Alltegrio AI Core
All your enterprise AI. One private platform.
Lower AI costs, every dollar of spend under control, your data isolated on your own installation, and access governed by role — inside the company and out to the web. One closed platform for every AI solution your business runs.
One question, three business systems, one answer — the morning briefing, assembled in seconds.
What your business gets
Savings, control over spend, isolated data and access by role. The things a company asks about first are the things AI Core is built around.
Lower cost of AI
One platform instead of a separate subscription for every tool and every seat. You pay for the model work you actually use, with your own keys, and routine tasks go to an inexpensive model while the hard ones get a strong one.
Plans and models ↓Spend under control
Plans and limits for a person, a team and a whole organization. Spend is visible by user, by model and by tool as it happens, so nobody runs up a bill unnoticed and every budget has an owner.
Limits and spend ↓Your data stays isolated
Your own installation on a dedicated server: in your cloud, in your data centre or in ours. No shared tenant, no public sign-up, nothing mixed with another company’s data.
Security and deployment ↓Multi-tenancy for groups
Subsidiaries, business units and clients work as separate organizations on one platform. Each has its own people, systems, budget and rules, and none of them sees the others.
Organizations ↓Access by role, inside
Every person and every role reaches only the systems and actions granted to them. The assistant cannot query what the person asking is not allowed to see, whatever they type.
Roles and access ↓Control of what goes outside
You decide which websites, web search, external agents and AI models the platform may use. Anything outside the policy is refused in plain words, and the refusal is on record.
Safety in action ↓Not another AI tool. The home for all of them.
Most enterprises end up with a patchwork: one assistant bolted onto the CRM, another inside the help desk, a chatbot for HR, a separate tool for documents, a dozen experiments nobody governs. Each knows a sliver of the business. None of them knows the others exist. AI Core replaces the patchwork with a single platform.
Everything under one roof
Every AI capability your company needs — assistants, agents, analysis, research, voice — lives in one place and works as one.
Integrates with anything
CRM, ERP, databases, data warehouses, help desks, HR systems, documents, email, your own AI agents, the open web. If your enterprise uses it, AI Core connects to it.
One answer instead of ten tools
Not separate AI products wired to separate systems, but one intelligence that sees across all of them.
Why AI Core and not another AI subscription
Enterprise editions of ChatGPT, Claude and Gemini give your people a capable assistant. They do not give you a platform that is yours: on your server, with any model, connected to the systems only your company has, with agents built for your processes.
Your own installation
AI Core runs on a server dedicated to your company — in your cloud, in your data centre or in ours. Your data does not share a tenant with anyone.
No dependence on one AI vendor
Models from OpenAI, Anthropic, Google, Mistral or your own private models, chosen per task, with your keys and a reserve provider. Change the model without changing the platform.
Built for groups of companies
Subsidiaries, business units and clients run as isolated organizations on one platform, each with its own people, systems, budget and rules.
Delivered turnkey, with your systems
We connect the systems you run today, including the old ones without an API, build the agents your processes need and hand over a working platform.
| Alltegrio AI Core | Enterprise AI subscriptions | Building it yourself | |
|---|---|---|---|
| Runs on your own dedicated server | Yes | No | Yes, by your team |
| Any model, your keys, a reserve provider | Yes | One vendor | Yes, by your team |
| Connects legacy systems without an API | Yes, built for you | Standard connectors | Your team builds it |
| Custom agents for your processes | Built by Alltegrio | Do it yourself | Your team builds it |
| Isolated organizations for a holding | Yes | Separate subscriptions | Your team builds it |
| Roles, rules, skills and limits by chat | Yes | Partly | No |
| Agents on a schedule and by event | Yes | Partly | Your team builds it |
| Time to a working platform | Weeks | Days, without your systems | Months |
How it fits together
One governed layer between your people and everything they work with.
People
Sales · Operations · Finance · Support · HR · Management
Chat, voice, phone, scheduled deliveries
Alltegrio AI Core
Orchestrator · Roles, rules and skills · Configurator · Enterprise Context
Access control · approvals · limits · audit trail
Agents
Built-in · your own · built by Alltegrio · third-party
On request, on schedule, on event
Your systems and data
CRM · ERP · databases · documents · mail · internal tools · the web
Connected through the open MCP standard
AI models
Cloud, private or hybrid — chosen per task
Your keys, a reserve provider, routing by sensitivity
The heart of the platform
Roles: an expert for every job in your company
A general AI model knows the world. It does not know how your company works. A role closes that gap: it turns AI Core into the specialist one team needs — the operations chief of staff, the revenue strategist, the finance controller, the people partner, the customer care lead, the market analyst.
People do not configure a model. They choose a role, the way they would walk over to the right colleague, and start talking. The role already knows the job, speaks the team’s language, follows the company’s standards, uses the right systems and proposes where to begin.
One platform holds as many roles as your company has jobs. One person can keep several and switch between them from one conversation to the next. Roles are written in plain language — or simply dictated to the Configurator.
What a role is made of
Instruction
Who the assistant is in this role and what the work is about: the team it serves, the goals it keeps in mind, the way it should reason.
Rules
The company standards this role follows in every answer, ticked from the shared library.
Skills
The playbooks this role can pick up: the procedures it follows when a task calls for them.
Connections
The systems and agents this role may use, and what it may read or change in each of them.
First questions
What the role offers the moment a chat opens, so the first click already leads to useful work.
Sharing
A role that works is shared with a team or the whole organization, and everyone starts with the same expert.
