WhatsApp, Instagram, Messenger and phone calls. It checks your real calendar and books, cancels or reschedules without your team stepping in. Then it reminds people, so they turn up.
In production today
We do not display logos of companies that are not clients. If you want to speak to someone already using Hachi, ask during the demo and we will put you in touch.
Powered by
Seamless Integrations
Unlike rigid automation tools like n8n or traditional LangChain agents, our LangGraph cognitive agents execute dynamic workflows that adapt to achieve objectives. They don't need pre-defined flows for each scenario—they reason, retry, and self-correct until they succeed.
Full support Partial/Limited Not supported
Messenger
Web Chat
Under the hood there is not one bot but several specialists that hand the conversation between them as needed. You see a single conversation; this is what happens behind it.
Urgency specialist that automatically detects critical situations and activates immediate escalation protocols.
Consultant with temporal awareness that understands "tomorrow", "next week" and verifies schedules in real time.
Massive automated campaigns microservice that sends up to 10,000 WhatsApp template messages per day without human intervention.
Key differentiators that set us apart from traditional automation tools
Unlike n8n or LangChain agents that follow pre-defined rigid workflows, our cognitive agents adapt dynamically to achieve objectives, retrying and self-correcting without manual intervention.
Upload well-structured documents and agents learn automatically via RAG. Change promotions or services instantly—no specialized support needed. Reduce maintenance time and costs dramatically.
Our tested backend architecture is ready to rapidly scale to hundreds of agents and integrations. Customer service agents are just the beginning—the platform supports any workflow type.
Agents remember context from the last 30 messages per client: names, services asked, previous promotions sent. Personalized interactions that feel human and build trust.
Sub-prompts are fully configurable and stored in PostgreSQL per user. Customize agent personality, tone, and behavior to match your brand without touching code.
Celery workers with per-chat locks prevent simultaneous processing of messages from the same conversation, ensuring coherent responses that never lose context.
Enterprise architecture designed for maximum performance, reliability, and scalability
Rule Engine for fast decisions + LLM for complex scenarios. 50+ nodes with conditional routing.
BM25 + semantic with adaptive alpha, cross-encoder reranking, and Reciprocal Rank Fusion (RRF).
Complete isolation per company: Redis prefixes, separate vector namespaces, independent configs.
SharedStateStore with typed dataclasses for pricing, scheduling, handoffs, and audit trail.
Encrypted credentials in PostgreSQL, automatic token refresh, multi-provider.
Asynchronous webhook processing, retry with exponential backoff, no message loss.
Things we can show you working in the demo, not brochure promises
We do the setup: your calendar, your services, your prices and your number. You do not build anything or draw any flow.
With a €200 ticket, between 2 and 8 saved appointments a month cover the fee, depending on the plan. The calculator runs it with your numbers and tells you if it does not pay off.
Also at night, at weekends and on public holidays, which is when people write most and nobody else answers.
More than 80 invariants stop the AI from confirming a taken slot, double-booking, leaving a booking half-done or inventing a price.
Changes to prices, services, opening hours or the assistant's behaviour are included in the fee. There are no flows to redraw.
See every conversation, step in whenever you like and test the assistant before it talks to anyone. Full metrics from the Essential plan.
Your team sees every conversation in a shared inbox and steps in whenever it wants. Hachi takes the repetitive work; people take the delicate part.
WhatsApp, Messenger, Instagram, web chat and WhatsApp calls arrive in the same inbox
Hachi responds automatically with specialized agents
When a person is needed, the conversation is marked pending and assigned to someone on your team or to a specific team
An inbox for several people: conversation assignment, teams, labels, canned responses and response-time reports
Embedded frontend to configure prompts, RAG, and behavior
Its own calendar as the source of truth, copied to Google Calendar. Also Calendly or your own system, via webhook
If you are collecting quotes, you will see wildly different numbers described identically: "an AI assistant that answers WhatsApp and books appointments." They are not the same. They are three different technologies.
