Where the assistant is headed. Tell us what you want next, and we build in the open: shipped, in flight, and planned, with no dates we cannot keep.
Tell the assistant to keep an eye on something and alert you when it changes, before it goes cold.
The desk in your pocket, with push for the few things that actually need you.
Single sign-on and shared workspaces, so a team can run on one assistant.
Teach the assistant a repeatable analysis once, then run it on demand. In private testing, not switched on for tenants yet.
Send a Data Room document as an expiring, view-only link and see who opened it. Built, in testing, not live yet.
Re-rank what Random Brain returns, so the most relevant memory reaches the draft first.
Raise signing from advanced to qualified (QES) for documents that need the higher bar.
People, companies, and deals that open into everything you have on them, in one view.
Tune how the assistant reads you: your voice, your rules, and what it should never do.
Memory recalled by meaning, so drafts and briefs carry the context.
Aug 2026Prepare a document and send it for e-signature, end to end.
Aug 2026Documents on durable EU storage, with a fail-loud guard.
Aug 2026Generate documents from your own templates, with a live preview.
Aug 2026Know the moment a connected tool drops.
Aug 2026The whole desk moved to a fast, modern web app.
Jul 2026Product updates in plain language, what changed and what it means for your day. Everything we have shipped since we started, in May 2026.
Your Meetings tab now fills itself from your calendar, so prep packets, transcripts, and briefings line up against the real schedule. Triage also reads your own context now, who you are and how you work, to sort and draft more like you would.
FeaturePeople, companies, and deals page instantly now, even at thousands of rows, and a dropped connector shows up right on the list instead of failing quietly.
ImprovementYour documents now sit on durable EU storage with a fail-loud guard, so an upload can never quietly go missing. Nothing about how you use it changes; it is simply harder to lose.
ReliabilityPrepare a document, add the signers, and send it for e-signature without leaving the assistant. Signing itself is never automated, and the document is stored in the EU; notification email goes via Resend (US, on the register).
FeatureWhen the assistant looks something up on the web, the search now runs through an EU provider, and we retired the US ones. Every message it drafts also got a brighter, plainer design, in German or English.
ImprovementRecall is now semantic, so the assistant surfaces the right memory even when you do not use the exact words. Every draft and brief opens already knowing the history.
FeatureConnector health tells you the moment a connected tool disconnects, instead of you finding out from a draft that never wrote.
ImprovementGenerate documents from your own templates, and build new ones with a live preview as you go.
FeatureRandom Assistant moved into a fast, modern web app: Brief, Inbox, Chat, Pipeline, People and Companies, Calendar, Meetings, Data Room, and Connections, all in one place.
FeatureMemory grew a graph, so the assistant links your people, companies, and deals to each other, and meeting names come through clean.
FeatureAdded Attio as a CRM connector, and the assistant began surfacing unanswered threads and follow-ups on its own, staged for your approval.
FeatureThe assistant began prepping meetings from your calendar and staging the follow-ups afterwards.
FeatureRandom Assistant started: inbox triage and drafting, the first connectors over Google and HubSpot, and encryption at rest from day one.
FeatureWhere an assistant fits, what EU residency really means, and how we compare to the other ways of getting this done.
A virtual assistant is real help. It is also a hire. Here is the honest version of where each one wins.
A good virtual assistant knows your calendar, chases the follow-ups, keeps the CRM tidy, and takes the small coordinating work off your desk. It is genuinely valuable. It is also a role you have to recruit, onboard, and eventually replace, that mostly runs during set hours in one timezone, and that costs a European salary plus everything that comes after it.
Random Assistant does the coordinating half of that role, and it is there the day you connect your inbox, calendar, and CRM. No recruiting, no onboarding week, no notice period, no gap when someone leaves. It drafts every reply in your voice, ranks the inbox so the few things that need you sit on top, remembers every preference and every deal without being reminded, and preps the day while you sleep.
