Interactive demos
Mini-apps to try in your browser, with fictional companies. They show what an agent really does: read, sort, check, draft, then wait for your approval.
NotHuman OS
Ten AI agents run a fictional company: emails, invoices, leads, content, chat, meetings, reports, video, SEO and GEO visibility, sync. Then try most of them, one by one, below.
▸ Alvélia Fiduciaire Sàrl (fictional) · Lausanne SIMULATION
Inbox on standby
→ ready for accounting
SOGC · new registrations
Radar on standby
3 personalised messages approved
Brief Remind clients to pay into pillar 3a before 31 December
Pillar 3a: your 2026 payment must reach your account by 31 December to be deducted from your taxes. Some banks set an earlier cut-off date, so check yours with your bank. Unsure how much to pay in? Write to us and we will check your maximum amount.
Scheduled · LinkedIn · Tuesday 08:30 · + 6 versions
Waiting for visitors
Minutes + 3 action items sent
Report ready for management · 18:00
Script 31 December is coming: 3 things to check for your pillar 3a.
30-second video ready · FR/DE/EN subtitles · Also as 9:16 reel, 16:9 cut, 4-min podcast, article
Site
alvelia-fiduciaire.example
FAQ · live Year-end closing and VAT: which deadlines apply to an SME in Lausanne?
Tracked question
“fiduciaire Lausanne bouclement TVA”
Site cited: 2/5 answers · simulated
1,287 records synced
Simulation · fictional company and data
19:00 · Day complete
Alvélia Fiduciaire Sàrl (fictional)
It is Not Human what we can do.
Simulation · fictional company and data
Demonstration simulation: companies, volumes, results and time saved are fictional. Every NotHuman system is custom-built, tested on your data, and you choose what a person approves before it goes out. ChatGPT, Perplexity, Gemini, Copilot and Google AI Overviews are trademarks of their respective owners; NotHuman is not a partner or affiliate of these companies.
AI without the jargon
No tech background needed. An AI agent is a simple chain: your data comes in, the AI understands it, code checks it, you decide. Follow the packet.
01 Input
An email, a PDF, a form, a phone call: the agent starts from what already reaches you. No new habits needed.
Under the hood IMAP · Microsoft Graph · Gmail API · webhook · OCR · Whisper
02 AI
A language model reads the way a person would: it sorts, pulls out what matters and writes a draft.
Under the hood LLM of your choice → schema-validated JSON
03 Code
Figures are never left to the AI alone: plain code recalculates the totals and checks the IBAN, the VAT and your rules. No guesswork.
Under the hood Python · IBAN checksum (mod 97) · VAT 8.1% · business rules
04 Human
You approve, the agent acts in your tools: reply sent, entry ready to import, file stored. Nothing goes out without your approval, unless you decide otherwise.
Under the hood Approval queue · your tools’ APIs · full audit log
What’s happening
AI reads and writes. Code checks. Humans decide.
Two useful tools, two different jobs.
Used as a chat
You Summarise this email and suggest a reply.
ChatGPT Here’s a three-point summary and a suggested reply…
Copy, then paste into your email
Great for thinking, writing and exploring ideas.
Digital colleague
Agent log · fictional example
Great for tasks that come back every day.
ChatGPT has an agent mode too; the difference here: your rules, code that re-checks every figure, your approval and the hosting you choose. Not one against the other: an agent can run on GPT, the model behind ChatGPT, on Claude, Gemini or Mistral, or on Apertus, the open Swiss language model built by EPFL, ETH Zurich and CSCS, hosted in Switzerland or on your own servers for sensitive data. You decide where your data is processed.
Want the whole team to use ChatGPT and other AI tools safely? Our half-day workshop
Read more: What is an AI agent? ChatGPT at work and data protection
Names and trademarks belong to their owners. NotHuman is independent: not a partner or affiliate. Legal notice
Demo 1/11 · Operations
Choose a (fictional) accounting firm, property manager or agency and run the agent: it labels each email, pulls out what matters and drafts replies in the sender’s language.
Reads, sorts, summarises and drafts your replies. You keep the final word.
Inbox
Select an email, or run the agent.
Everything runs in your browser: nothing is sent or stored. Simplified demo: answers are prepared in advance or computed by simple rules in your browser. In production, a language model suited to your data does this work, and someone on your team signs off.
I want this for my businessWhat’s happening, step by step
Emails arrive Input
The agent watches your inbox, like a colleague opening the post the moment it lands.
The AI understands AI
A language model reads the message in its own language and recognises what it is: an invoice, something urgent, a new client…
It pulls out what matters AI
Deadline, amount, file number, attachments: everything important is picked out and summed up in one line.
To the right person Code
Rules agreed with you send each request to the right person. No AI here: clear rules.
A draft is ready AI
For routine requests, the agent writes a reply in the client’s language, based on your templates.
You approve Human
Nothing goes out without your say-so: you read, adjust if needed, and send.
Run the demo: each step lights up here.
ModelApertus 70B · Swiss open source
Data processedA server in Switzerland or yours
Picture a colleague who reads every email the second it arrives, understands what it’s about, passes it to the right person and drafts a reply. It never sleeps, never loses anything, and always waits for your go-ahead.
Words to know
Why it’s safe
The agent’s pipeline
AI reads and writes. Code checks. Humans decide.
