Skip to content
en vi

AI Transformation

Agents that do the work.

We build AI systems that run inside the business. Each agent signs in as the person it works for — it can open exactly what they can open, nothing else, and every action it takes is recorded against their name.

Running in production right now writing up the meeting, answering from the company handbook, building the sprint report, translating the call, live, filing the bug with steps to reproduce.

Talk to an engineer See how it works
Since 2022
Building AI products for enterprises
Vortex
Agentic platform, running today
On-premise
Or a custom API, when the business needs it

Works with what you already run

  • Microsoft 365
  • Google Meet
  • Zoom
  • Microsoft Teams
  • Slack
  • GitHub
  • Claude
  • ChatGPT

Rules we do not bend

Trustworthy because it knows where to stop.

  1. An agent sees what you see — and nothing more

    When an agent opens your mail, your calendar or your files, it is signed in as you, carrying your permissions. If you cannot open the Leadership folder, neither can it. There is no service account behind the scenes holding keys to everything.

  2. Any access you give can be taken back

    Access is given for a specific set of things and a specific length of time. Take a permission away from the person and the agent loses it on its very next request — nothing to redeploy, no ticket to raise, no waiting for an overnight job to catch up.

  3. Built for production, not for the demo

    We work against your real data, your real sign-in and your real traffic from the first week. The measure of the work is whether it still runs in month six, not whether it presents well in month one.

  4. One team, from design to operations

    The people who design the system are the people who write it and keep it running. Nothing is subcontracted, and nothing is handed to a delivery partner once it launches.

A Nguyễn Văn A Head of Operations signed in as themselves
  • Mail · Microsoft 365
  • Calendar
  • Wiki — Engineering
  • Project management
  • Wiki — Leadership not their access
  • Support not their access
Operations agent acting as Nguyễn Văn A same list
  • Mail · Microsoft 365
  • Calendar
  • Wiki — Engineering
  • Project management
  • Wiki — Leadership not their access
  • Support not their access
The agent inherits the list — it never extends it. Take a line away from the person and it is gone here on the next request.

The hard part was never the model. It was making sure an agent can open exactly the files its owner can open — and nothing else, on every single request, for years.

What it looks like

Work that nobody has to do by hand any more.

  1. Monday, 09:02

    The meeting wrote itself up

    A bot joined the call on Meet, Zoom or Teams and captured it — including Vietnamese, on a model tuned for it. By the time people are back at their desks, the transcript, the decisions and the action items are filed where the team already looks, and the tasks exist in the project.

    • Meetings
    • Wiki
    • Project Manager

  2. Any afternoon

    Ask the company, not the internet

    “What did we agree with this supplier about penalties?” The answer comes back with its sources, drawn only from the documents that person is already cleared to read. Someone without that clearance asks the same question and does not get the answer.

    • Knowledge base
    • Wiki

  3. A partner call

    Everyone speaks their own language

    Three people, three languages, one room joined from a QR code. Each hears the others in their own — speech straight to speech, with subtitles labelled by speaker underneath. Guests need no account; the host can require one.

    • Live Translation
    • Meetings

  4. End of sprint

    A report nobody had to assemble

    The sprint report is built from the work items themselves, instead of from chasing six people for status. It arrives as a deck in the house style, with every number pulled from the system that owns it.

    • Project Manager
    • Presentation
    • Office

The platform

One connector. Everything behind it.

Vortex puts a single address in front of every system the company already runs. An assistant connects to it once, the person signs in once, and from then on it reaches those systems as that person — never as a shared service account, never with more access than they have.

Microsoft 365 Wiki Knowledge base Meetings Project management Presentation Office Design Support /spaxe-mcp/mcp
One address in the middle, every system around it. What comes back is whatever the person who asked is allowed to see.
/spaxe-mcp/mcp the only address anything needs a sample of what it reaches
  • Microsoft 365 Mail, calendar, tasks, files, chat
  • Wiki Read, write, search and translate internal pages
  • Knowledge base Question answering over document stores
  • Meetings Transcripts, recaps, decisions, action items
  • Project management Projects, sprints, work items
  • Presentation Build and edit decks
  • Office Word, Excel, PowerPoint
  • Design Read, create and share design files
  • Support Customer conversations and contacts
  • The assistant signs itself in the first time you point it here. There is no key to copy and nothing to configure for each tool.
  • Works with any assistant that speaks MCP, the open standard for connecting assistants to tools. Claude and ChatGPT both do — and when needed we build on-premise AI or custom APIs to order.
  • Set it up either way: install a one-click bundle, or paste one web address.

Available today

One identity across every application.

Each is a real product with real users, not a demo. They share one login, one permission model and one audit trail — the part that takes years to build and cannot be bought in.

Knowledge

  • Wiki Real-time co-editing, bilingual pages, per-folder permissions the agent obeys.
  • Knowledge base Retrieval over handbooks, with access enforced at the dataset, not the prompt.
  • Office Read and write Word, Excel and PowerPoint in place.
  • Presentation Decks built from the organization’s own material, in its own template.

Collaboration

  • Meetings A bot joins Meet, Zoom or Teams and files the transcript, decisions and tasks.
  • Live Translation Everyone speaks their own language — speech to speech, subtitles labelled by speaker.
  • Support Customer conversations answered from the knowledge the team already maintains.
  • Design Design files readable and shareable under the same identity as everything else.

Delivery

  • Project Manager Run projects from chat — sprints, work packages, reports that assemble themselves.
  • Spec-Driven Development Software built through a governed pipeline, every agent pinned to a spec version.
  • Automation Multi-step flows in plain language, with schedules, webhooks and connectors.
  • BugBase In-app bug reports with AI triage and enough context to reproduce first time.

Platform & people

  • AgentOps The console for the agent fleet: model, workflow and skills per role, audited.
  • Admin Identity, organizations, RBAC, the app and tool registries, and billing.
  • Wallet Employee benefits in one place — vouchers redeemed at the counter by rotating QR.
  • Rewards Campaigns drawn against a real budget, where whoever created one cannot approve it.

Start here

Tell us what you are trying to automate.

A short description of the process and who owns it is enough to begin. We reply within one business day.

Based in
Hồ Chí Minh City, Vietnam