VidéPro CRM + Ava
One system where the whole company comes together — with a company brain that knows everything and acts on it.
VidéPro ran on FileMaker with a handful of separate tools around it. Today the entire company runs on one system: client contact, its own phone exchange, quotes and invoices, bank transactions, planning, video delivery and the public website. And because everything lives in one place, Ava could be built on top: a company brain you ask in plain language what's going on — and that then does the work.
57
Tools at Ava's disposal
6
Channels she's reachable on
±339,000
Lines of code
30+
Connected systems
What it does
One question, every source at once
Ava searches email, WhatsApp, calendar, invoices, tasks, phone calls and documents in parallel, and turns it into a single answer. Not seven searches in seven systems.
She knows what's going on
A bird's-eye brief refreshed every fifteen minutes: calendar, urgent tasks, outstanding balances, missed calls and fresh leads from the website.
And she actually does it
Draft and email a quote, send an invoice with a payment QR, schedule or move an appointment, call a client back, create a project. Knowing is half of it; doing is the rest.
Reachable wherever you are
In the CRM, through the phone exchange, on the CarPlay screen in the car, over WhatsApp, by email — and as a participant in a Zoom or Teams meeting, under her own name and face.
Permissions by caller ID
Who is calling decides what she may do: owner, team member, client or unknown. Not what someone claims to be.
Draft first, send second
Messages and emails arrive with you as a proposal. Only after an explicit 'yes' do they actually go out.
She learns from corrections
Correct her and that correction becomes a standing rule. Past mistakes are stored literally, as lessons in her memory.
Its own phone exchange
Menu, queue, opening hours and voicemail on our own infrastructure. Every call is recorded, transcribed and summarised — with action items landing as tasks in the CRM.
EditRoom for delivery
Clients review video frame by frame and draw their notes straight onto the picture; those corrections become tasks automatically. With 720p proxies that already have the right camera LUT baked in.
From quote to bank statement
Quotes, invoices, prepaid day packages, purchase invoices through an AI inbox, and a bank link that matches incoming payments to the right invoice by itself.
The website runs from the same system
videpro.nl comes out of the same codebase, with its own CMS and a content engine that researches, writes, translates, illustrates and publishes articles.
What that sounds like in practice
“Wat speelt er?”
In één beeld: je agenda van vandaag en morgen, urgente taken, openstaande posten — inkoop én verkoop — relevante mail, gemiste oproepen en verse leads van de website.
“Dat document dat ik gisteren binnenkreeg — waar ging dat over?”
De bijlage is bij binnenkomst al uitgelezen en geïndexeerd, dus ze antwoordt uit de inhoud van de pdf in plaats van te zoeken naar het bestand.
“Wie stond er ingepland op de opnamedag van vorige week?”
De crew per categorie — camera, geluid, regie — met per persoon of het bevestigd was, plus de ingehuurde apparatuur per leverancier.
“Reageer op dat appje, zeg dat ik morgen langskom.”
Ze zet een concept klaar en stuurt dat eerst naar jou. Pas na een expliciet 'ja, verstuur maar' gaat het bericht de deur uit.
“Stuur me om kwart voor 7 een herinnering dat ik moet terugbellen.”
Ze plant de actie in en meldt zich 's avonds uit zichzelf — geplande acties zijn haar eigen hartslag.
“Wanneer was ik voor het laatst bij deze klant en wat bespraken we?”
Semantisch zoeken door het hele bedrijfsgeheugen: de afspraak met locatie en notities, plus de mails van rond die datum — ook die van collega's.
“Maak een offerte voor twee draaidagen en mail hem.”
Ze stelt de offerte op met de prijslijst erbij, maakt de pdf en verstuurt hem. Bij een factuur gaat er een betaal-QR mee.
“Bel de opdrachtgever terug dat de opnames doorgaan.”
Ze belt zelf uit met jouw opdracht als leidende instructie — en hangt netjes op als ze een voicemailbandje krijgt.
How we built it
Next.js 16 with React 19 on Vercel, Firestore as the only database — no ORM, no migrations. Realtime listeners keep every screen current by themselves: 111 collections, 342 API routes and 78 dashboard pages in one application.
Ava is not a chatbot beside the system but an agent inside it: one engine with dozens of tools that ask targeted questions of the company data, backed by a semantic memory layer. Every new email, call and appointment is indexed the moment it arrives.
Purely semantic search turned out to be unusable: short contact cards hijacked the entire top of the results while the email holding the answer sat at position 84. The ranking was therefore tuned empirically, with per-type limits and a mild recency bonus — measured, not assumed.
Heavy lifting deliberately sits outside the app: separate workers for video transcoding, virus scanning and the realtime voice bridge, with media on object storage behind temporary links. Several AI providers run side by side with explicit roles and a fallback chain when one goes down.
Every deploy requires a version bump plus a changelog line, and the version number sits bottom-left in the sidebar — live proof of what is running. A release gate binds approval to the exact commit; a new push invalidates earlier authorisation automatically.
Why we built it
Company knowledge is scattered across mailboxes, chat threads, phone calls, PDFs and someone's head — and it is unreachable at exactly the moment you need it, on the road or sitting with a client.
So we built the foundation first: everything in one system. Only that made a company brain possible — you cannot be smart about data you do not have. The result is a company you can ask instead of search, and that answers in actions.
Need something like this in your company?