| The question that separates them | Tier 1 Drawn flows |
Tier 2 Assistants that talk |
⭐ Tier 3 Talk AND execute |
|---|---|---|---|
| Price per month | €29 – €199 | €150 – €500 | €460 – €1,840 |
| Implementation | €0 – €300 | €0 – €1,000 | €2,000 – €15,000 |
| Does it understand free text? | Only if you press the right button | Yes, it converses well | Yes, including voice notes and mixed intents |
| Does it check your real calendar before offering a slot? | No | No | Yes |
| Does it write the booking into your calendar? | It sends a link | It pings you to do it | Books, cancels and reschedules |
| Can it confirm a slot that is already taken? | Yes — it cannot see your calendar | Yes — it can hallucinate one | No. Code prevents it, not the prompt |
| Who maintains it | You, redrawing flows every time something changes | You, rewriting instructions | Nobody. You upload a document and it learns |
Hachi is tier 3, and it costs less than its own tier: our most complete plan is €990/month where the market reaches €1,840, and our highest implementation fee is €1,690 where that range starts at €2,000.
When you compare €99 against €390, the question is not "why is it more expensive?" It is: what does it cost you when the €99 one confirms a slot that was already booked?
They are good tools and each one wins at its own thing. This is what each does best, and what Hachi adds.
Messaging for teams
It is better if your priority is coordinating a large sales team inside the chats: advanced routing and productivity metrics per person.
Hachi adds a conversation that ends in an appointment written into your calendar with nobody stepping in, under the guardrails above. And your team steps in whenever it wants, through an inbox with assignment, teams and reports.
Voice infrastructure
They are excellent at transcribing, speaking and handling interruptions. In fact, Hachi's voice runs on Retell. What they do not bring is the agent: with them alone you have to program who checks the calendar, what may be promised and what happens when something fails.
Hachi adds the agent already built — booking, changes, cancellations, sales and emergencies — with the same guarantees as in chat, in Spanish, English or Portuguese. Honestly: it answers in about 4 seconds on average, because it checks first; meanwhile it says a short phrase. A voice that only chats answers sooner, but checks nothing.
Guided flows
They are the right choice if your whole process fits in a menu: the client taps fixed options and there is no room for misunderstanding.
Hachi adds what you look for in buttons — an AI that does not make up what matters — without forcing anyone to tap them: it understands free text and voice notes, and the critical part is enforced by code, not by the model.
Prices in euros, VAT not included. No lock-in. Nothing is charged until the 7-day trial ends and you decide to continue.
You work on your own
+ €290 implementation (one-off)
One calendar, one number
No team behind you
You have a team
+ €690 implementation (one-off)
Everything in Solo, plus:
Plus the phone line
+ €1,190 implementation (one-off)
Everything in Essential, plus:
⭐ Most chosen
+ €1,690 implementation (one-off)
Everything in Professional, plus:
Groups and resellers
Implementation quoted per case
Talk about resellingSell it under your own brand
€49/month base fee, plus €4 for each appointment Hachi books over chat and €9 for each one it books during a call. Unlimited conversations: talking is free, only bookings are charged.
Reminders included and a monthly cap of €990: you never pay more than the Complete plan. Implementation: €290. Proactive follow-up and campaigns are not included. If two months in a row it costs you more than a fixed plan, we tell you and switch you.
| Moment | What you pay |
|---|---|
| Demo video call | €0 |
| Setting up your assistant | €0 |
| The 7-day trial | €0 · no card required |
| If you decide to continue | First month + implementation |
| If you decide not to | €0 |
Pay a quarter up front and the implementation fee is waived.
Conversations beyond your plan: €0.25 each. Extra call minutes: €0.20. Service is never cut off.
In the United States the same plans are billed in USD: Pay-per-booking $59 + $5 per booking, Solo $179, Essential $459, Professional $809 and Complete $1,159 a month, with reminders and follow-up by SMS as well as WhatsApp.
The most honest arithmetic we can show you: how many saved appointments per month cover the fee.
| Plan | Monthly fee | Saved appointments/month €200 ticket · 65% margin |
|---|---|---|
| Solo | €149 | 2 |
| Essential | €390 | 3 |
| Professional | €690 | 6 |
| Complete | €990 | 8 |
At a €400 average ticket, halve every figure. What is counted is contribution margin, not revenue: a €200 appointment is not €200 of profit.
Automatic reminders cut no-shows by 40-60%. Every slot that does not sit empty is revenue you had already lost.