Stack
Guardrails
What the agent produces (JSON)
{
"email_id": "f2",
"language": "de",
"category": "urgent_tax_office",
"priority": "high",
"summary": "Reminder from the tax office, due 5 November. The client thinks he has already paid.",
"fields": {
"client": "Thomas Keller",
"deadline": "2026-11-05",
"action": "verify_payment"
},
"route_to": "marc@alvelia-fiduciaire.example",
"draft": {
"language": "de",
"status": "awaiting_approval"
},
"confidence": 0.94
}Choose the engine
How we set it up
What we need from you
Demo 2/11 · Operations
An invoice with a QR-bill, a rental application or a bank statement: the agent extracts the data, then code checks every figure. Tick “scanned document” to see what happens when the reading goes wrong.
Reads invoices, statements and applications. Code re-checks every franc. You approve.
Fictional document, in French as is common in Romandy
Facture n° 2026-1187
| Désignation | Qté | Prix | Montant |
|---|---|---|---|
| Flyers A5, recto-verso | 2’000 | 0.34 | 680.00 |
| Affiches A2 | 50 | 8.40 | 420.00 |
| Mise en page et préparation | 1 | 140.00 | 140.00 |
| Sous-total HT | 1’240.00 | ||
| TVA 8,1 % | 100.44 | ||
| Total CHF | 1’340.44 | ||
Section paiement
Compte / Payable à
CH44 3199 9123 0008 8901 2
Imprimerie Calame Sàrl
Référence
21 00000 00003 13947 14300 09017
Monnaie CHF · Montant 1’340.44
Demande de location
Pièces jointes
Banque Démo SA
| Date | Libellé | Débit | Crédit |
|---|---|---|---|
| Solde initial | 24’310.55 | ||
| 02.10 | Paiement QR, Favre & Fils (réf. …0417) | 2’160.00 | |
| 05.10 | Loyer bureau octobre | 2’450.00 | |
| 09.10 | Paiement QR, C. Rochat (réf. …0422) | 540.00 | |
| 14.10 | Télécom Démo SA, facture mensuelle | 189.90 | |
| 20.10 | Virement, Atelier Nivéo Sàrl | 1’296.80 | |
| 27.10 | Salaires octobre | 18’430.00 | |
| 29.10 | Virement, J. MARTIN | 800.00 | |
| Solde final | 8’037.45 |
Extracted data AI
Run the agent to read this document.
Checks done by code Code
Everything runs in your browser: nothing is sent or stored. Simplified demo: answers are prepared in advance or computed by simple rules in your browser. In production, a language model suited to your data does this work, and someone on your team signs off.
I want this for my businessWhat’s happening, step by step
The document arrives Input
A PDF invoice by email, a scan, a photo or a bank statement: the agent picks it up wherever it lands, like a colleague opening the post.
The AI reads the page AI
It reads the text and understands the layout: where the supplier, the lines, the total and the payment part are.
It pulls out the fields AI
Supplier, number, dates, amounts, IBAN, QR reference: each value goes into its own box, ready to be checked.
Code checks every figure Code
No AI here: plain code redoes the maths (IBAN checksum, QR reference, VAT at 8.1%, totals). If an amount was misread, the sums don’t add up and the document is blocked.
A proposal is ready Output
An accounting entry, a bank reconciliation or a list of missing documents with the reminder drafted: the work is prepared, not yet carried out.
You approve Human
Nothing is imported or sent without your go-ahead: you check at a glance, correct if needed and approve.
Run the demo: each step lights up here.
ModelOpen-source OCR + Apertus 70B
Data processedA server in Switzerland or yours
Picture a colleague who opens every invoice, statement or application the moment it arrives, copies the right details into the right place and redoes every calculation before showing it to you. It never tires, and you always have the final say.
Words to know
Why it’s safe
The agent’s pipeline
AI reads and writes. Code checks. Humans decide.
Stack
Guardrails
What the agent produces (JSON)
{
"document": "facture-2026-1187.pdf",
"type": "supplier_invoice",
"supplier": {
"name": "Imprimerie Calame Sàrl",
"vat_id": "CHE-123.456.789"
},
"number": "2026-1187",
"date": "2026-10-30",
"due_date": "2026-11-29",
"amounts": {
"currency": "CHF",
"subtotal": 1240,
"vat_rate": 8.1,
"vat": 100.44,
"total": 1340.44
},
"payment": {
"qr_iban": "CH4431999123000889012",
"reference": "210000000003139471430009017"
},
"checks": {
"passed": 9,
"failed": 0,
"iban_mod97": true,
"qr_reference_mod10": true,
"vat_recomputed": true,
"totals_match": true
},
"proposal": {
"debit": "6600 Advertising",
"input_vat": "1171 Input VAT",
"credit": "2000 Accounts payable"
},
"status": "awaiting_approval"
}Choose the engine
How we set it up
What we need from you
Demo 3/11 · Operations
Choose a fictional dataset and a language: the agent gathers the figures, computes the KPIs, spots anomalies, writes the commentary and draws the charts. You review it before it goes out.
Gathers your figures, calculates, spots the gaps and writes the report. You approve before it goes out.
Sources Input
Calculations Code
Anomalies Code
Log
Everything runs in your browser: nothing is sent or stored. Simplified demo: the figures are calculated in your browser from fictional data, and the summary is a template filled in with those figures. In production, a language model writes from the calculations, and someone on your team signs off.
