Agentix AI Managed Services
Product Commitments
The practical promises that shape how AgentixAI designs, sells, and operates managed AI services for small businesses.
Current service guide · October 2026
Our service
Agentix AI Managed Services is a monthly managed AI operations service for small businesses, led by Virgil, your named human guide. Veylo is the system that runs the automated back-office work. Virgil oversees it, reviews exceptions, and answers for the service.
What every plan includes
Every plan begins with bookkeeping workflow support. The exact scope is agreed with each client and may include work around receipts, invoices, transaction organization, financial task coordination, and related follow-up. Automation is activated only after the relevant workflow is ready to support.
Human oversight and client control
AgentixAI is human-led. Virgil provides guidance and manages the service. Clients remain responsible for their business decisions and approvals. Every activated Agentix workflow is designed with the appropriate approval and audit rules for its scope. AgentixAI does not make consequential decisions, send commitments, move money, or take other material actions on a client's behalf without the client's specific authorization and an approved, supportable process.
Payment approvals and customer messaging
Every payment needs the owner’s approval. Customer messaging follows a written policy the owner approves. FrontDesk can send routine replies within that policy automatically, while important, sensitive, uncertain, or out-of-scope messages are escalated to a human. Owners can change or revoke their messaging policy. Approvals and message actions are logged.
Systems and integrations
We assess a client's existing business systems and may design support around approved accounting, email, scheduling, point-of-sale, phone, and AI-workspace tools. A system discussed during a review is not a promise that it is connected or available as a live service.
A feature or integration becomes active only after it is connected, tested, permissioned, monitored, and supportable for that client.
Plans and service scope
Books & Back Office covers bookkeeping and agreed back-office work. Books & Back Office Plus adds FrontDesk email handling, customer follow-up, and a broader agreed scope of daily admin. Custom Managed includes the Plus scope with daily human support, quote preparation, advanced reporting, and specialized workflows. Monthly fees and setup charges are confirmed in writing before work begins. Priced by the shape of your business, not metered by its success.
Privacy and secure access
We use the minimum appropriate access for an agreed workflow. Clients should not send passwords, banking information, financial-account credentials, or sensitive bookkeeping data through this website. Connected services must be authorized through the appropriate account, workspace, integration, or API—not by sharing personal passwords. This marketing-site notice does not replace the client-data privacy policy that will accompany the operational application when it is live.
Before we're connected
Most of this page describes how we work once your systems are connected. There's a stage before that, and it deserves its own answer.
Before an engagement begins, we usually look at something you already have — most often a stretch of your email — to establish a baseline.
What we ask for. Wherever we can, an authorization you grant and can withdraw: a read-only connection through your provider's own consent screen, or a forwarding address you point at us and switch off whenever you like. We never ask for a personal account password, at this stage or any other.
Where an export is the only option. Some mail providers offer no read-only authorization to grant. When that's the case we'll say so plainly and give you the choice, rather than presenting an export as the only way. You can always skip the historical look and let us start from the day you say go.
What we put in writing first. Before you send or authorize anything, you get a signed agreement covering it: what the data will be used for, who can see it, where it's held, whether AI processing is involved and where, and the date it gets destroyed.
What happens to it afterwards. Evaluation data is destroyed on a fixed schedule, and sooner if you ask. What survives is the finding itself, which is yours, and a dated record that the destruction happened.
The Data Bucket: where your records live
What goes in
Emails and attachments as received; texts and receipt photos; voice-call recordings and transcripts where authorized and captured; uploaded documents of any file type; and records from connected systems.
We store the original, not our interpretation of it. Extraction reads from the bucket; it does not decide what gets in. If a file cannot be machine-read, it is kept with a note. A correction is a new record, never an overwrite of the original.
Every client we serve gets a Data Bucket. It is the single place where your original records land exactly as we received them: bank statements, vendor bills, invoices, receipts, spreadsheets, and the photo of a receipt someone texted in. Nothing is retyped, summarized, or “cleaned up” before it is stored. The original is the original, and it stays that way.
We built the Data Bucket on Cloudflare R2. We chose it for reasons that come down to security, privacy, and control.
Your documents are stored as opaque, encrypted objects.
R2 encrypts everything at rest and in transit. The storage layer never needs to understand what a file is; it just holds the bytes. The system that reads and interprets your records is separate, and it only fetches a document once it already knows exactly which one it needs.
Nothing can be quietly altered.
Every file is named by its own cryptographic fingerprint. If a single byte of a document changed, the fingerprint would no longer match, and we would know. The bucket is also configured so stored originals cannot be overwritten or deleted through the normal working path. The credentials our software runs with can write and read. They cannot erase.
Erasing is a separate, deliberate act.
When a client leaves, or asks us to remove something, deletion happens through a distinct path with its own credentials, under a human's approval, and it is logged. That separation is on purpose: the everyday system physically cannot destroy your records, and the removal path cannot be triggered by accident.
One bucket per client. Your data never mixes with anyone else's.
Each client's records are scoped and isolated. Our people see only what they are authorized to see, and sensitive documents can be flagged so they are withheld even from tools that would otherwise search them.
We can tell you plainly where your data is.
R2 is a standard object store run by a major infrastructure provider, in a defined region, under a data processing agreement. That is a sentence you and your accountant can act on. We looked hard at decentralized “web3” storage and decided against it for one reason above all others: on those networks, you cannot guarantee a document is ever truly gone. We are not willing to make a promise about your data we cannot keep.
An optional scheduled copy of your records
For an additional fee, a full client copy of your Data Bucket can be scoped for delivery every 3, 6, or 12 months. It includes raw records and a readable index organized by period: date, source, sender, filename, and what each record was filed as.
The delivery method, schedule, and fee are confirmed before this option is added to your service.
The standard we hold ourselves to
Never represent a feature as live until it is ready to support.
We do not promise an automation, integration, dashboard, messaging channel, notification, or other capability until it is connected, tested, permissioned, monitored, and supportable.