AI Corporate Assistant
Your company's AI assistant.
You speak. It gets it done. And everything stays on your server.
The problem
Your team loses hours every week to work that isn't the work.
Reporting and oversight
Building the weekly report, knowing who is working on what, spotting blockers.
Coordinating and following up
Reminders, pending tasks, preparing every meeting.
Finding context
“What did we decide with this client?” “How is the project going?”
Repetitive work
Quotes, meeting minutes, the email triage every morning.
The trade-off
Until now, AI made you choose.
Cloud AI
Capable and modern, but your data, your clients and your decisions live on someone else's server.
- Capability: yes
- Privacy: no
Local AI
Private and yours, but it only chats: it doesn't do real work or know your company.
- Capability: no
- Privacy: yes
No one gave you both.
The solution
The capability of a modern agent. Full sovereignty over your data.
Hands-free voice
Say “Andri” and ask for what you need. Without touching anything, from any device.
Executes, not just answers
It drafts quotes, runs reports, prepares meetings and saves files. Anything with consequences, you approve.
Everything runs on your server
The model, the voice and the memory run on your infrastructure, not in an AI vendor's cloud.
Your company's memory
It learns from every conversation and remembers clients, decisions and processes.
Hands-free voice
There are no menus to learn.
Say “Andri” and ask for what you need, in your own words. It answers in a natural voice, and it is designed to do so in under two seconds.
Try an order
Voice interaction is available in Spanish today. The orders on this page are shown translated.
“Andri, what should I work on today?”
using GitLab…
You have 12 issues assigned. I would start with three:
· Error 500 when exporting PDF reports: it is blocking support.
· Optimize loading of the client dashboard: due on Thursday.
· The payment gateway merge request has gone three days without review.
Shall I read you the details of any of them?
“Andri, what should I work on today?”
using GitLab…
You have 12 issues assigned. I would start with three:
· Error 500 when exporting PDF reports: it is blocking support.
· Optimize loading of the client dashboard: due on Thursday.
· The payment gateway merge request has gone three days without review.
Shall I read you the details of any of them?
“Andri, draft a quote for ACME: 40 hours of development at $35 and 10 of QA at $25”
using doc_generate_quote…
Done. I generated quote COT-2026-041 for ACME:
· Development: 40 h × $35 = $1,400
· QA: 10 h × $25 = $250
· Total: $1,650
I left it in your documents. Shall I send it once you've reviewed it?
“Andri, run the weekly report”
using playbook_run…
I launched the Weekly team report playbook as a background mission.
You can close the app: I'll let you know when it's ready.
Mission completed · Weekly team report
“Andri, email the quote to ACME”
using Outlook…
I prepared the email to ACME with the quote attached. I need your confirmation to send it.
Needs your confirmation
Outlook · send email
- to
- compras@acme.example
- subject
- Quote COT-2026-041
- attachment
- COT-2026-041.md
Sent. It has been recorded in the audit log.
Cancelled. Nothing was sent.
Background missions
Delegate the goal. Close the laptop. Andri tells you when it's done.
- You ask by voice
- It works alone, on your server
- You carry on
- It notifies you with the result
- Up to four missions at a time.
- Missions never run sensitive actions: there would be no one to confirm them.
Control and audit
Nothing with consequences happens without a person's approval.
- Andri proposes
- You confirm
- It runs
- It's logged
Every sensitive action—sending an email, creating an issue, writing a file—is shown in full detail before it runs. Afterwards it is recorded in the audit log: who, what, when and with what result.
If no one confirms within 15 minutes, the action expires.
every action is recorded in the audit log
Governance, built in.
100% self-hosted core
The model, the voice, the memory and the database run on your infrastructure.
Optional connectors
External services such as Outlook connect only if you decide so. They ship turned off.
Encrypted credentials
Tokens and credentials are stored encrypted with AES-256.
Per-user identity
Every tool acts with the account of whoever asks, with admin and member roles.
Private memory
Personal memory is isolated by design. Everyone can review and delete their own.
Files with limits
On your computer, Andri only reaches Documents, Downloads and Desktop. Writing asks for confirmation.
Made to fit
It works your way, not the other way round.
Playbooks
Your workflows, not generic ones.