Two libraries every role draws on: rules and skills
Roles do not repeat the same instructions. The company keeps its standards and its procedures in two shared libraries, and each role takes what it needs. Change a rule or a skill once, and every role that uses it changes with it.
How it must behave
Rules
A rule is a company standard applied to every answer: tone, reporting conventions, confidentiality. Written once, followed by every role you tick.
See it in the tour ›How the work is done
Skills
A skill is a playbook: the steps of a briefing, a review, an escalation. The assistant opens the right one when the task calls for it and follows your procedure.
See it in the tour ›Enterprise Context: AI that knows your business
AI becomes more valuable when it understands not only the request, but the enterprise in which that request must be completed. Enterprise Context is the governed knowledge layer that gives every authorized agent the right operational background, so answers and actions match how your organization actually works.
Business knowledge
Terms, entities and processes: what a "file", a "lane" or a "renewal" means in your company.
Policies and access
Roles, rules and approvals: who may see what, and which actions wait for a person.
System context
Sources, records and relationships: where each fact lives and how systems relate.
More relevant outcomes
Agents interpret requests using shared business definitions and approved knowledge.
Consistent execution
Workflows reflect your rules, responsibilities and approval paths.
Safer adoption
Access to context and connected systems is scoped by role and task.
Reusable intelligence
New agents and use cases draw on one maintained context foundation.
Everything in one platform
Twenty-one capabilities that work as one. Each opens the part of the tour that shows it.
Agents
Built-in, your own, third-party and custom-built agents working as one team.
Scheduled and event-driven runs
Agents that start by timetable, by event or by threshold and deliver the result.
Document search and memory
Answers from your documents with the passage shown, and context kept between conversations.
Orchestrator
Decides which systems and agents hold the answer, calls them in parallel and writes one reply.
Transparent answers
Every call, every source, the time and the cost behind each answer, one click away.
Roles
A specialist for every team: its own instruction, rules, skills, systems and first questions.
Rules
Your standards applied to every answer: tone, reporting conventions, confidentiality.
Skills
Your playbooks, picked up by the assistant exactly when the task calls for them.
Configurator
An AI administrator you talk to. It sets up roles, rules, skills and access, and shows every change first.
Voice
Full voice control: speak, listen and keep the conversation going hands-free.
Connections
Any system, database or agent joins through an open standard, with read and write permissions per action.
Built-in agents
Ready-made agents for research, documents and analysis beside your own.
Google Workspace
Gmail, Docs, Drive and Sheets under your company’s own account.
Web research
Live web search and reading under your policy, with every source cited.
People and access
Invitations, administrators, and access to each system person by person.
Plans and limits
Predictable budgets: plans per person and ceilings per group of systems.
Model freedom
The right AI model for each kind of task, and your own provider keys if you prefer.
Analytics and spend
Conversations, answers, cost and system usage by person, by day, by tool.
Many organizations
Subsidiaries, business units or clients side by side, fully isolated from each other.
Mobile
The full platform on a phone: the same conversations, roles and answers.
Your way
A personal profile, three themes and the interface size that suits your screen.
Step 1 · Ask
Ask anything. Get the whole answer.
One conversation reaches every system you own.
Type a question the way you would ask a senior colleague. The Orchestrator decides which systems and agents hold the answer, calls them in parallel, and writes a single reply — with the numbers, where each came from, and what to do next.
Sales, finance, operations, customer care, strategy: every team asks in its own words and gets an answer built from live company data. When the answer is outside your walls, AI Core researches the live web and cites every source.
1.1 · Operations
One question across three departments
A follow-up to the morning briefing pulls shipment tracking, open support tickets and the sales pipeline into a single answer. The assistant connects the delayed shipment to the customer behind it and the renewal at risk, then finishes with the message to that customer, ready to send.
A follow-up joins operations, customer care and sales data — and ends with the message ready to send.
1.2 · Sales forecast
The quarter, forecast from live CRM data
A sales leader asks where the quarter will land against quota. AI Core reads the pipeline, weighs the deals by stage, shows the gap to target and names the deals that decide it — with who should do what this week.
"Where will Q4 land against quota?" — forecast, gap and the deals that close it.
1.3 · Meeting brief
A one-page brief before every meeting
Before a pricing call the assistant assembles the relationship history, the numbers, the open issues and the ask on one page. Nobody walks into a customer meeting unprepared, and nobody spends an hour collecting what five systems already know.
The relationship, the numbers, the open issues, the ask — on one page before the call.
1.4 · Finance
Cash and receivables, answered
Finance asks which invoices to chase first. The answer ranks overdue invoices by amount and age, proposes a collection plan and adds a 30-day cash outlook — an analysis that used to take a spreadsheet and an afternoon.
Which invoices to chase first, and what the next 30 days of cash look like.
1.5 · Operations review
A comparison with a recommendation
Operations compares four carriers on on-time delivery, damage rate and cost. The platform lays the figures side by side, names the weak point and makes a clear recommendation instead of leaving a table for someone to interpret.
Four carriers compared on three measures — and a clear recommendation.
1.6 · Reply draft
A reply worth sending
The next request drafts the answer to the customer. It is written from the real facts — where the shipment is, why it is late, when it lands — in the tone the company’s rules prescribe, ready to review and send.
The reply is drafted from the real facts: where the shipment is, why, and when it lands.
1.7 · Market research
Research beyond your own walls
When the answer is not in your systems, AI Core researches the live web: it searches, reads the pages, summarises what matters for your business and cites every source so the reader can check it.
Live market research: searched, read, summarised and sourced.
1.8 · Business review
One request, five calls across three systems
"Prepare the quarterly business review" starts a chain: the account from the CRM, overdue invoices from finance, on-time performance and delays from operations, open deals — then the document, written to the skill's outline.
