Xalios
Book a discovery call
The AI assistant and agents your teams can answer for

Answers you can stand behind. Work you can
answer for.

Every answer sourced.Every action owned.Every decision on record.

Xalios answers your teams’ questions from your own documents and does routine work across the apps they already use. Every answer shows its source or says it can’t back it. Every agent works for a named person, checks with your team until their approvals say it needn’t, and keeps a record of every decision.

HFHead of Finance OperationsAnswers for the work
Marks payments and bank-detail changes as serious. Those never run alone
3 people4 agents
RKFinancial ControllerResponsible for 2 agents
Approves, edits or declines in the review queue
Payables
AIPayment statusLevel 3
150 approvals in a row
AIRemittance chasersLevel 2
Stepped back after 2 mistakes in its last 20
JDAccounts AssistantPerson
Takes what the agents send back
MOAP ManagerResponsible for 2 agents
Set up both agents from a job description
Supplier enquiries
AIInvoice copiesLevel 4
750 approvals in a row
AIExpense queriesLevel 1
New task. MO approves every item
30 approvals150 approvals750 approvals
Finance operations
answers
Answers with their source

The assistant cites the document behind every claim. When it can’t back an answer, or can’t confirm you may see a source, it says so and why.

agents
A named person for every agent

Each agent holds a role in a department and works for a named person. It can carry its own Microsoft Entra identity.

record
One record of every decision

Every action, the check that let it run and the person who approved it go into one append-only record.

compliance programme
All four in progress, with a target of the end of Q4 2026. None is certified yet.
SOC 2security controls, in progress
ISO 27001information security, in progress
ISO 42001AI management, in progress
GDPRdata protection, in progress

Built for everyone who has to sign it off

The head of the function buys it and the CFO signs. IT, risk and data protection must approve it, and in Germany and France so must works councils.

Buys it · finance, HR, partner services, procurement
Function head

Hand your team’s routine work to AI and stay the one who decides how much. Your team’s approvals set the pace, task by task.

Signs
CFO

You can attribute value item by item, because Xalios is priced by the item handled. The spending cap can’t be switched off. Outsourcing spend can be replaced without fewer colleagues.

Must approve
CIO and Microsoft 365 admin

Sign-in runs only through your identity provider, and agents can carry their own Entra identities. Each customer has its own database in UK and EU Azure regions. Business teams build inside one governed product instead of adding to the sprawl.

Must approve
Risk, audit and data protection

Actions you mark as serious never run alone. One record holds every action, the check that let it run and who approved it. We bring a pre-filled data protection impact assessment.

AI has moved from answering questions to doing work.
Someone has to answer for what it does.

Once AI sends replies and changes records, someone in the business has to explain what it did and why. Companies without that person lose out in one of three ways.

Three ways companies lose
They check everything and pay twice

Workday’s research found that 37% to 40% of AI time savings are lost to rework.

They let agents run, then switch them off

Gartner expects 40% of enterprises to demote or decommission autonomous agents by 2027.

IT controls the AI and nobody in the business answers for it

In a vendor survey by Dataiku, responsibility for agents that go wrong split almost evenly: shared teams 23%, IT 21%, data and AI 20%, risk 18%.

AnswersAgentsAutonomyRecord

What lets your team answer for the work

The assistant shows where each answer came from. Each agent has a place in your organisation and a person responsible for it. Your team decides, task by task, how much it does alone. Everything goes into one record.

01 / Answers
Answers you can stand behind

The assistant answers from your SharePoint, OneDrive or Google Drive and cites every claim. It follows document permissions. When it can’t confirm the asker may see a source, or can’t back an answer, it refuses and says why.

Answers →
02 / Agents
Every agent belongs somewhere

Each agent holds a role in a department with a named person responsible. You set it up from a plain-language job description and rules written in plain English.

Agents →
03 / Autonomy
Autonomy set by the people who answer for it

Each task starts with every item approved by a person. It does more of that task alone after each run of approvals (30, 150 and 750 in a row by default), and steps back after 2 mistakes in its last 20.

Autonomy →
04 / Record
One record

An append-only log of every action, the check that let it run and who approved it. The history is tamper-evident, so you can trace one decision end to end.

The record →

One review queue for everything that needs a person

Your team approves, edits or declines each item. Those decisions set how much of each task the agent does alone.

Autonomy is per task, so invoice copies can be at level 4 while every expense reply still waits for a person. Each item arrives with its sources attached. Approvals with edits count as neutral. Anything you mark as serious lands here at every level.