I want this for my businessWhat’s happening, step by step
The figures come in Input
The agent reads your exports (Excel, CSV) or your software’s API: accounting, CRM, online shop, advertising.
Code calculates Code
Revenue, margin, change on last month: fixed formulas, like a spreadsheet. No AI here.
Gaps stand out Code
Thresholds agreed with you flag anything unusual: receivables up, margin down, stock running low.
The AI writes the summary AI
A language model receives the calculated results and explains them in plain words, in the reader’s language. It explains; it doesn’t calculate.
The report takes shape Output
Key figures, charts and points to watch: a document you can read in two minutes, online or as a PDF.
You approve Human
You read it, correct it if needed, and the report goes to management, the client or the board.
Run the demo: each step lights up here.
ModelApertus 70B · Swiss open source
Data processedA server in Switzerland or yours
Picture a financial controller who, on the first morning of the month, has already gathered all your figures, calculated every KPI, spotted what’s off and written a clear summary. Code calculates, AI explains, you decide.
Words to know
Why it’s safe
The agent’s pipeline
AI reads and writes. Code checks. Humans decide.
Stack
Guardrails
The facts passed to the AI (JSON)
{
"report": "monthly_management",
"company": "Alvélia Fiduciaire Sàrl (fictional)",
"period": "2026-10",
"language": "en",
"facts": {
"fees_chf": 168300,
"fees_change_pct": 6.1,
"operating_margin_pct": 15.3,
"margin_change_pts": -4.7,
"cost_drivers": [
{
"line": "temporary staff",
"change_chf": 9800
},
{
"line": "licences",
"change_chf": 4200
}
],
"receivables_over_30d_chf": 37750,
"receivables_change_pct": 38,
"billable_rate_pct": 64.9
},
"alerts": [
{
"rule": "receivables_over_30d_change > 20%",
"value": 38,
"level": "alert"
},
{
"rule": "margin_change < -2 pts",
"value": -4.7,
"level": "alert"
},
{
"rule": "billable_rate < 65%",
"value": 64.9,
"level": "watch"
}
],
"summary": {
"status": "awaiting_approval",
"numbers_checked": true
}
}Choose the engine
How we set it up
What we need from you
Demo 4/11 · Operations
Ask a question about a fictional SME’s staff rules and procedures. The assistant answers from the documents, cites its source, and says when it doesn’t know.
Answers your team from your own documents, cites the source and says when it doesn’t know.
Boisvif assistant Answers from 6 internal documents
Boisvif assistant
Hello! I answer the team’s questions using Menuiserie Boisvif SA’s internal documents: staff regulations, expenses, safety, warranty, onboarding and prices. I always cite my source, and if the answer isn’t there, I tell you.
Pick a question below, or ask your own.
Suggested questions
Agent log
Knowledge base
Menuiserie Boisvif SA (fictional) · 18 staff · canton of Vaud
threshold: score ≥ 2.0 and 50% of the question covered
The assistant only knows these documents. Click one to open it.
Source document
Click a cited source to see the exact passage, highlighted in the document.
Demo: the keyword search really runs in your browser. The answers are prepared in advance from the passages; in production, a language model writes them from the passages found and always cites its sources. Internal rules of a fictional company, not legal texts.
Everything runs in your browser: nothing is sent or stored. Simplified demo: answers are prepared in advance or computed by simple rules in your browser. In production, a language model suited to your data does this work, and someone on your team signs off.
I want this for my businessWhat’s happening, step by step
You ask a question Input
In your own words, as you would ask a colleague. The assistant picks out the keywords and their synonyms: “vacation” means “holiday”.
Search through your documents Code
A search engine scans every passage of your documents in a few milliseconds, like the index of a giant book.
The best passages Code
Each passage gets a relevance score. Only those above the threshold are kept; otherwise, there’s no answer.
A grounded answer AI
The AI writes a clear answer only from the passages kept. Nothing reliable enough? It writes nothing.
The source is cited Output
Every answer names the document and the section. One click shows you the exact passage, highlighted.
Otherwise, HR takes over Human
No reliable source? The assistant says “I don’t know” and offers to pass the question on to HR.
Run the demo: each step lights up here.
ModelApertus 70B · Swiss open source
Data processedA server in Switzerland or yours
Picture a colleague who has read all your internal documents (staff handbook, procedures, prices) and answers every question by pointing to the exact paragraph. If it can’t find the answer, it says so and hands over to HR instead of making something up.
Words to know
Why it’s safe
The agent’s pipeline
AI reads and writes. Code checks. Humans decide.
Stack
Guardrails
What the assistant produces (JSON)
{
"question": "How many weeks of holiday do I get?",
"keywords": ["weeks", "holiday"],
"scanned": { "documents": 6, "passages": 18 },
"ranked": [
{ "passage": "r2", "source": "Staff regulations, art. 4.2", "score": 4.5, "kept": true },
{ "passage": "r3", "source": "Staff regulations, art. 4.3", "score": 1.8, "kept": false },
{ "passage": "r6", "source": "Staff regulations, art. 6.3", "score": 1.4, "kept": false }
],
"threshold": { "min_score": 2, "min_coverage": 0.5, "coverage": 1, "passed": true },
"answer": {
"text": "You are entitled to 5 weeks of holiday per calendar year…",
"citations": ["r2"]
},
"escalated_to": null
}Choose the engine
How we set it up
What we need from you
Demo 5/11 · Growth
Choose your offer, a sector and a canton: the agent spots new (fictional) companies, scores their fit and drafts a personalised first email plus two follow-ups, for you to approve.