One request, five calls across three systems: a quarterly business review.
1.9 · Tables
Numbers arrive as tables, ready to read
Ask for a table and the answer is one: revenue by month with the change, margin by service line beside it. The figures come straight from the finance system.
Numbers arrive as tables, ready to read.
1.10 · Action
Not only answers
When a connected system offers an action, the assistant can take it: here a deal moves to the Proposal stage. Which actions change data is marked per system by the administrator.
Not only answers: the deal is moved in the CRM, and the answer says so.
1.11 · Churn risk
Revenue: which accounts may leave, and what each is worth
The pipeline, the account health and the open service issues are read together — the kind of answer that normally takes three screens and a meeting.
Revenue: which accounts may leave, and what each is worth.
1.12 · Response mode
Cheap, Optimal or Quality
The response mode is one tap in the message box. It lets a person trade speed and cost against depth, answer by answer.
Cheap, Optimal or Quality — choose how much thinking an answer deserves.
1.13 · Same question finance
One question, asked as the finance desk
Ask "How are we doing with Helix Medical Supply?" as the finance desk and the answer is about revenue, the open deal and what is overdue. The role decides what matters.
One question, asked as the finance desk: the money.
1.14 · Same question care
The same question, asked as customer care
The same words in the Customer Care Desk role bring back tickets, the delayed shipment and the relationship — a different, equally right answer, from the same company data.
The same question, asked as customer care: the service.
Step 2 · See the work
See the work behind every answer.
No black box. Ever.
Every answer shows which systems and agents were called, what each was asked, what it returned, how long it took and what the step consumed. Open any step and read the full exchange.
When someone asks "where did this number come from?", the answer is one click away. Your auditors, your security team and your board get the same view your people do.
2.1 · Opened call
Every call can be opened
Above each answer sits the list of systems and agents that were called. Open any line to see exactly what was asked and exactly what came back, with its time and size. "Where did this number come from?" is answered in one click.
Open any call to see exactly what was asked and what came back.
2.2 · Call CRM
Open the CRM call
Any row above an answer opens on a click and shows the whole exchange with that system — what was asked and what was returned — with a copy button on each block.
Open the CRM call: the pipeline summary it asked for and the figures that came back.
2.3 · Skill opened
See which playbook the answer followed
When the assistant uses a skill, the skill is named above the answer and opens to the exact text it read. Nothing about how an answer was made is hidden.
See which playbook the answer followed — and read it.
Step 3 · Agents
Agents that do the work.
Built-in agents, your own agents and agents we build for you — one team under one orchestrator.
An agent is a specialist that does a job from start to finish: it reads the data it needs, uses the tools it is allowed to use, and returns a result or carries out the action. AI Core runs any number of them side by side and makes them work as a team.
The Orchestrator hands each request to the right agents, runs them in parallel, checks what they return and assembles one answer. Your people talk to one assistant; behind it works as many agents as the task requires.
Built-in agents
Web research, page reading, documents and text analysis are ready from the first day.
Your own agents
Agents your teams have already built join the platform through the open MCP standard and become available to every role you allow.
Custom agents by Alltegrio
We design and build agents for your processes: claims, pricing, procurement, scheduling, verification, reporting — and deploy them on the platform.
Third-party agents
Agents and AI services from other vendors are connected and governed the same way as your own.
Multi-agent work
One request can pass through several agents in sequence or in parallel: collect, check, analyse, write, act.
Agents under control
Each agent has its permissions, its limits and a full trail of what it did. Which actions may change data, and which of them wait for a person’s approval, is set for each system.
Step 4 · On schedule
Work that starts by itself.
Agents run on a schedule and react to events — nobody has to remember to ask.
The morning briefing is on the leadership team’s phones at 8:00. The cash review arrives every Monday. The moment a key customer’s shipment slips, the escalation playbook starts and the account manager gets the facts and a drafted message.
Any request a person can make in a chat can be scheduled or tied to an event in a connected system. The result is delivered where people already work — in the platform, by email or in the team messenger — and every run is recorded with the same trail as a conversation.
By timetable
Daily, weekly, monthly or on a custom calendar, in each person’s time zone.
By event
A new ticket, a delayed shipment, an overdue invoice, a deal changing stage: the event in your system starts the agent.
By threshold
A figure crosses a line you set — margin, stock, response time — and the right people hear about it with the analysis attached.
Delivered anywhere
To the chat, to email, to the messenger channel of the team, as a message or as a document.
A list of runs
Every scheduled agent, its next run, its last result and its cost in one list. Pause, change or stop any of them.
The same rules
A scheduled agent works in a role: it follows the same rules, skills, permissions and limits as a person in that role.
Step 5 · Knowledge
Your documents and your memory.
Answers come from your company’s documents as well as its systems — with the passage shown.
Contracts, policies, manuals, reports, presentations, meeting notes: AI Core searches them by meaning, quotes the passage it relied on and links the source document. Access follows the permissions the documents already have, so each person finds only what they are allowed to open.
Memory keeps what matters between conversations: the decisions taken, the preferences of a team, the context of a customer. The assistant picks up where the last conversation ended instead of starting from zero.
Document search with sources
Every statement taken from a document carries the passage and a link, so it can be checked in seconds.
Permissions respected
Search results follow the access rights of the source: a document a person cannot open never reaches their answer.
Any repository
Shared drives, document management systems, wikis, cloud storage and mail archives are indexed where they live.
Files in the conversation
Attach a spreadsheet, a PDF or a presentation to a message and ask about it directly.
