I'm building Verbasil. Every AI you use forgets everything the moment the chat ends. Meanwhile your company's real memory, the decisions, the reasons, the recurring signals, ends up scattered across tools or trapped inside someone else's model. I think that's the wrong default. A company's memory should be proactive (it surfaces what matters before you ask), verifiable (every claim traces to a source, no inventing), and sovereign (it stays yours, in Europe, portable across any model). This week, someone became the first to pay for it. That's the signal I needed: the problem is worth solving for someone other than me. I'm building it in the open. Follow along for the build, and the thesis behind it.

48h at the OpenAI DevDay Hackathon Paris: I built Murmure, a live radio commentator for blind people. Your phone watches the TV and tells you what’s happening, in your ear. 🎥 youtu.be/pU1EvKQauzo 🔑 Try it (BYOK): murmure-phrnl.vercel.app 💻 github.com/phrnl0xyz/murmure @borvibe @openaidevs #parisdevdayhack @OpenAI

48 hours of hacking, edited into 2 minutes. The demo video is done… and I can’t show it to you yet. 🤐 The crowd explodes. Soon you’ll know why. @borvibe @OpenAIDevs @oleksandrtsp @OpenAI

Since yesterday, I’ve been building something at the OpenAI DevDay Community Hackathon in Paris. Today, I’m testing it out in the streets. One hint: the most important part won’t be on screen. More soon. @borvibe @openaidevs #parisdevdayhack

Verbasil now joins your meetings. Invite it to a Google Meet. It listens, transcribes, pulls out every decision and who made it, then emails the recap to everyone. And those decisions don’t die in an inbox. They go into your company memory, next to the ones made in email, docs and chat. Ended 12:00. In memory 12:01. Recap sent 12:03. Your meetings already happened. Now what gets decided in them stays.

1250 commits 185 migrations de base 230 fichiers de tests en 4 mois Verbasil.com

DAMNNNN’´N’´´NNNNNNNNNN

BIG news: @VerbasilHQ just joined Anthropic’s @claudeai for Startups program! 🎉 As a founder who builds with Claude every day, this is a small dream come true. Verbasil is the memory of your team’s decisions. This is just the beginning. 🔥

On rentre dans une nouvelle ère de presentation

I’d lost track of where I was After a few days away from my project, I came back trying to remember what I needed to do. What was still in progress? What mattered most? Where should I start? I kept turning those questions over in my head. Then I remembered I’d kept all of this in @VerbasilHQ. I opened it and found several interesting things I’d already started. I’d almost gone looking for new ideas when I already had plenty to work on. What surprised me was that first instinct: asking my brain to reconstruct something I’d already taken the time to save elsewhere. A break can change how you see your work. Some ideas feel less important when you return. Others suddenly make more sense. There’s nothing wrong with changing direction. But I want that to be a choice. Not something that happens because I forgot where I was going. This time, finding my unfinished work helped me choose what to pick up again. One small rule I’m keeping for my next break: before looking for something new to start, check what’s already waiting for me.

Documentation is not failing because teams are lazy. It fails when we make it a second job. The decision happened in a conversation. The constraint was explained on a call. Customer feedback arrived in a message. Then, later, someone has to recreate all of it in the right document. By then, nuance is gone. So are rejected alternatives and the reason the team changed its mind. A memory layer should start with the work a team already produces, keep the sources, and let people validate what deserves to remain. That is the problem I’m building Verbasil around.

Your dashboard can tell you that a number moved. It cannot always tell you which customer feedback warned you, which decision changed the conditions, or what happened after an experiment. One data point is not a learning. One comment is not a pattern. One correlation is not causation. The useful work is connecting: source → verified signal → decision → action → result → learning. That is the context layer I’m building Verbasil around.

The model is not the whole problem. Your AI does not know the decisions, failed experiments, customer feedback and constraints that shaped your company. So every new chat becomes a partial restart. A capable model is still working from an incomplete brief. Chat memory is not company memory. That is the context layer I’m building Verbasil around.

Most onboarding teaches new hires what to do. It rarely teaches why. “We always do it this way” is not context. It does not explain the problem the rule was designed to prevent, whether the constraint still exists, or when to challenge it. That is how teams repeat old mistakes while following the rules perfectly. Good onboarding transfers judgement, not only procedures. That is part of the context I’m building Verbasil to preserve.