Your company defines its playbooks once. Andri runs them by voice, with a tap or on a schedule. Four ship by default; your team edits them and creates its own.
Daily review
Technical
every day · 08:00
Weekly team report
Management
Mondays · 08:00
Email summary
Management
every 2 hours
Prepare meeting
Management
by voice
“Andri, run the weekly report”
Example schedules: each team sets its own.
Company memory
Every conversation makes it more yours.
Personal memory
Private to each user. No one else can see it.
Company memory
Clients, decisions, processes. The organization's shared knowledge.
AI models come and go. The accumulated knowledge stays, and it's yours.
Multi-device
The same assistant, on all your devices.
In the browser on your PC or phone, and in a desktop app with a floating window.
CtrlShiftAsummons Andri from any application
Microsoft 365 and your tools
It works inside the apps you already use.
It reads the mail, summarizes the Teams thread, updates the sheet and drafts the document. You just talk. Anything with consequences, you approve first.
- O
Outlook
Reads, summarizes and drafts mail; checks and books your calendar.
- T
Teams
Summarizes threads and channels, and posts what you approve.
- X
Excel
Reads your sheets and updates rows, totals and trackers.
- W
Word
Drafts minutes, proposals and reports from what happened.
Every connection uses each user's own identity, through the open MCP standard, and inherits voice, confirmation of sensitive actions and the audit log.
Integrations
It connects to what you already use, with no custom development.
Andri
the hub for your tools
GitLab
issues · merge requests · projects
Outlook
mail · calendar
Your computer
files · applications
And whatever comes next
open MCP standard
Every new connection inherits voice, confirmation of sensitive actions and the audit log.
Who it's for
From the first team to the whole organization.
Startups
Start small, without paying per seat.
- A pilot starts on an entry-level server for 3 to 5 people at a time.
- No per-user licenses: hiring doesn't make your AI more expensive.
- The whole system comes up with Docker Compose.
Large corporations
Governance and sovereignty, by team or by division.
- Deployed per team or division, inside your own infrastructure.
- Roles, human approval and an audit log of every action.
- Scales with hardware: from 10–20 to more than 25 people at a time per server.
Why Andri
What a cloud agent can't give you.
| Capability | Cloud agents | Andri |
|---|---|---|
| Your data stays in your company | No | Yes |
| Voice as the primary interface | No | Yes |
| Your organization's own workflows | generic plugins | Yes |
| Fixed cost, no paying per token or per user | No | Yes |
Cost model
Unlimited use. Predictable cost.
$0
per token · per user · per query
- Runs on your own hardware
- The investment is fixed and defined in your proposal
Frequently asked questions
What people usually ask us.
Does my data leave the company?
Not to an AI vendor. The model, the voice, the memory and the database run on your infrastructure. Connectors to external services, such as Outlook, are optional and ship turned off: you switch them on.
Can Andri act without permission?
No. Every action with consequences is proposed on screen with its details and only runs if a person confirms it; if no one does within 15 minutes, it expires. Background missions never run sensitive actions.
What hardware do I need?
An application server and a GPU server. The size depends on how many people use it at once: 3 to 5 on the entry tier, 10 to 20 on the recommended tier and more than 25 on the top tier. We size it with you in the proposal.
What do I need to get started?
Today, sign-in uses your company's GitLab account. The complete system is deployed with Docker Compose on your servers.
Which tools does it integrate with?
GitLab, Outlook (mail and calendar) and the files on your computer through the desktop app. Other tools are added by configuration through the open MCP standard, with no custom development.
Which AI models does it use? Can I change them?
Open models served on your own hardware through standard APIs. Changing the model means changing an address in the configuration; your company's memory and workflows are preserved.
Which languages does voice work in?
Spanish, today. Andri listens and answers in Spanish; the examples on this page are shown translated.
Which devices does it work on?
In the browser on your PC or phone, and in a desktop app with a floating window and a global shortcut (Ctrl + Shift + A). Updates reach every device on their own.
How much does it cost?
There is no cost per token, per user or per query. The investment is fixed and defined in a proposal based on your team and your hardware.
Listens. Remembers. Executes.
And everything it knows, and everything it does, belongs to your company.