Memory you control
What the assistant remembers is visible, editable and scoped: to a person, a team or the organization.
Tables and charts
Numbers come back as tables and charts ready for a report, not as paragraphs to retype.
Step 6 · Roles
Give every team its own expert.
One platform. A specialist for every seat.
A role turns AI Core into the expert a team needs: the operations chief of staff, the revenue strategist, the finance controller, the people partner, the customer care lead, the market analyst. Pick a role when you start a conversation, and the platform already knows the job.
Every role is built from four parts. The instruction says who the assistant is and what the work is about. The setup switches on the rules it follows, the skills it has and the systems it may use. The first questions are what it offers the moment a chat opens, so nobody faces an empty screen.
Create as many roles as your company has jobs. One person can hold several and switch between them from one conversation to the next; a role that works can be shared with the whole team, so a new colleague starts with the same expert on day one.
6.1 · Start with a role
Choose a role, start from its questions
A new chat begins with a role. The moment it is chosen, the platform offers that role’s first questions — the morning briefing for operations, the forecast for sales — so nobody faces an empty screen and wonders what to type.
Choose a role and start from its suggested questions — nobody faces an empty screen.
6.2 · Instruction
The instruction: who the assistant is
Each role carries its own instruction: who the assistant is in this job, which team it serves, what it should pay attention to and how it should reason. The list on the left shows seven roles kept by one leader, each a different expert.
Seven roles for one leader; each with its own instruction.
6.3 · Setup
The setup: rules, skills and systems
On the setup tab a role is assembled from the company’s libraries: which rules it follows, which skills it has, which connected systems it may use. Tick a box and the role changes from the next answer.
Rules, skills and connections, switched on per role.
6.4 · First questions
The first questions: where the work begins
Every role opens with questions written by the people who know the job. They teach new colleagues what the assistant is good at and turn the best prompts of the team into a button anyone can press.
The questions each role opens with — written by you.
6.5 · All roles
Eight roles, one window
Every role is listed with what it is made of: its main instruction and how many rules, skills and connected systems it carries. Add one, remove one or share one from the bar under the list.
Eight roles, one person — each a different expert.
6.6 · A new role
A new role in a minute
Press plus and a new role appears in the list, ready to be named and given its instruction, rules, skills and connections. Nothing is saved until you press Save.
A new role starts with a name; the rest is added tab by tab.
6.7 · Sharing
Share a role with a colleague, a team or everyone
An administrator hands a finished role to colleagues in one step. They get the whole role — instruction, rules and skills — read-only, and every later edit reaches them from their next answer.
Share a role with the whole organization or with chosen people.
6.8 · Operations Control Tower: instruction
Operations Control Tower: the chief of staff for operations
The main instruction tells the assistant who it is in this role and what the work is about. In Operations Control Tower it becomes the operations chief of staff, who joins shipments, tickets, finance and sales into one picture of the day. The text is yours to edit, formatted or in Markdown.
Operations Control Tower: the operations chief of staff, who joins shipments, tickets, finance and sales into one picture of the day.
6.9 · Finance Desk: instruction
Finance Desk: the finance controller
The main instruction tells the assistant who it is in this role and what the work is about. In Finance Desk it becomes the finance controller's analyst, precise about periods, units and cash. The text is yours to edit, formatted or in Markdown.
Finance Desk: the finance controller's analyst, precise about periods, units and cash.
6.10 · Revenue & Pipeline: setup
Revenue & Pipeline: its rules, skills and systems
Setup is where a role gets its character. Rules are listed in the order the assistant reads them and can be dragged; skills are added from a picker; each connected system is a switch.
What Revenue & Pipeline works with: its rules, its skills and the systems it may use.
6.11 · Customer Care Desk: first questions
Customer Care Desk: the questions it opens with
Up to ten questions a new conversation in this role offers, in the order you set. They turn an empty screen into a starting point and teach a team what the role is for.
The questions Customer Care Desk opens a new chat with.
6.12 · People & Hiring: new chat
A new chat as the people partner
The role is chosen before the first message, and the screen changes with it: the questions offered are that role's own. A conversation keeps the role it started in.
Start a chat as People & Hiring — its first questions are already waiting.
6.13 · Market Intelligence: new chat
A new chat as the market analyst
The role is chosen before the first message, and the screen changes with it: the questions offered are that role's own. A conversation keeps the role it started in.
Start a chat as Market Intelligence — its first questions are already waiting.
Step 7 · Rules
Write your standards once.
Rules make every answer sound and behave like your company.
A rule is a standard the assistant applies to every answer: the tone of voice, the reporting convention, the confidentiality line, the wording your compliance team insists on. "Conclusion first." "Name the source after every figure." "End with one next step." "Never quote a salary."
You write a rule once and tick the roles it belongs to. From the next answer every one of those roles follows it. Change the rule, and the whole company changes with it: no retraining, no memo, no chasing people.
The organization keeps its shared rules; each person adds their own on top. That is how one platform answers the finance desk in the language of finance and the customer in the language of care, while both stay inside the same company policy.
7.1 · Rules library
One library of company standards
All rules live in one library, each with the roles it belongs to: the tone of voice, the reporting convention, the confidentiality line. Write a rule once and tick the roles; every one of them follows it from the next answer.
Write a rule once; every role you tick follows it.
7.2 · A new rule by chat
A new rule, shown before it applies
Asked for a new rule, the Configurator writes it and presents a confirmation card with the full text and the roles it will affect. Approve it, decline it, or decline and explain what to change. Nothing takes effect unseen.
The new rule is shown in full before it takes effect.
