An AI agent for the whole team, without handing company data to a hosted platform. One self-hosted VAF serves everyone, what the agent knows stays in the building, and you pay for the machine rather than for every head.
An AI agent is software that does not just answer questions, it carries out tasks. Given a goal, it plans the steps and works through them with tools: the web, files, documents, code, checking its own results as it goes. Where a chatbot gives you text, an agent hands back finished work.
In a company, that covers the recurring work that quietly fills the week: pulling figures into a morning summary, researching a supplier, reading and writing documents, preparing the numbers before a meeting. A person states the goal and reviews the result; the agent does the steps in between, on a schedule if you want it to.
VAF runs all of this on your infrastructure, and the code is open: your team can read exactly what the agent does. No vendor can switch it off, and no subscription sits between you and it. That is the difference between renting an agent and owning one.
Which path fits depends on who works with the agent: a whole team, individual desks, or your own software.
One machine in the building, and everyone works with the same agent. Accounts and memory stay separate, and it is set up only once.
The agent sits in its own window on the machine and needs neither a server nor a network. Every workstation stands on its own.
The agent moves into your own software, with your tools and under your name.
The recurring work that eats a team’s week, phrased the way you would hand it to a person.
It reads the exports your tools produce, spreadsheets, CSV, PDF invoices, cross-checks them against the rules you give it, and flags what looks off for a person to review. The judgment call stays human; the tedious first pass does not.
It reads the exports your tools produce, spreadsheets, CSV, PDF invoices, cross-checks them against the rules you give it, and flags what looks off for a person to review. The judgment call stays human; the tedious first pass does not.
It answers recurring internal questions from what it remembers about your company, and drafts replies where your customers actually write: mail, WhatsApp or Telegram, for a person to approve. The team stops retyping the same explanation for the tenth time.
Give it the supplier quotes and price lists as they arrive, PDF or spreadsheet, and it compares terms, lead times and totals, then summarizes the trade-offs and flags what is missing, like absent liability coverage, before you sign.
Its research sub-agent searches the web, reads ten-plus sources properly, and returns a structured report with citations: for a campaign, a market question, or the article your team needs to write. You state the question, not the process.
New people ask it what they would ask the person at the next desk: where things live, how the project is set up, who owns what. Because it remembers your projects, tools and conventions, the answers are about your company, not generic.
It checks contracts and internal documents against the checklists you define, and on a schedule it watches the pages where rules and documentation are published, reporting what changed. A first pass that never gets bored, with findings a person verifies.
On a schedule it checks competitor pages and price lists, then runs the comparisons in sandboxed code and hands you the spreadsheet: what moved, where you stand, what a change would do to your margin.
Point it at the folder everyone dreads, locally or in Google Drive and OneDrive: it extracts the data from PDFs, renames and sorts files by your convention, and keeps registers up to date, so finding a document stops being an archaeology project.
It works through reviews, survey answers and the mail inbox itself, clusters what people are actually saying, and drafts responses for a person to approve. You see the themes instead of scrolling the pile.
These roles are not fixed products. They are the same agent with over a hundred tools: web, files, documents, code, mail, calendar, messaging, memory and schedules. Describe a role in plain words and the agent takes it on. Content researcher plus records keeper plus a Friday schedule is your weekly market report, assembled in a sentence.
The difference from a tool: a task does not wait for somebody to operate it.
You choose the model, the hardware and where the data lives; VAF brings the agent, the tools and the memory. In local mode, prompts, files and the agent’s memory stay on hardware you own. When you want stronger models, the hosted Veyllo API is one key and a pay-as-you-go bill, and speech-to-text through it is processed in the EU. Veyllo is a German GmbH, and the privacy policy discloses every transfer of data we make.
From first install to day-to-day operation, we take responsibility for the system, on request as an ongoing engagement with a fixed service agreement and a named contact.
We install VAF on your infrastructure, set up the host, network sharing and accounts, and hand it over running.
We keep your install current and healthy: updates, monitoring the host, and fixing what breaks before it matters.
We adapt VAF to your environment and workflows, from configuration and providers to how the agent fits your processes.
We build new tools for your internal systems, so the agent can work with the software your company actually runs on.
Yes. Install VAF on one server (or even a spare desktop), turn on network sharing, and it is reachable across your network over HTTPS. Each person signs in with their own account and gets isolated memory and tasks, so one host serves everyone without a per-seat subscription.
Not in local mode. VAF can run entirely on your own hardware with local models, in which case prompts, files and memory stay inside your network. If you choose to connect the hosted Veyllo API or another provider for stronger models, those requests go to that provider; speech-to-text through the Veyllo API is processed in the EU, as documented in our privacy policy.
Nothing per seat. VAF is open source under the AGPL-3.0, and running it internally, even commercially, is free. A paid commercial license exists for companies that want to embed VAF in a closed-source product or offer it as a proprietary SaaS. Model usage through the Veyllo API is billed pay as you go.
Pick a machine, run the installer script for your platform (it sets up the container runtime, Python and Node), and share the host with your team. People connect from their browsers over your network, or install the desktop app on individual machines. If you get stuck, write to support@veyllo.app.
Yes. Beyond the open-source software itself, we offer hands-on services in the classic open-source model: installation on your infrastructure, ongoing care and updates, adapting VAF to your environment, and building custom tools for your internal systems. Write to support@veyllo.app with what you need.
Rollout, services, licensing or the API: your message goes straight to the people who build VAF, not to a sales layer.
