read cannot be unsent.

Plans and chores

Work that takes longer than a reply runs as a plan a person approved.

The panel only reads. When you want the assistant to do something — several steps, or the same job every Monday — it becomes a plan: every step written out, approved by a person, run by a worker, stopped the moment a step fails.

A plan approved by a person runs its steps in order. The first step succeeds, the second fails and stops the plan with its reason kept, the third never runs, and the person is notified. Once the cause is fixed the plan resumes from the failed step without repeating the first.The planevery step, and theexact request it sendsApproved by youStep 1Gather the countssucceededStep 2Build the reorder liststopped — reason keptStep 3Save the reportnever runs on a failureYou are told, and it resumes fromstep 2 once the cause is fixedClose the tab: a worker carries it on.

A plan you approve, request by request

Before anything runs you see every step and the exact request each will send. If a step is changed after you approved it, it does not run.

It stops where it fails

A step runs only when the one before it succeeded. A failed step stops the plan and keeps the reason; fix the cause and it resumes from there, without repeating what already finished.

Close the tab

A worker carries the plan on without you, and you are told when it finishes, when it stops, and when it needs an approval.

On a timetable

A chore can run on a schedule. It cannot overlap itself, cannot outlive the grant behind it, and switches itself off after three failures in a row rather than paying for a fourth.

Grants that end

Letting it work while you are away takes an explicit grant: named apps, never more than you hold, and an end date you pick — there is no “until revoked”. Revoke it and the next call stops.

Your organization decides

Switches in your admin console

  • Switch the assistant on or off for the whole organization
  • Turn it off in particular apps
  • Turn voice input off
  • Require that no data leaves your own network — which limits it to what a model running inside that network can do
  • Fix the capability tier it may use
  • Add your own entries to the list of things it may never do

What it costs

One prepaid balance

The assistant runs on AI credits your organization buys in advance — one balance, not a separate bill per feature. A quick question costs like a quick question: what each call used is metered in four separately priced parts rather than rounded into one.

The assistant’s console shows the balance and what it has cost, by what it was asked to do. Running out stops it with a message that says so and where to top up — never a silent failure.

From outside the suite

Bring your own assistant

An assistant you already use can connect through the Model Context Protocol with your organization’s own key, and meets the same checks. It gets 6 tools:

  • list_agentsfind the right agent
  • search_endpointssearch the API
  • describe_endpointread how an endpoint works
  • readread through it
  • propose_changepropose a change for a person to approve
  • list_proposalssee what it has proposed

Connected read-only, the change tools are not offered at all. Otherwise a change is only ever proposed, and waits for a person to approve it.

Questions

What people ask about Rutba Assistant

Does Rutba Assistant change my data?

Not from the panel. There it only reads, and the one thing it writes is a report into your own Drive when you ask for one. Changes happen only as a plan a person has approved, step by step, under a grant that ends — and a short list of things, from running payroll to deleting records, stays out of its reach whatever anybody approves.

Can it see anything I cannot?

No. It has no account of its own. Every call it makes carries your sign-in and is checked exactly as the screen you would otherwise use is checked, so it sees what your role shows you and nothing else.

Is my data used to train a model?

Rutba never uses your records to train anything, and one organization’s records never inform another organization’s answer. Which model provider processes a question, and on what terms, will be listed with our sub-processors before the assistant processes anybody’s data.

How is it paid for?

With prepaid AI credits held by your organization — one balance, not a separate bill per feature. The assistant’s console shows the balance and what it has cost, broken down by what it was asked to do, and running out stops the assistant with a message saying so rather than failing quietly. Credit prices will be on the pricing page when it is released.

Can we turn it off, or limit where it works?

Yes. Your administrators can switch it off entirely, turn it off in particular apps, turn voice off, require that no data leaves your own network, and add to the list of things it may never do. They cannot remove anything from Rutba’s own list — and neither can the assistant.

Which apps does it work in?

The Ask button is in the top bar of 39 apps, and there is a named agent for each of 41 apps and 10 areas. It is not in the public storefront or the delivery-tracking page, which have no signed-in user to act as; an assistant for your shoppers is a separate product.

Where it is not finished: Not released. It is built into the suite and exercised against it running, but it is not switched on for customers and it is not for sale yet.How well it answers is measured against a set of real business questions before it is released, and the release waits on that number.Credit prices are not published yet.The assistant for your shoppers, on your own storefront, is a separate product. Its engine is built; the part a shopper sees is not.

The rest of what is in every app is on its own page: dashboards that answer, and a way to tell us something is wrong.

Want it the day it switches on?

Register interest and tell us the first thing you would ask it. We will come to you when it opens.

Share this

One family

The rest of Rutba

One account across all of it. Sign in once and the products know each other.

PORTAL-ASSISTANT · db013a8