7.3 · Name the source system
Rule: Name the source system
A rule is a short instruction the assistant always follows in the roles it belongs to. With this one, every figure says which system it came from. Tick the roles it applies to and it takes effect from the next answer.
"Name the source system" — every figure says which system it came from.
7.4 · The company voice
Rule: The company voice
A rule is a short instruction the assistant always follows in the roles it belongs to. With this one, messages to customers are written in the company's own voice. Tick the roles it applies to and it takes effect from the next answer.
"The Northpeak voice" — messages to customers are written in the company's own voice.
7.5 · Escalate above USD 250K
Rule: Escalate above USD 250K
A rule is a short instruction the assistant always follows in the roles it belongs to. With this one, anything above USD 250,000 at risk is flagged in the first line — a rule created by chat. Tick the roles it applies to and it takes effect from the next answer.
"Escalate above USD 250K" — anything above USD 250,000 at risk is flagged in the first line — a rule created by chat.
Step 8 · Skills
Turn your playbooks into skills.
Your best practice, picked up exactly when the task calls for it.
A skill is one of your company’s playbooks, written down for the assistant: how the daily operations briefing is built, how a quarterly business review is prepared, what the escalation procedure is when a key customer is at risk.
A rule applies always; a skill is taken off the shelf when it is needed. Ask for the morning briefing, and the assistant recognises the task, opens the briefing skill and follows your procedure step by step: which systems to check, in what order, what to compare, how to lay out the result.
Every answer names the skill it used, so people learn the company’s way of working by watching it done. When the procedure changes, you edit the skill once, and tomorrow’s briefing is built the new way.
8.1 · Skills library
A library of the company’s playbooks
Skills are procedures written down for the assistant: how the daily operations briefing is built, how a business review is prepared, what to do in an escalation. Each lists its steps and the roles that carry it.
Your best practice, packaged — the briefing, the business review, the escalation procedure.
8.2 · A skill at work
A skill at work
Asked for the morning briefing, the assistant recognises the task, opens the "Daily operations briefing" skill and follows it: ten calls to three systems, then one answer in the company’s format. The skill it used is named at the top of the answer.
The "Daily operations briefing" skill at work: ten calls to three systems, one answer.
8.3 · Daily operations briefing
Skill: Daily operations briefing
A skill is company know-how the assistant picks up only when the task calls for it: the morning briefing: service, shipments at risk, customers waiting, decisions for today. The "when to use it" line is how the assistant decides; the text is the procedure it then follows.
"Daily operations briefing" — the morning briefing: service, shipments at risk, customers waiting, decisions for today.
8.4 · Daily operations briefing: the procedure
Daily operations briefing: the procedure itself
The skill's text is written the way you would brief a new colleague, with reference material attached where it helps. Below it, tick the roles that may use the skill.
The procedure behind "Daily operations briefing", step by step.
8.5 · Delay escalation playbook: the procedure
Delay escalation playbook: the procedure itself
The skill's text is written the way you would brief a new colleague, with reference material attached where it helps. Below it, tick the roles that may use the skill.
The procedure behind "Delay escalation playbook", step by step.
8.6 · Competitor snapshot: the procedure
Competitor snapshot: the procedure itself
The skill's text is written the way you would brief a new colleague, with reference material attached where it helps. Below it, tick the roles that may use the skill.
The procedure behind "Competitor snapshot", step by step.
Step 9 · Configure by chat
Configure everything by chat.
If you can say it, AI Core can set it up.
There is no admin console to master. The Configurator is an AI administrator you simply talk to. Ask for a new role for the procurement team, a new rule for the finance desk, a new skill, a change to who can reach what. It asks the one question it needs, then shows you exactly what will change.
Nothing happens behind your back. Every change arrives as a confirmation card: approve it, decline it, or decline and explain. Full administration of the platform, by conversation.
9.1 · Conversation
Administration as a conversation
"Create a role for our purchasing team." The Configurator asks the one thing it needs to know — who the role is for — and then builds it: instruction, rules, skills, systems and first questions.
"Create a role for our purchasing team" — the Configurator asks who it is for, then builds it.
9.2 · Configurator plan change
"Put Grace on the Business plan"
Administration is a sentence too. The Configurator finds the person, reads the plans and shows the change for approval. What it may change depends on who is asking.
"Put Grace on the Business plan" — people and plans, managed by chat.
9.3 · Configurator skill card
A whole playbook, written from one request and waiting for approval
Describe what the skill should do and the Configurator writes it — the "when to use it" line and the procedure — and attaches it to the roles you named.
A whole playbook, written from one request and waiting for approval.
9.4 · Configurator edit role
Existing roles are edited the same way
Changing a role does not mean opening it. Ask for the skill to be added and the card shows the role before and after.
Existing roles are edited the same way: say what should change.
9.5 · Configurator own prompt
Even the Configurator's own instruction is yours to change
The Configurator runs on an instruction each person can rewrite and restore to the default. It is the same platform, pointed at itself.
Even the Configurator's own instruction is yours to change.
Step 10 · Voice
Talk to it.
Full voice control. Hands-free, eyes-free.
Speak to AI Core the way you speak to a colleague. It listens, understands natural phrasing, answers out loud and keeps the conversation going — at your desk, in the car, on the warehouse floor.
Studio-grade input with noise suppression, echo cancellation and automatic levels. You choose how it answers: voice, speed and tone.
10.1 · Listening
Speak instead of typing
Press the microphone and talk. A live level shows the platform is listening, and your words become the next question — at the desk, in the car or on the warehouse floor.
Press the microphone and just talk; your words become the next question.