Judgement calls. Relationships. The phone call that needs a human on the line. The work that is not on a screen at all. We do not replace that, and we are not going to pretend we do. If your assistant work is mostly people and judgement, hire a person.
Volume, memory, and patience. It does not forget the thing you told it once, it does not take holiday, and it does the same careful coordination on the four hundredth email as it did on the first. And it hands you every decision, so nothing leaves your account without you.
The honest split: hire a person for the judgement, run the assistant for the coordination. Most desks need both.
The part people miss is that this is not either-or. Let the assistant carry the coordination, and the person you hire spends their day on the work that only a person can do. That is a better job for them and a better result for you.
A chatbot is a brilliant answer machine. It also starts every conversation knowing nothing about you.
Ask a chatbot anything and it will give you something useful. But it cannot see the thread you are replying to, the deal you are closing, the meeting you had last week, or the way you write to that one client. So it gives you a good generic answer, and you do the work of making it specific.
The gap is not cleverness. It is context. Random Assistant is wired into your inbox, calendar, and CRM, so when you ask it something it answers from your real work: the actual overdue count, the actual person, the actual history. Then it can act on the answer, draft the reply, prepare the document, update the deal, and stage it for you.
A chatbot answers. An assistant that knows your desk can act on the answer, with your hand on the final move.
An assistant that can do things is only safe if there are things it can never do on its own. Ours never sends, signs, or moves anything without you. It prepares everything and waits in a queue for your approval. You get the speed of the draft and the safety of the human hitting send.
So the question was never which model is smarter. Use whichever you like; we route to strong ones. The difference is that the assistant knows your desk, can act on it, and still leaves the last call to you.
The modern way to run a small business is five subscriptions that do not talk to each other. You are the integration.
A CRM here, an inbox there, a notes app, a calendar, a signing tool. Each is good at its one thing. The problem is the space between them, and you are what fills it. You copy context from the email into the CRM. You retype the meeting decision into a task. You go looking for the file in the third app.
That tax is quiet but real. Every switch costs a little focus, and the day is made of switches. The tools are not the problem; the gaps are.
Not one more app. One assistant across the desk you already have.
Random Assistant does not ask you to leave the tools you like. It connects to them over OAuth and works across all of them at once. Ask it something and it looks in your mail, your meetings, your documents, and your CRM together. Reply to a thread and it can update the deal in the same move, once you approve it.
The tools stay where they are. You stop being the glue.
Most software puts the important safety settings on a page you can find. Ours is not on a page.
The assistant can read everything you connect. It can draft, prepare, analyse, and stage. But there is a short list of actions it can never take on its own: sending an email, signing a document, moving money. In the code, sending is a hard stop. Even if you ask it to send, it will not. It writes the reply, puts it in your Drafts, and waits.
We built it this way for a simple reason. An assistant that can act without you is a liability the first time it is wrong, and it will be wrong sometimes. An assistant that prepares everything and hands you the final move is fast and safe at the same time.
It does the work. You make the call. That is the whole philosophy in one line.
Everything the assistant stages lands in one Decisions queue, and nothing acts until you approve it there. You are never hunting through the app to find what it wants to do. It is all in one list, waiting, and the list is the only way anything happens.
A setting can be changed by accident, or by someone who should not have. A limit in the code cannot. That is the difference we wanted, and it is why this one is not a setting at all.
"EU-hosted" is easy to put on a page and hard to mean fully. Here is the honest account.
Your content, the mail, the documents, the CRM records, the memory, is stored and encrypted in the EU, with a separate key for every tenant. Your data is never used to train any model. We publish the list of the sub-processors we rely on, so you can see exactly who touches what.
The AI reasoning itself. Today, some of the model calls run outside the EU. We are moving that onto EU-hosted models, and until it is fully there, we will not claim it is. The distance between "your data is stored in the EU" and "everything, including the inference, runs in the EU" is real, and we would rather tell you which one is true than blur the two.