Spots new companies and writes each one a tailored message. You approve every send.
Companies kept
Choose your offer, a sector and a canton, then run the agent.
The Swiss rule. In Switzerland, the Unfair Competition Act (art. 3 para. 1 let. o) bans mass advertising by email without prior consent. Our approach: few messages, truly individual, approved by a person, with the sender clearly identified and an easy way to say no.
Everything runs in your browser: nothing is sent or stored. Simplified demo: answers are prepared in advance or computed by simple rules in your browser. In production, a language model suited to your data does this work, and someone on your team signs off.
I want this for my businessWhat’s happening, step by step
New companies appear Input
The agent reads the new registrations published in the SOGC, the Swiss Official Gazette of Commerce. Public data, updated every working day.
Your criteria Code
Simple code keeps only the sectors and cantons you serve. No AI here: clear filters, set with you.
The AI rates the fit AI
A language model reads each company’s registered purpose and gives it a fit score, with its reasons in plain words.
A tailored message AI
For each company kept: a short, personal first email and two spaced follow-ups, in the language and tone you chose.
Swiss rules checked Code
Code checks every message: identified sender, a free and easy way to say no, an individual message, a capped volume (Unfair Competition Act, art. 3 para. 1 let. o).
You approve Human
Nothing goes out without your say-so: you read each sequence, adjust it if needed, and decide.
Run the demo: each step lights up here.
ModelApertus 70B · Swiss open source
Data processedA server in Switzerland or yours
Picture a sales assistant who reads the new entries in the commercial register every morning, spots the companies that genuinely need what you offer, and drafts a personal first email plus two follow-ups for each. No mass mailing: a handful of well-targeted messages that you read and approve one by one.
Words to know
Why it’s safe
The agent’s pipeline
AI reads and writes. Code checks. Humans decide.
Stack
Guardrails
What the agent produces (JSON)
{
"source": "SOGC · 2026-10-27",
"company": "Café Zéphyrine Sàrl",
"town": "Morges (VD)",
"purpose": "Running a café-restaurant and catering service.",
"contact": "info@cafe-zephyrine.example",
"fit_score": 91,
"reasons": [
"Registered 3 days ago",
"Sàrl: bookkeeping required by law",
"VAT likely above CHF 100,000 turnover"
],
"sequence": [
{
"day": 0,
"language": "en",
"status": "awaiting_approval"
},
{
"day": 4,
"only_if": "no_reply"
},
{
"day": 10,
"only_if": "no_reply",
"last": true
}
],
"checks": {
"sender": "marc@alvelia-fiduciaire.example",
"opt_out_in_every_email": true,
"individual_message": true,
"daily_volume": "5/20"
},
"human_approval": "pending"
}Choose the engine
How we set it up
What we need from you
Demo 6/11 · Growth
Choose a goal and a segment of fictional clients: the agent removes contacts without consent, writes an email sequence, a text message and a post in each contact’s language, schedules everything and checks the rules before you approve.
Targets, writes, personalises and schedules your campaigns. You approve before the first send.
Sarine Langues Sàrl (fictional) · Language school in Fribourg · 1,286 contacts · FR · DE · EN
Goal
From client list to permitted audience
Client list (3 years)1,286
No course in 12 months—
Email: consent checked—
SMS: number and consent—
Rule last_course < 30.10.2025
| Contact | SMS | |
|---|---|---|
| Claire MorandFRFribourg · German B1 · ended March 2025 | ✓opted in | ✓opted in |
| Lukas AebischerDEDüdingen · French B2 · ended June 2024 | ✓opted in | ✓opted in |
| Emma ClarkeENFribourg · French A2 · ended February 2025 | ✓opted in | ✓opted in |
| Nadia BapstFRBulle · Italian A2 · ended May 2025 | ✗no consent | ✗no consent |
| Julien FavreFRLausanne · English B2 · ended December 2023 | ✓client, opt-out offered | –no number |
| Stefan RiedoDEBern · English B2 · ended June 2025 | ✗unsubscribed | –no number |
| Léa RossierFRMarly · German A2 · ended March 2024 | ✓opted in | ✓client, opt-out offered |
| Reto ZbindenDEMurten · Italian A1 · ended November 2023 | ✗no consent | ✗no consent |
Preview: 8 contacts out of 214
✓ opted in · ✓ client: existing client who was told they could refuse when their details were collected (similar offers only) · ✗ excluded
Everything runs in your browser: nothing is sent or stored. Simplified demo: answers are prepared in advance or computed by simple rules in your browser. In production, a language model suited to your data does this work, and someone on your team signs off.
I want this for my businessWhat’s happening, step by step
The right audience Code
Code filters your list for the goal, then removes everyone who hasn’t given consent or has unsubscribed. No AI here: clear rules.
The AI writes AI
A language model writes the sequence (3 emails, an SMS, posts) in each contact’s language, with two subject lines to test.
Every message becomes personal Code
Code fills in the first name, the course, the date… and adds the sender and the unsubscribe link. The AI never sees a contact.