10.2 · Sound window
A sound window for every situation
The sound window gathers the microphone and speech controls in one place: noise suppression for a loud floor, echo cancellation for a meeting room, automatic levels so nobody has to adjust anything.
Studio-grade input: noise suppression, echo cancellation, automatic levels.
10.3 · Voice output
Choose how it answers
Voice output settings pick the voice, its speed and its pitch. Answers can be read aloud automatically, turning a briefing into something you listen to on the way to the meeting.
Choose how it answers: voice, speed and tone.
10.4 · Reading aloud
Listen instead of reading
Any answer can be read aloud, and in speaking mode every answer is. The sentence being said is highlighted and a bar shows how far through the answer the voice is.
Listen instead of reading: the sentence being spoken is highlighted.
Step 11 · Connect
Connect everything your enterprise runs on.
Any system. Any database. Any agent.
AI Core connects to the tools your business already uses through an open standard — CRM, ERP and finance, logistics, HR, help desk, databases, documents, email, your own AI agents. Add a system in minutes; the platform discovers what it can do and puts it to work at once.
You decide what the assistant may read and what it may change, action by action. Built-in capabilities and your own agents sit beside your systems, and the open web is a connection too — under your policy.
11.1 · Connected systems
Your systems, connected and answering
Five business systems of the demo company — CRM, ERP and finance, fleet and shipments, HR directory, support desk — are connected through an open standard. Each shows its status; adding one takes minutes, and the platform discovers its actions by itself.
Five business systems connected and live — added in minutes each.
11.2 · Built-in agents
Built-in agents, ready from day one
Beside your systems sit the platform’s own agents for research, documents and analysis. An administrator decides which of them the organization uses, and your own agents join the same list.
Built-in agents, ready from day one.
11.3 · Google Workspace
Mail, documents, files and spreadsheets
Gmail, Docs, Drive and Sheets connect under your company’s own Google account. The assistant can find a message, read a document, create a file or fill a sheet — within the access each person already has.
Mail, documents, files and spreadsheets — connected under your company’s own account.
11.4 · Web policy
The open web, under your policy
You decide which websites the assistant may read: any site, an approved list for the whole company, or a list each person keeps. A site outside the boundary is simply not opened.
Any site, your approved list, or each person’s own — you set the boundary.
11.5 · Web search
How the platform searches the web
Web search settings choose the provider and how deep and how wide a search goes, balancing thoroughness against speed and cost for the whole organization.
How the platform searches the web: provider, depth and scope.
11.6 · CRM
CRM: checked on your schedule, with the sign-in and the guidance the assistant gets
One card per connected system. This one gives the assistant the sales pipeline, deals and customer accounts. The card sets how often the system is checked, how long to wait for it, how the platform signs in, and what the assistant is told about using it.
CRM: checked on your schedule, with the sign-in and the guidance the assistant gets.
11.7 · Add
Add your own system
Any system that speaks the open MCP standard joins the platform from this form — a database, an internal service, an AI agent of your own. The platform asks it what it can do and lists its actions.
Add your own system: a name, an address, how to sign in.
11.8 · Agent web search
Built in: Web search — searching the web for current information
Built-in agents come with the platform. Each card shows what the assistant is told about the agent and the actions it offers, with the same read and write marks as your own systems.
Built in: Web search — searching the web for current information.
11.9 · Google gmail
Gmail, connected under the company's own Google app
Gmail is built in. Once the company's Google app is set up, each person connects their own account and the assistant works with their mail and files — and only theirs.
Gmail, connected under the company's own Google app.
Step 12 · Govern
Govern it all from one place.
Enterprise control is not a feature. It is the foundation.
People and access: invite your team, assign plans, decide who reaches which system. Budgets you can predict: plans and limits per person and per group of systems. Model freedom: choose the AI model for each kind of task, and bring your own provider keys.
Full visibility: conversations, answers, cost and system usage — by person, by day, by tool. Many organizations, total isolation: run subsidiaries, business units or clients side by side, each with its own people, systems and data.
12.1 · Analytics
Your company’s month of AI at a glance
Analytics show chats, answers, tokens, cost and calls to connected systems, day by day. It is the picture of how the company really uses AI: which teams, which systems, which days.
Your company’s month of AI at a glance.
12.2 · Spend
Spend by person, down to the cent
The spend screen breaks the month’s cost down by person. Finance sees where the budget goes, managers see who gets the most from the platform, and nobody is surprised by the invoice.
Spend by person, down to the cent.
12.3 · People
Everyone in one list
All the people of the organization are in one list with their plan and status. Select a person to see and change everything about their work with the platform.
Everyone in your organization, in one list.
12.4 · Plan and limit
A plan and a limit for each person
A person’s profile sets the tariff plan and the message limit. Changing them takes seconds and applies at once.
Each person’s plan and limits, set in seconds.
12.5 · Access
Access to systems, person by person
For each person you tick the systems they may reach. The sales team sees the CRM, finance sees the ERP, and an intern sees neither until someone decides otherwise.
Access to each system, person by person.
12.6 · Personal spend
What one person used and what it cost
Each person has their own spend history: chats, answers, tokens and cost over time. It answers the question "is this seat worth it?" with data.
What each person used and what it cost.
12.7 · Plans
Plans that make the budget predictable
Tariff plans define how much a person may use: messages, tokens, spend. Assign a plan and the ceiling is set before the first question is asked.
Plans that define how much each person may use.
12.8 · Limit groups
Separate ceilings for separate systems
Limit groups put a ceiling on categories of connected systems. An expensive data source can have its own allowance without restricting everything else.