Someone who writes at 23:40 and hears nothing until 10:00 has already messaged three competitors. Replying in seconds is not convenience — it is the sale.
They asked the price and never wrote again. Proactive follow-up picks them back up on its own, without anyone having to remember.
A receptionist in Spain costs €25,000-€35,000 a year fully loaded, and takes days off. Hachi does not replace your team: it takes off their plate what they should not be doing.
Put in your own ticket, appointments and no-show rate, and see the whole account: what walks out of the door with every no-show, and the reception hours that go into answering the same questions again. No sign-up, and every formula on screen.
Calculate my returnIt may tell you it does not pay for itself yet. If that is the answer, it says so.
Hachi grew up serving clinics, but what it solves is not a medical problem: it is the problem of any appointment-based business.
Psychology, physiotherapy, nutrition, podiatry, therapies. One calendar, one number, and a phone buzzing while you are with someone.
Aesthetics, dental, spa, beauty, veterinary. Several calendars, several rooms, one overwhelmed front desk.
Hair, barbers, tattoo studios, photography, consultancies, academies, personal trainers. If people book a time, it fits.
Sole trader, your own tax ID, no legal entity. The Solo plan exists exactly for that; there is no minimum below which we turn you away.
See how it works in your industry:
Med spas and aesthetic clinics • Dental practices • Veterinary clinics • Physical therapy clinics • Therapists and counselors • Hair salons and barbershops • Beauty and nail salons • Day spas and massage studios • Auto repair shops • Hair restoration and laser clinics • Plumbers, electricians and home services
See all industries →
You handle personal data and, in healthcare, special-category data. This is not a checkbox.
The assistant does not start conversations on its own. Proactive follow-up only acts on conversations that person opened themselves. Nobody gets a message without having written first.
Campaigns. You supply the list, and with it the responsibility for holding consent. We neither buy nor provide contact databases.
You are the data controller; we are the processor. We sign the Article 28 GDPR agreement before touching a single record.
Treated as special category (Art. 9), with the safeguards that requires. And you can exercise erasure for a specific person when they ask.
Today the question that matters about any AI is not whether it converses well, but who is in control when it gets something wrong. This is our answer, without jargon.
A language model is persuasive even when it is wrong. If you let it decide on its own whether Thursday 17:00 is free, one day it will say yes when it is not — and the person who deals with that is your client, at your front desk.
It works out what the person wants, even in a voice note, half-finished, or with three requests in one message. That it does well, and that is what it is for.
Checking availability, writing to the calendar, not double-booking, not dropping a message: that runs on verified code. If the slot is not free, there is no booking, whatever the conversation says.
It checks the real calendar before offering a time. If there is no free slot, there is no booking.
Even if the client insists three times or sends the details in pieces, the booking is created once.
If a step fails while moving an appointment, what was done is rolled back: the old one is never left cancelled with the new one missing.
Dates are calculated by the code, in your business's time zone, not by the model.
An amount that is not in your documents and that the client did not say is removed before sending, and it says the figure will be confirmed.
If someone tries to make it introduce itself as another company, it does not, on the phone either.
If asked, it says it is an AI assistant, as Article 50 of the EU AI Act requires.
Faced with an emergency or a request to speak to someone, it escalates to a person, even if something fails internally.
invariants enforced in code, each with its own automated test, not entrusted to the model's instructions.
Each one answers the same question: if the model gets it wrong right here, what stops it? It is slow, unglamorous work, and it is exactly the difference between an assistant that dazzles in the demo and one that survives six months without you checking what it says.
Almost nobody explains this before you sign, and it drives the most important decision of all: your number.
| WhatsApp Business app | WhatsApp API | |
|---|---|---|
| Where you use it | On the phone | From a computer, in the browser |
| Who answers | One person | Several people at once |
| Can it be automated? | Quick replies only | It is the only way |
| Anything to install? | Yes, the app | No |
A new number for the assistant. Your usual WhatsApp stays untouched on the phone. Fastest route, nothing is lost.
Clients keep writing to the number they know. In exchange, the history and contacts stored in the app stop being there. Meta does not allow the same number on the app and the API at once.