Everything is scheduled Tool
Your email tool schedules the sends on day 0, 3 and 7, and stops the sequence as soon as someone replies or unsubscribes.
The rules are checked Code
Consent, sender, unsubscribe, SMS length, banned phrases: everything is checked, and fixed where needed.
You approve Human
Nothing goes out without your say-so. After that, the campaign runs on its own and you get the results.
Run the demo: each step lights up here.
ModelApertus 70B · Swiss open source
Data processedA server in Switzerland or yours
Picture a colleague who knows every one of your clients, writes to them in their own language at the right moment, checks it is allowed to, and never forgets a follow-up. It prepares the whole campaign, then waits for your go-ahead.
Words to know
Why it’s safe
The agent’s pipeline
AI reads and writes. Code checks. Humans decide.
Stack
Guardrails
The campaign plan the agent prepares (JSON)
{
"campaign": "reactivation_2026_11",
"segment": {
"rule": "last_course < 2025-10-30",
"matched": 214,
"excluded_no_consent": 37,
"excluded_unsubscribed": 9,
"email_eligible": 168,
"existing_client_exception": 47
},
"languages": {
"fr": 104,
"de": 49,
"en": 15
},
"sequence": [
{
"step": 1,
"channel": "email",
"day": 0,
"subjects": [
"A",
"B"
],
"split": "50/50"
},
{
"step": 2,
"channel": "email",
"day": 3,
"when": "no_click"
},
{
"step": 3,
"channel": "email",
"day": 7,
"when": "no_click"
},
{
"step": 4,
"channel": "sms",
"day": 4,
"when": "no_click",
"audience": "sms_consent"
}
],
"exit_on": [
"reply",
"booking",
"unsubscribe"
],
"checks": {
"sender": true,
"unsubscribe_link": true,
"consent": true,
"sms_segments": 1
},
"status": "awaiting_approval"
}Choose the engine
How we set it up
What we need from you
Demo 7/11 · Growth
Pick a fictional company’s website: the agent crawls it like a search engine, checks the technical basics, groups what customers search for, looks at whether AI assistants quote the company, then proposes an action plan for you to approve.
Makes your business findable on Google and quotable by AI assistants. You decide on every action.
https://alvelia-fiduciaire.example
Engine Apertus 70B · Swiss open source
Like Google’s crawler, the agent follows every link and maps the site.
Ready to crawl alvelia-fiduciaire.example. Run the agent: crawl, technical checks, demand, AI visibility, then a plan for you to approve.
Code checks what Google and AI assistants look at first.
What your customers actually type, grouped by intent.
The same 8 questions put to 4 AI assistants: is your company cited?
Scores computed by code, actions ranked by impact and effort. Your call.
Everything runs in your browser: nothing is sent or stored. Simplified demo: answers are prepared in advance or computed by simple rules in your browser. In production, a language model suited to your data does this work, and someone on your team signs off.
I want this for my businessWhat’s happening, step by step
The site is crawled Code
Like Google’s crawler, a program follows every link and maps your pages: the ones that work and the ones that are broken.
The technical side is checked Code
Code checks speed, titles, languages and structured data. No AI here: measurable rules.
Demand is mapped Tool
A keyword tool shows what your customers actually type, grouped by intent: learn, compare, find nearby, get in touch.
AI assistants are asked AI
The same questions go to several AI assistants. An AI notes whether your business is cited, and which pages are.
A plan is proposed AI
Actions are ranked by impact and effort, with a 90-day goal. A goal, not a promise.
You approve Human
You choose which actions to start. Nothing goes live on your site without your review.
Run the demo: each step lights up here.
ModelApertus 70B · Swiss open source
Data processedA server in Switzerland or yours
Picture an expert who goes through your whole website in seconds, checks what Google sees, finds out what your customers really search for and asks AI assistants whether they know you. It hands you a clear plan, ranked by priority, and you decide.
Words to know
Why it’s safe
The agent’s pipeline
AI reads and writes. Code checks. Humans decide.
Stack
Guardrails
What the agent produces (JSON)
{
"site": "alvelia-fiduciaire.example",
"crawl": {
"pages": 38,
"redirects_301": 3,
"errors_404": 2,
"languages": "fr, en"
},
"checks": {
"critical": 1,
"to_improve": 5,
"ok": 2
},
"seo_score": {
"total": 51,
"technical": 56,
"content": 44
},
"ai_visibility": {
"questions": 8,
"assistants": 4,
"answers": 32,
"citing_site": 3,
"rate": 0.09
},
"top_action": {
"id": "structured_data",
"impact": "high",
"effort": "low",
"status": "awaiting_approval"
},
"goal_90_days": {
"seo_score": 75,
"ai_visibility": 0.25,
"type": "objective_not_promise"
}
}Choose the engine
How we set it up
What we need from you
Demo 8/11 · Growth
One idea becomes LinkedIn and Instagram posts, a newsletter and a Google update, in French, German and English.
One brief, four channels, three languages. Nothing is published without your go-ahead.
One brief, four channels, three languages. Choose a company and a topic: the agent writes each version following the brand’s rules, code checks it, then the agent suggests a publishing plan. You approve.
1 brief → 4 channels × 3 languages = 12 posts
Brief
Boulangerie Dumoulin (fictional)Sourdough week
Sourdough week from 16 to 21 November, with a free tasting on Saturday 21, from 9 am to 12 noon. A different loaf every day, and starter to take home.