Residency is not a badge you earn once. It is a set of specific claims, each either true or not.
So when you read our security page, you will find it says storage and encryption are in the EU, and it does not yet say inference is. That gap is deliberate, and it will close. We would rather be the company that tells you exactly where the line is today than the one that rounds up.
There is a new job nobody applied for: keeping up with which model is best this month, and for what. It is a job the product should do, not you.
Ask most people using AI at work how they choose a model, and the honest answer is a shrug and a habit. One tool for writing, another that a colleague swore by, a third because it was in the tab. The landscape changes every few weeks, and staying current has quietly become unpaid work layered on top of the actual job.
That is backwards. You do not choose which database your bank runs, or which codec your calls use. You care about the outcome. The plumbing is the vendor’s problem. AI should be the same.
Pick the outcome, not the engine. Which model wrote your draft is our problem, not yours.
Random Assistant chooses the right approach for each task behind the scenes. A quick classification, a careful long draft, a search across your own records: different jobs, different engines, and you never see the switch. You do not write prompts, you do not compare benchmarks, and you do not re-learn a tool every quarter. You connect your work and approve the results.
Absorbing that choice is not us being lazy about giving you control. It is us doing the part that is genuinely ours to do. The control that matters to you is over the outcome: your voice, your rules, and the final send. The control that does not matter, which model ran, is exactly the part we should carry so you can stop thinking about it.
The measure of good infrastructure is that you forget it is there. If we have done our job, you will never wonder which AI is best, because you will never have to.
Watch where an hour actually goes and most of it is not spent deciding. It is spent assembling the context to decide.
You open the thread. Then the earlier thread. Then the calendar, to see when you last spoke. Then the CRM, to check the stage. Then the shared drive, for the proposal you sent. By the time you have the picture, the decision itself, reply or wait, discount or hold, call or email, takes ten seconds. The ten seconds was never the problem. The twenty minutes of gathering was.
Most software has answered this with more places to look: another dashboard, another inbox, another feed. That adds to the chase. It does not end it.
The work is the decision. The gathering is tax. An assistant should pay the tax so you keep the work.
Random Assistant does the assembling before you arrive. The morning brief already knows what is waiting, what changed overnight, and what needs a call. Each item comes with the thread, the history, and a draft, so you are reading a decision, not building one. You approve, adjust, or decline. The context is there because it was gathered for you, not by you.
Reclaiming twenty minutes once is a convenience. Reclaiming it on every decision, all day, changes what the day is for. You spend it on judgement, the part that is actually yours, instead of on the errand of getting ready to judge. That is the shift: from chasing information to making decisions.
Everyone gets the same 168. The question is how many of yours go to work that a machine would happily do instead.
Sleep takes about a third. What is left is the week you actually run a business in, and a surprising slice of it never touches the decisions only you can make. It goes to the shuffle: reading mail to sort it, rewriting the same replies, prepping for meetings, writing up what was agreed, and keeping the CRM honest afterwards.
Run the tally on a normal week. An hour a day drafting and re-drafting email. A few hours of meeting prep and follow-up. An hour lost to context-switching between the inbox, the calendar, and the CRM that do not talk to each other. It adds up to most of a working day, every week, spent on coordination rather than judgement.
You cannot buy more than 168 hours. You can stop spending them on the work a machine does well.
These are exactly the tasks Random Assistant is built for. Mail arrives triaged, with a reply already drafted in your voice. Meetings arrive prepped, and leave as a briefing with the follow-ups staged. Your people and deals stay current without you tending them. None of it sends or changes anything until you approve, so the time you save does not cost you control.
For one week, note every task you do that does not need your judgement, only your time. Rewriting a reply you have sent a hundred times. Copying a meeting outcome into three tools. Hunting for a file. That list is your reclaimable week. It is usually larger than people expect, and it is the week we are trying to give back.
Request early access and we will set your assistant up for your desk.