Separate ceilings for separate kinds of systems.
12.9 · Models
The right model for each kind of task
You choose which AI model answers in chats, which one serves the administrator’s chat and which writes titles and suggestions — and you can run the organization on your own provider keys.
The right model for each job — and your own keys, if you prefer.
12.10 · Organizations
Many organizations, total isolation
Subsidiaries, business units or clients run side by side on one platform. Each organization has its own people, systems, data and spend, and nothing crosses the line between them.
One platform, many organizations — each with its own people, systems and data.
12.11 · Invitation new
People join by invitation
There is no public sign-up. A person gets in because an administrator invited them, and the invitation says with which role.
People join by invitation: an address and a role.
12.12 · Plan enterprise
The Enterprise plan
A plan decides how much a person may use: an overall ceiling, a separate one for the assistant's own reasoning, and one per group of connected systems, so a single expensive system cannot eat the whole allowance.
The Enterprise plan: an overall limit, one for the assistant, one per group of systems.
12.13 · Analytics tools
Which systems your company actually asks
Below the totals, every connected system: how often the assistant called it, how often a call failed and how long it took. Open a row for its tools, one by one.
Which systems your company actually asks — and how each performs.
Step 13 · Safety
Safe by design.
The platform says no when it should, and says so plainly.
Control is not a settings page somebody has to remember. It shows in the conversation itself: a website outside your policy is not opened, a system a person has no access to is not queried, and when the data does not hold the answer the assistant says exactly that instead of guessing.
Every refusal is visible and explained, so people trust the answers they do get — and your security team can see the boundaries working.
13.1 · Site not allowed
A site outside the allowed list is not read
The organization decides which websites may be read. Asked for a page outside the list, the platform refuses the call, shows the refusal in red above the answer and tells the person what it can do instead.
A site outside the allowed list is not read — and the answer says why.
13.2 · Data not there
When the data is not there, it says so
Asked for a discount and a shipment count that no system holds, the assistant reports what the records do say, states plainly that it cannot verify the rest and names who can.
When the data is not there, it says so — and does not invent it.
13.3 · System not granted
No access means no answer from that system
This person's administrator switched the finance system off for them. Their assistant does not see it at all, and says so instead of guessing at the company's cash.
No access means no answer from that system — for that person.
Step 14 · Everywhere
Everywhere you work.
Your desk, your phone, your way.
One account, one conversation, every screen. Start at your desk, continue on the way to the meeting: the full platform runs on a phone.
Your profile and preferences travel with you: light, blue or dark, and the interface size that suits your screen.
14.1 · Phone: chat
The same answer on a phone
A conversation started at the desk opens on a phone exactly where it stopped: the same answer, the same calls behind it, the same next steps.
The same answer, the same next steps — wherever you are.
14.2 · Themes
Three themes and the size that suits you
Appearance settings switch between light, blue and dark and set the interface size — for a large monitor, a laptop or tired eyes late in the evening.
Light, blue or dark — and the interface size that suits your screen.
14.3 · Dark theme
The dark theme
And in the dark theme, for control rooms, night shifts and people who simply prefer it.
And in the dark theme.
14.4 · Phone roles
Roles on a phone
Roles, settings and administration open as full-screen windows on a phone and close with the Back gesture.
Roles on a phone: every window goes full screen.
14.5 · Phone configurator
Approve a change from your phone
The Configurator and its confirmation cards work the same on a phone — a change proposed at the desk can be approved on the way home.
Approve a change from your phone.
14.6 · Tablet chat
On a tablet
Below laptop width the history folds into a menu and the conversation gets every pixel.
On a tablet: the conversation takes the whole screen.
Integration first: preserve and extend
AI Core complements the systems you already run. You keep your systems of record and add an intelligent interaction and automation layer above them. Connected agents retrieve information, prepare reports, update records and initiate approved actions.
| System category | What connects |
|---|---|
| CRM | HubSpot, SugarCRM and other customer management systems |
| ERP and finance | Financial, accounting and operational applications |
| Enterprise data | SQL databases, data warehouses, documents and knowledge repositories |
| AI solutions | Internally developed and third-party agents |
| Business applications | Internal tools, APIs, reporting and workflow platforms |
| Workspace | Gmail, Docs, Drive and Sheets under your company’s own account |
| The open web | Live search and reading, inside the policy you set |
Enterprise security and data governance
Deployment is tailored to your security policies, data residency needs and infrastructure strategy. Three architectures give you the choice.
Cloud AI
External AI models combined with enterprise orchestration, controlled access and data safeguards.
Private AI
Models and processing hosted in a dedicated environment that your enterprise controls.
Hybrid AI
Internal and external models together, with selective routing and anonymization of sensitive inputs.
- ✓ Role-based permissions
- ✓ Access to tools and systems, person by person
- ✓ Read and write permissions for every action
- ✓ Approval checkpoints for sensitive actions
- ✓ A complete trail of every agent call
- ✓ Data-flow restrictions and web policy
- ✓ Spending plans and limits
- ✓ Isolated organizations
What your IT and security teams will ask
The answers to the first questions of a large company’s architecture and security review.