Artisan bakery near Fribourg · French- and German-speaking customers
Brand rules
| FR | DE | EN |
|---|---|---|
| pain au levain | Sauerteigbrot | sourdough bread |
| tresse | Zopf | tresse loaf |
| dégustation | Degustation | tasting |
Generated texts
Your texts will appear here: one card per channel, each language one click away, with brand checks and a copy button.
Everything runs in your browser: nothing is sent or stored. Simplified demo: answers are prepared in advance or computed by simple rules in your browser. In production, a language model suited to your data does this work, and someone on your team signs off.
I want this for my businessWhat’s happening, step by step
You give a brief Input
A few lines are enough: the topic, the dates, the offer. Just as you’d explain it to a colleague.
Your rules are loaded Code
Formal address, glossary, ‘ss’ instead of ‘ß’: your brand rules are written down. The agent gets them before it writes a single line.
The AI writes each version AI
A language model writes one version per channel and language: short for Google, with hashtags for Instagram, with a subject line for the newsletter. Adapted, not translated word for word.
Code checks Code
Each channel’s length limit, ‘ß’ replaced by ‘ss’, informal address, banned words: every text goes through fixed checks. No AI here: rules.
The week’s plan Tool
Each post gets a slot from Monday to Friday, following the rhythm agreed with you.
You approve Human
You read, adjust if needed, then approve. Nothing is published without your say-so.
Run the demo: each step lights up here.
ModelApertus 70B · Swiss open source
Data processedA server in Switzerland or yours
Picture a copywriter who knows your tone, your vocabulary and your rules, and turns a simple idea into ready-to-post content for LinkedIn, Instagram, your newsletter and your Google profile, in French, German and English. It writes fast, never tires, and always waits for your approval.
Words to know
Why it’s safe
The agent’s pipeline
AI reads and writes. Code checks. Humans decide.
Stack
Guardrails
What the agent produces (JSON, excerpt)
{
"brief": {
"company": "Boulangerie Dumoulin (fictional)",
"topic": "levain",
"tone": "warm"
},
"versions": [
{
"channel": "linkedin",
"lang": "fr",
"chars": 507,
"limit": 3000,
"checks": {
"vouvoiement": true,
"banned_words": 0,
"within_limit": true
},
"slot": {
"day": "tue",
"time": "08:30"
},
"status": "awaiting_approval"
},
{
"channel": "newsletter",
"lang": "de",
"subject_chars": 53,
"limit": 60,
"checks": {
"sie_form": true,
"swiss_spelling": true,
"banned_words": 0,
"within_limit": true
},
"fixes": [
{
"from": "Grüße",
"to": "Grüsse"
}
],
"slot": {
"day": "tue",
"time": "07:30"
},
"status": "awaiting_approval"
},
{
"channel": "instagram",
"lang": "en",
"chars": 500,
"limit": 2200,
"checks": {
"british_spelling": true,
"banned_words": 0,
"within_limit": true
},
"slot": {
"day": "fri",
"time": "18:00"
},
"status": "awaiting_approval"
}
],
"versions_total": 12,
"published": 0
}Choose the engine
How we set it up
What we need from you
Demo 9/11 · Growth
A fictional brief becomes a script, a storyboard, a voice-over and subtitles, then a video you can watch in the player. Bonus: the outline of a podcast episode.
Writes, storyboards, voices, subtitles and renders. You approve before anything goes live.
Pick a brief, a format and the voice language, then run the agent. Everything is fictional and nothing is published.
Engine
The brief
Régie Pervalis SA (fictional)
Run the agent: the script is written here, scene by scene.
The 5 storyboard scenes will appear here.
The preview comes to life during production.
CSS demo · no sound · fictional dataAfter the render, click in the timeline to move around.
After the render, the video waits here for your approval.
Everything runs in your browser: nothing is sent or stored. Simplified demo: answers are prepared in advance or computed by simple rules in your browser. In production, a language model suited to your data does this work, and someone on your team signs off. Here the video is drawn in CSS, with no sound and no real file. In production: your visuals, a real voice-over and an MP4 file.
I want this for my businessWhat’s happening, step by step
The AI writes the script AI
From your brief, a language model writes a short text, split into scenes, in your brand’s tone of voice.
It builds the storyboard AI
Each scene gets a visual (your logo, your photos, licensed footage) and a duration.
The voice-over is generated AI
A synthetic voice reads the script in the language you chose. Your own voice is possible, only with your consent.
Code times the subtitles Code
No AI here: code splits the text, syncs it with the voice and checks readability (line length, reading speed).
The video is rendered Tool
An editing tool puts pictures, voice, music and subtitles together into one video file, in the format you need.
You approve Human
You watch, ask for a change or approve. Nothing is published without your say-so.
Run the demo: each step lights up here.
ModelApertus 70B, Whisper and open-source voices
Data processedA server in Switzerland or yours
Picture a full production team (scriptwriter, editor, voice-over artist, subtitler) that turns a three-line brief into a video ready for review, in your languages and formats. Then it waits for your go-ahead.
Words to know
Why it’s safe
The agent’s pipeline
AI reads and writes. Code checks. Humans decide.