That history can be rescued before migrating. It is not automation: it is database work, and we quote it separately. We say so now, not once there is no way back.
The ones we get most, answered without hedging.
Six options: Pay-per-booking, €49/month plus €4 per appointment booked over chat and €9 per call, capped at €990; Solo, €149/month; Essential, €390/month; Professional, €690/month; Complete, €990/month; and Multi-site from €1,690/month. On top there is a one-off implementation fee of €290 to €1,690 depending on the plan, charged only if you decide to continue after the trial.
Yes. AI voice reception, which checks the calendar and books during the call, is included from the Professional plan (400 minutes a month). On Solo and Essential it is an optional add-on for €190/month with 300 minutes. The €149 plan does not include voice on its own.
Free, and no card. First a video call where we show it working, then 7 days of trial with your assistant already configured. If you do not continue, you pay nothing.
No. Hachi does not start conversations. Follow-up only acts on conversations that person opened themselves. The exception is campaigns, where you supply the list and are responsible for holding consent.
No, because the booking is not decided by the language model. Availability checks and calendar writes run on verified code. If the slot is not free, there is no appointment.
What matters does not depend on it. More than 80 critical rules are enforced in code: it does not confirm a taken slot, does not double-book, does not leave a booking half-done, does not quote a price that is not in your documents, does not pose as another business, does not claim to be a person and, in an emergency, escalates to someone on your team. The model understands and converses; the code enforces.
Yes, but decide it with the facts in front of you. Meta does not allow the same number on the WhatsApp Business app and the API at the same time. Either you use a new SIM, or you migrate your usual one accepting that the history stored in the app stops being accessible. It can be rescued beforehand, quoted separately.
Three to five working days once we have your details and the number question settled. We do the setup; you build nothing.
Service is not cut off. Extra conversations bill at €0.25 each and extra call minutes at €0.20. If the overage is constant we suggest moving plan, because it works out cheaper for you.
No. You can leave whenever you want. Implementation is charged once on signing, and is waived if you pay a quarter up front.
No, and anyone who says otherwise is selling you smoke. It takes the repetitive part off their plate — opening hours, prices, directions, moving an appointment — so their time goes where a person is actually needed. And it covers nights and weekends, when there is nobody.
What you lose by migrating your number, when Hachi is not worth it for you, the case in which a €29 tool is the right choice, and the four questions you should put to any quote — ours included.
See all 85 questions →The three questions that come up most when someone is comparing quotes. Answered in full, without asking for your email in exchange.
Why quotes run from €29 to over €900 a month and what exactly you are paying for at each tier. Includes what to check in anyone's small print.
The exact mechanism behind a duplicated booking, what goes wrong when data arrives in pieces, and the silent failure that shows up in no error report.
Why numbers get blocked, what the 24-hour window is, and exactly what you lose if you migrate your long-standing number. Worth knowing beforehand.
If you got this far looking for engineering rather than pricing, this part is for you.
Graph nodes emit typed signals that a router with no LLM in the loop turns into flow decisions. The model proposes; the state machine disposes. Check-Signal-Act in every node.
More than 80 properties the system guarantees in code, with an explicit cut-off rule: if Python guarantees the outcome it is an invariant; if only the prompt asks for it, it is a suggestion.
Per-tenant isolation down to the conversation checkpointer key. Every query is parameterised and scoped by tenant: one client's state cannot leak into another's.
The graph can be compiled from the database: a new agent is defined with typed extraction fields and its own tools, with no code deploy.
Safe retry across the whole external surface. A duplicated webhook does not create two appointments; a retried campaign does not send twice.
The inference layer is decoupled: provider and model are selected per tenant. No dependency on a single vendor or a single model family.
The variable cost of serving one client — messaging, inference and telephony — is measured in tens of euros per month, against fees of €390 to €990. Margin does not depend on scaling headcount, because maintaining the assistant is autonomous: there are no flows to redraw when a business changes its prices or its hours.
And the moat is not the model, which anyone can rent. It is the slow work of turning more than 80 critical behaviours into verifiable guarantees. That does not get replicated in a weekend with a flow builder.
Investing, or building things like this? Write to ceo@hachi.live with "technical" or "investment" in the subject and I will answer personally.