Your team’s review queue
Tuesday · 09:00
done alone overnight
on the record
Payment status replies
Invoice copies sent
Remittance chasers sent
Supplier questions answered
needs you
3 items
serious06:42 · bank details

A supplier has asked to change its bank details. You marked this as serious, so it waits for a person at every level. The request and the supplier record are attached.

ApproveDeclinesources attached
approval08:05 · expense queries

Draft reply on mileage rates, citing the expenses policy. This task is new, so a person approves every item.

ApproveEditDecline
stepped back08:20 · remittance chasers

Two mistakes in its last 20 stepped this task back, so more of its items come to you.

Your team decides how much it hands over. The record shows who decided what.

How it is built
01 / Triggers
Starts where the work arrives

Agents pick up work from email, Teams, Slack, files or a schedule.

02 / Apps
Works in about 200 business apps

Routine follow-up happens in the apps your team already uses. Longer work is tracked as a case from start to finish.

03 / Shared memory
Facts kept where they apply

Each fact belongs to the organisation, a department, a role or a source. When two conflict, a person decides.

04 / Pricing
Priced by the item handled

You pay for each item the agents handle, not for each seat.

03 / shared memory, in practice

Two sources disagree. A person decides which stands.

Each fact belongs to the organisation, a department, a role or a source. When a new document contradicts one, the conflict goes to the person responsible. The version they replace stays in the history.

Payment terms / Kelmore Supplies
example record
30 days
replaced
effective 01 apr 2026 → 11 jun 2026
Master agreement, signed 22 Mar 2026
conflict sent to the financial controller
45 days
in force
effective 12 jun 2026 →
Contract variation, approved by the financial controller on 09 Jun 2026

The 30-day terms stay in the history, with the person who replaced them and the date.

Fixed limits

What never moves

Autonomy changes with your team’s decisions. These limits hold at every level and for every task. No agent sets its own autonomy.

Actions you mark as serious never run alone

You choose what counts as serious, such as payments or bank-detail changes. Those wait for a person at every level.

The spending cap can’t be switched off

Where money moves, a spending cap applies. There is no setting that turns it off.

Only a run of approvals moves a task up

Autonomy rises only after each run of approvals, 30, 150 and 750 in a row by default or the steps your organisation chose at setup. Nobody can skip a step.

2 mistakes in the last 20 step a task back

The step back happens on its own. Nobody has to spot the pattern first.

5% of the work is spot-checked

A person reviews a sample at every level, including work the agent did alone.

Edited approvals count as neutral

An item your team had to change before approving doesn’t count toward the next level.

Answers follow document permissions

The assistant refuses when it can’t confirm the asker may see a source.

You choose what counts as serious. The rest hold for every task.

Where teams start

Pick the work one team already answers for

The best first stream has a few hundred items a month across a handful of request types. That is enough volume for autonomy to move. The assistant, the agents, the limits and the record work the same way in every team.

All teams →
Finance

Supplier and staff questions about payments, invoices and remittances

Cited answers on terms and look-ups. Bank-detail changes and payments are never left to the agent.

Answers for it: Financial controllerSee the team page →
HR

Policy questions, leave and benefits requests

Answers that respect restricted documents. When two policy versions conflict, a person decides.

Answers for it: HR operations leadSee the team page →
Partners and customers

Partner and customer questions through the embedded assistant

Each partner tier sees only its own documents, and AI disclosure is built in.

Answers for it: Partner or customer service leadSee the team page →
Procurement

Supplier onboarding and purchasing

Multi-step cases run under a spending cap, and serious actions never run alone.

Answers for it: Head of procurementSee the team page →
Compliance

Compliance and policy questions

Every answer cites its source, so your team can verify it before relying on it.

Answers for it: Compliance or legal operations leadSee the team page →
Sales

Sales enablement and RFP answers

Answers come from approved material. Where no approved source exists, the assistant says so.

Answers for it: Bid or sales operations leadSee the team page →
The 30-day pilot

Start with one stream of work

A paid 30-day pilot on one stream of work your team already answers for. Autonomy moves only as your team’s approvals build up, and you pay by the item handled.

Book a discovery call→paid · 30 days · one stream of work
Every item approved at the start

Your team sees and decides every item first, so you can judge the answers before handing anything over.

The admission pack

A pre-filled data protection impact assessment, a SOC 2 audit date and a works-council pack for your approvers.

No promise of fewer colleagues

The agents take routine items and your people decide the rest. If the aim is to cut staff, we’re the wrong fit.