Stack
Guardrails
What the agent produces (JSON)
{
"video_id": "pervalis-launch-en",
"format": "9:16",
"language": "en",
"duration_s": 30.1,
"scenes": [
{
"n": 1,
"type": "kinetic_text",
"on_screen": "Trouble at home?",
"voice": "A dripping tap, a stuck blind? …",
"duration_s": 7.2,
"source": "motion_design"
},
{
"n": 2,
"type": "app_mockup",
"on_screen": "Report it in 30 seconds",
"duration_s": 6.4,
"source": "client_screenshots"
},
{
"n": 4,
"type": "cutaway",
"duration_s": 4.9,
"source": "licensed_stock",
"licence_ref": "STK-48213"
}
],
"voice": {
"type": "synthetic",
"real_person": false,
"disclosure": "end_card"
},
"subtitles": {
"files": [
"fr.srt",
"de.srt",
"en.srt",
"it.srt"
],
"max_chars_per_line": 32,
"max_cps": 15.3
},
"status": "awaiting_approval"
}Choose the engine
How we set it up
What we need from you
Demo 10/11 · Growth
A fictional studio’s website and its assistant: it answers from the site’s pages and cites its source, qualifies the request, offers a time slot and hands over to a person when needed. Ask it your own questions.
Answers, qualifies and books at any hour, in the visitor’s language. Your team stays in charge.
A visitor asks about prices, then a trial class. The assistant answers with its sources, qualifies the request and offers slots.
Behind the scenes: what your team gets
UnderstandingAI
Context used: page /en/prices · browser language · no ad cookies
QualificationAI
CRMAPI
The lead will appear here as soon as a slot is booked.
Everything runs in your browser: nothing is sent or stored. Simplified demo: answers are prepared in advance or computed by simple rules in your browser. In production, a language model suited to your data does this work, and someone on your team signs off.
I want this for my businessWhat’s happening, step by step
A visitor asks a question Input
On your website, at 10 pm or on a Sunday, in their own language. No hunting for the right page, no form to fill in.
The AI understands AI
It recognises the language and what the person wants (a price, opening hours, an appointment) and picks out the useful details.
It answers, with its source AI
The answer comes from your approved content (pages, prices, FAQ) and names the page. If it can’t find it, it says so.
It qualifies the request AI
Like a good adviser, it asks two or three questions: the need, the timing, how to get back in touch.
Appointment and CRM Tool
It offers free slots from your calendar and creates the lead in your CRM, with a summary.
Your team takes over Human
You confirm the booking, or take over the chat when the visitor asks or the topic is sensitive.
Run the demo: each step lights up here.
ModelApertus 70B · Swiss open source
Data processedA server in Switzerland or yours
Picture an adviser on your website who replies within seconds, day and night, from your own pages and in the visitor’s language, then qualifies the request, offers an appointment and drops the lead into your CRM. It always says it’s an AI, and hands over to your team when it should.
Words to know
Why it’s safe
The agent’s pipeline
AI reads and writes. Code checks. Humans decide.
Stack
Guardrails
The lead created in the CRM (JSON)
{
"lead_id": "lead_7f3k2",
"source": "website_chat",
"page": "/en/prices",
"language": "de",
"name": "Jonas Meier",
"email": "jonas.meier@mail.example",
"need": "trial_class",
"timing": "saturday",
"contact_preference": "email",
"slot": {
"start": "2026-11-07T09:00",
"timezone": "Europe/Zurich",
"status": "held"
},
"sources_used": [
"/en/classes",
"/en/hours"
],
"summary": "Trial class on a Saturday morning. Conversation in German; confirmation by email.",
"next_step": "confirm_booking",
"marketing_consent": false
}Choose the engine
How we set it up
What we need from you
Demo 11/11 · Tool
Enter the time your team spends on repetitive tasks. The calculator estimates the hours and the money automation could free up each year.
Estimate
0 % of a full-time job (42 h)
Where to start
Get a detailed analysisIndicative estimate calculated in your browser. Nothing is sent unless you click the button.
Everything runs in your browser: nothing is sent or stored.
Where we help
Pick a department: here are the agents that take the work on, what they do, and a rough idea of the time involved.
Every agent follows the same circuit
Invoices, reminders, receipts: the same clicks, every week.
Combined example ≈ 13 h / week less repetitive work
Every supplier invoice is read, QR-bill included, then matched with the payments on your bank statement.
example ≈ 4 h / week for 250 invoices a month
See the demo : Invoice reading and bank reconciliationOverdue invoices are spotted and a polite reminder is drafted, in the client’s language.
example ≈ 2 h / week for 40 reminders a month
Learn more : Unpaid invoice remindersEvery email to info@ is read, summarised and passed to the right person, with a draft reply.
example ≈ 5 h / week for 60 emails a day
See the demo : Shared inbox triageReceipt photos are read, amounts and VAT extracted, your expense policy applied: ready for accounting.
example ≈ 2 h / week for 80 receipts a month
Learn more : Expense claimsFind, qualify, follow up. Your salespeople should be selling, not copying and pasting.
Combined example ≈ 8 h / week less repetitive work
New companies in your canton are spotted and scored against your offer. You keep the good ones.
example ≈ 3 h / week for 30 new companies a week
See the demo : Lead radarFor each lead you keep, a short personal email plus two follow-ups. A person reviews everything: no mass mailing.
example ≈ 2 h / week for 10 first messages a week
Learn more : Personal first messagesA quote with no answer for a week? The agent flags it and drafts a follow-up that picks up the context.
example ≈ 1 h / week for 15 open quotes a week
Learn more : Quote follow-upsEvery request from your website or inbox is qualified (need, urgency, size) and handed to the right salesperson, with a summary.
example ≈ 2 h / week for 10 requests a day
See the demo : Qualifying inbound requestsOne idea, five channels, three languages. Without losing your week to it.