| Sign-in | Single sign-on through your identity provider: SAML 2.0 and OpenID Connect (Microsoft Entra ID, Okta, Google Workspace, Keycloak). |
| People and groups | Automatic provisioning by SCIM and synchronization of groups from the corporate directory; access to roles, systems and plans is granted by group. |
| Rights inside connected systems | Each person works in CRM, ERP and other systems under their own account, so the assistant sees exactly what that person may see. |
| Approval of actions | Reads run freely. For each connected system the administrator marks which actions change data and which of them wait for a person. The platform enforces it, whatever a prompt says. |
| Audit | A complete trail of every conversation, call and configuration change, with export to your SIEM. |
| Configuration control | Version history for every role, rule and skill: see the difference, test the change on a set of control questions, apply it or roll it back. |
| Data location and retention | Data stays in the region and infrastructure you choose; retention periods are set per organization. |
| What reaches the model | Private and hybrid deployment with routing by sensitivity and anonymization of personal and confidential data before it leaves your perimeter. |
| Budgets | Plans, limits per person and per group of systems, alerts before a ceiling is reached. |
Business intelligence and operational visibility
A shared AI layer shows how work really moves through the organization. Answers arrive as structured text, tables and charts, and anyone can ask the business a question in plain language: "Where are our longest processing delays?" or "Which workflows are the strongest candidates for automation?"
Process performance
Reveal delays, handoffs and recurring bottlenecks.
Resource utilization
Understand workload and operational capacity.
Automation opportunities
Prioritize repetitive, high-volume activities.
System efficiency
Identify overlapping functionality and fragmented workflows.
Executive reporting
Assemble insights from every approved data source.
What each function uses it for
Sales and CRM
Customer information retrieval, pipeline reporting, forecasts, meeting briefs, CRM updates, administrative automation.
Finance and accounting
Data collection and validation, receivables and cash outlook, reporting, preparation of files for accounting systems.
Business operations
Cross-team task coordination, document processing, daily briefings, multi-stage workflows.
Customer care
Ticket triage, replies drafted from live order and shipment data, escalation by your procedure.
People and HR
Headcount and workload questions, policy answers, onboarding with the right role from day one.
Management
KPI monitoring, consolidated reporting, operational insights, improvement analysis.
From a first pilot to the whole enterprise
A phased adoption proves the value on targeted workflows before it scales. The platform is tailored at every step: scope, integrations, security controls and rollout follow your requirements.
- 1
Assessment and prioritization
Identify high-value workflows, dependencies, success metrics and enterprise requirements.
- 2
Focused pilot
Validate selected scenarios with a limited user group and measurable outcomes.
- 3
Department expansion
Connect additional teams, agents and systems based on demonstrated results.
- 4
Enterprise-wide adoption
Extend centralized governance and orchestration across business functions.
- 5
Continuous optimization
Measure outcomes, refine workflows and expand automation.
Less repetitive manual work
Less fragmented AI usage
More output per team
Lean and scalable operations
From AI automation to an intelligent enterprise
The goal is an operating environment where AI is part of how work is requested, coordinated, completed and improved — not a collection of disconnected assistants. People increasingly state the outcome they want instead of operating each application; your systems of record stay authoritative.
Connected AI ecosystem
A governed foundation linking people, agents, enterprise applications and business information.
Intelligent process automation
Coordinated execution of multi-system workflows with human oversight where it matters.
AI-driven business operations
Routine work is handled by agents while employees focus on judgment, exceptions and higher-value decisions.
Adaptive intelligent enterprise
Operational insight keeps pointing to the next workflow worth refining or automating.
Stop collecting AI tools. Start running on one platform.
Alltegrio AI Core is the private enterprise platform that unites every AI solution your company needs, connects it to every system you already have, and puts it all under your control. We set it up for your organization, connect your systems and write the first roles, rules and skills together with your teams.
The screens on this page show a demo company with invented data.
Talk to our teamQuestions about AI Core
Can agents work on a schedule, without being asked? +
Yes. Any request can run by timetable, by an event in a connected system or when a figure crosses a threshold, and the result is delivered to the chat, by email or to the team messenger.
Does it support single sign-on and our corporate directory? +
Yes. Sign-in works through your identity provider by SAML 2.0 or OpenID Connect, people are provisioned by SCIM, and access is granted by directory group.
Where does the platform run and where is our data? +
On a server dedicated to your company — in your cloud, your data centre or ours — in the region you choose. Cloud, private and hybrid model deployments are available, with anonymization of sensitive data.
Can you build agents for our processes? +
Yes. Building custom agents is our main work: we design them around your process, connect them to your systems and run them on the platform under your rules.
What is Alltegrio AI Core? +
A closed, private enterprise AI platform that brings all your AI solutions under one roof, puts one intelligent assistant on top of all your business systems and lets every employee work through conversation.
How is it different from a chatbot? +
A chatbot talks. AI Core works: it orchestrates specialised agents across your live systems, follows your rules, applies your playbooks and can act — under your permissions.
Do we need to rebuild our systems? +
No. AI Core connects to what you already run through an open standard. New systems join in minutes.
How do we configure it? +
You talk to it. The Configurator creates roles, rules, skills and access from plain language and shows every change for approval.
Can anyone sign up? +
No. AI Core is a closed platform: it is set up for your organization and your people join by invitation.
What can it integrate with? +
Anything your enterprise runs on — CRM, ERP, databases, help desks, HR systems, documents, email, your own AI agents and the open web.
Is it safe? +
Roles, per-person access, read and write permissions, spending limits, isolated organizations and a complete trail of every agent call. Control is the foundation, not a feature.
Can we use our own AI models? +
Yes. Choose the model per task and bring your own provider keys.
Does it work by voice and on mobile? +
Yes — voice conversation, and the same conversations on every screen.
Can it be adapted to our company? +
Yes. The platform is set up for each organization: we connect your systems, write the first roles, rules and skills with your teams, and add the capabilities your processes need.
Let's talk about your project
We'll help you harness AI and ML technologies to create innovative software solutions