Vero in Practice: What an In-Tool AI Assistant Can and Cannot Do

What this is about

Vero is the AI assistant inside dVersum. It does not live in a second browser tab. It lives in the same workspace as your projects and invoices. This piece covers what that changes day to day, where the limit sits, and why we put it there on purpose.

The problem is not the model

Here is the usual loop. You want to know how many hours went into the redesign this month and how much of that is still unbilled. You switch to an AI chat. It knows nothing about you.

So you export. A CSV from your time tracker, the invoice list next to it, paste both, write the prompt. The answer reads well. It is also exactly as old as your export.

Then comes the part nobody talks about. The answer is text, and it stays text. To make it do anything, you retype it somewhere else, as a time entry or an invoice line. The assistant thought. You worked.

Run the numbers on that loop. Four times a week, seven minutes each for export, paste, prompt, and re-entering the result in the tool that actually holds your data. That is 28 minutes a week. Across 46 working weeks it adds up to 1,288 minutes, roughly 21 hours a year. At an hourly rate of €85.00, that is €1,785.00 nobody pays you for.

There is a second cost that is easy to miss. Every one of those exports puts a pile of client data on your clipboard. Names, rates, project budgets, sometimes your full client list. If you signed a DPA with any of those clients, that clipboard is not a detail. It is processing.

The model is not the problem here. The missing access to your real data is.

What Vero actually does

Vero reads and writes in your workspace. Not in a copy, not in an index from last week, but in the data you are looking at right now.

One week, as it actually runs:

You come out of a call and type into the Vero chat: "Log 2.5 hours on Redesign Acme, yesterday, wireframes." The entry sits on the project afterwards, with date, duration, and description. No form, no calendar clicking.

[IMAGE: Vero chat with the command and the resulting time entry on the project below it]

Two weeks later you ask: "How many hours went into Acme in August and how much of that is still unbilled?" Vero adds up the entries. Say the answer is 14 hours open. At €85.00 an hour that is €1,190.00 sitting nowhere except in your time tracker.

The next sentence is: "Turn that into an invoice for Acme, August." The draft appears with line items built from the tracked time. You check it, fix the descriptions, send.

[IMAGE: Invoice draft with line items carried over from tracked time]

Between "14 hours are open" and "the invoice is drafted" there is exactly one sentence. No handoff, no second tool.

The same mechanic covers tasks, time off, and finance numbers. "Book me off Monday through Wednesday" lands in team planning. "Show me receivables older than 30 days" returns a list instead of a guess. Voice works too, which matters when you are driving and the hour would otherwise disappear.

The most useful case is also the dullest one: the follow-up after a client call. Instead of creating five cards by hand, you dictate a sentence. "Create three tasks for Acme: revise wireframes by Friday, send image selection to the client, prepare the phase 2 quote." The cards land on the board with due dates. That is the difference between a documented call and one whose results live in a notebook.

Forms work the same way. If you collect briefings regularly, you describe the form instead of dragging it together in the builder. And because Vero has a memory function, you do not re-explain recurring context in every chat, such as which client is on which rate.

One detail for the technically curious: the same tool layer is reachable over an MCP server. If you work in Claude or another MCP client, you connect your workspace there instead of copying data back and forth.

The decision behind it, and what it costs

Vero has no free write access to your workspace. Every action has a defined tool behind it, or the action does not happen. An assistant allowed to write anywhere in a database is more flexible. It is also the one where, three months later, you cannot explain why an invoice says €480.00 instead of €840.00.

That costs flexibility, noticeably. If you want something no tool covers, Vero says so instead of inventing a path. I prefer that trade. On accounting data, "I cannot do that" beats a well-phrased fabrication.

The practical effect shows up when something goes wrong. Vero logs time on the wrong project because two clients have similar names. The entry is visible in the project overview, and you fix it in two clicks. With an assistant that rewrites values in the background, you first have to work out what it did at all. That traceability is why we chose the narrower design.

The second decision is about keys. Vero runs on BYOK, so you bring your own API key. You decide which provider sees your prompts, and you see the cost directly with them. The price is one more setup step and a second bill that does not come from us. If that bothers you, it will keep bothering you.

Your workspace data is unaffected by that choice. It sits in German data centers, specifically Nuremberg, ISO 27001 certified, encrypted at rest and processed under the GDPR. For freelancers outside Germany billing European clients, that is often what decides whether a tool is an option at all, well before anyone looks at features.

What Vero cannot do

Vero makes no decisions for you. It knows two projects are running in parallel. It does not know which one matters more. It knows your hourly rate but cannot judge whether this client should be paying a higher one. What gets priority and what it costs stays with you.

It also knows only what is in the workspace. The agreement from Tuesday's phone call is not there unless you recorded it. An assistant inside the tool does not raise your data quality. It makes the gaps visible faster.

Vague phrasing makes it guess. "Log yesterday's time" without a project or a duration gets you a follow-up question or a wrong match. Short sentences work better than paragraphs.

The larger point sits one level up and applies to AI in project management generally. On facts already in the system, an assistant like this is strong: count, filter, convert, create. On anything requiring judgment, it produces noise with decent grammar. Auto-generated status summaries are the clearest example. They read well, and after two weeks nobody opens them. Estimated remaining effort is the same trap: the number looks precise, its basis is not. We do not build those features, even though they would look good in a comparison table.

And the limit I consider the most important: I would not send an invoice I had not read. Vero builds the draft. Approval stays with you.

Who this fits, and who it does not

It fits if your work already runs the chain from project to time to invoice, and the handoffs in between cost you hours. Solo freelancers with five to ten active clients notice it first, because nobody else is doing their catch-up entry for them.

For small agencies the value moves. There, the problem is rarely a single time entry. It is the month-end question of which hours landed on which project and which of them can be billed. Someone used to assemble that by hand in a spreadsheet. The Team plan covers ten users at €79.00 a month, or €63.00 a month billed yearly. Measured against half an hour of spreadsheet work a week, it pays for itself quickly.

It does not fit if you want a writing tool. For drafting and research, a general AI model is better, and Vero does not try to replace it. It also does not fit if your processes are wired deep into an existing stack and you only want an AI layer on top. Vero is part of the workspace, not an attachment for other systems.

Bottom line

The difference between an assistant inside the tool and a chat tab next to it is not answer quality. It is whether the answer turns into an action or into more typing. Model size and phrasing come second.

dVersum starts at €19.00 a month, or €17.00 a month billed yearly. Seven days free, no credit card. If you want to know whether it holds up for your workflow, take one real client during the trial and run them end to end: log the time, query the hours, create the invoice. You will know after that one round.