Combined example ≈ 10 h / week less repetitive work
One brief becomes LinkedIn and Instagram posts, a newsletter and a Google Business Profile update, in French, German and English.
example ≈ 4 h / week for 3 topics a week
See the demo : One idea, every channelYour texts adapted into French, German, Italian or English, with your glossary and your tone. A person reviews before publishing.
example ≈ 3 h / week for 5 pages a week
Learn more : Translation and adaptationEvery new Google review gets a draft reply, in the customer’s language and your tone. Negative reviews are flagged to you.
example ≈ 1 h / week for 40 reviews a month
Learn more : Review repliesEvery Monday, the key news from your sector and five topic ideas, with sources.
example ≈ 2 h / week for one weekly round-up
Learn more : Market watch and topic ideasThe same questions every day, in several languages. Accurate answers, faster.
Combined example ≈ 10 h / week less repetitive work
Opening hours, lead times, documents needed: the reply is drafted from your templates, in the customer’s language.
example ≈ 5 h / week for 25 requests a day
See the demo : Replies to routine requestsPrices, terms, procedures: your team asks, the assistant answers from your documents and cites its source. If it doesn’t know, it says so.
example ≈ 2 h / week for 40 questions a week
See the demo : Answers from your documentsEvery voicemail is transcribed, summarised and passed to the right person, with the number to call back.
example ≈ 2 h / week for 10 messages a day
Learn more : VoicemailReviews, emails and surveys grouped every month: the recurring topics and the customers to call back.
example ≈ 1 h / week for 200 pieces of feedback a month
Learn more : Customer feedback analysisThe right numbers at the right time, without chasing them.
Combined example ≈ 8 h / week less repetitive work
Every Monday at 7 am: revenue, open invoices, pending requests. Code does the maths, the AI explains the changes.
example ≈ 3 h / week for one report a week
See the simulation : Weekly reportEvery morning, the essentials in five lines: urgent emails, today’s meetings, this week’s deadlines.
example ≈ 1 h / week at 15 minutes a day
See the demo : Morning briefingYour notes, or a meeting recording made with everyone’s consent, become clear minutes: decisions, tasks, owners, deadlines.
example ≈ 2 h / week for 4 meetings a week
See the simulation : Meeting minutesExpected payments, bills to pay and the forecast balance over eight weeks, updated daily. Unusual delays are flagged to you.
example ≈ 2 h / week for one review a week
Learn more : Cash-flow trackingLess paperwork, more time for people.
Combined example ≈ 7 h / week less repetitive work
Holidays, expenses, procedures: the assistant answers from your staff handbook and cites the section. If it doesn’t know, it says so.
example ≈ 2 h / week for 30 questions a week
See the demo : Staff questionsEach application is summarised against your criteria, with missing documents flagged. A person reads the files and decides, always.
example ≈ 2 h / week for 30 applications a week
Learn more : Application summariesAs soon as the contract is signed: task list, accounts to open, documents to sign and a welcome message, all ready.
example ≈ 2 h / week for 2 new starters a month
Learn more : New starter onboardingA first draft written from the employee file and reviews, clear and fair. HR reviews, adjusts and signs.
example ≈ 1 h / week for 3 certificates a month
Learn more : Work certificatesOrders, appointments, call-outs: everything flows, with no double entry.
Combined example ≈ 11 h / week less repetitive work
Orders received by email or as PDFs are read and prepared for your software, with quantities and prices checked.
example ≈ 4 h / week for 25 orders a day
See the demo : Orders and delivery notesAppointment requests are understood, slots offered from your calendars, confirmations and reminders drafted.
example ≈ 2 h / week for 40 appointments a week
Learn more : Appointments and schedulingA leak, a breakdown: the item, address and urgency are identified, and the job order is prepared for the right tradesperson.
example ≈ 3 h / week for 15 requests a week
See the demo : Repair requestsA new client in your CRM? They’re carried over to accounting and the calendar, through their API or an export. No more double entry.
example ≈ 2 h / week for 30 records a week
Learn more : Tools that talk to each otherExample figures, based on the volumes shown: not client results, and not a promise. Your real numbers are measured during the audit, on your own data. Estimate my own hours
With NotHuman
With a single agent, chosen with you. We aim for a first automation in service 2 to 4 weeks after the audit.
A 30-minute call, or the 3-minute quiz: we find what costs you the most time.
Tested on your data, with your approval rules.
Hosting, monitoring, small improvements and a monthly report.
Use-case builder · 1 minute
Answer 3 quick questions and watch your AI agent being designed live. Then ask for it: we reply personally.
They call no live language model: answers are prepared in advance or computed by simple rules in your browser. Nothing you type leaves your computer.
In production, we connect the model suited to your task and data (a leading model, or an open-source model such as Apertus hosted in Switzerland) to your tools, with your approval rules. The companies, people and documents in these demos are fictional.
NotHuman
Let’s talk: in a free 30-minute audit, we look at how to adapt it to your tools and your data.