# Mindsmiths

The big AI companies are launching personal agents, and each one wants to own the relationship with your customers. Your company is building AI too. All of it is book smart.

What earns a place in someone's life is street smarts: when to speak up, and when to shut up. Mindsmiths is street-smart AI for the companies people rely on.

It's software, and it runs as the reliance layer of your customer stack. It sits on top of the systems you already run, such as CRM, billing, support and your app, next to your service AI and marketing tools. Every day, for every customer, it decides whether anything would actually help: sometimes a heads-up before something runs out, sometimes weeks of useful notes before an offer, most days nothing at all. Over time more of your customers read you, answer you and come to you first, and that shows up where you're measured: acquisition, retention and growth.

Our customers include national telecom operators, fintechs and an airport, in Europe and the US. Mindsmiths, Inc. is a US company. Customers in the EU sign with our Croatian subsidiary, Mindsmiths d.o.o.

**To see it running on one of your use cases,** write to Jordan Yates, our VP of Sales, at [jordan@mindsmiths.com](mailto:jordan@mindsmiths.com).

This is the whole website, in one file. Nothing is behind a form or a sales call. It's written so your AI can read all of it before you talk to us, and so can you.

## At a glance

Everything a security, legal or procurement review asks for is on this page, in full, before we ever talk.

- **Contract.** You sign our licence agreement and an order form. Everything else it's made of is published in full, the version you sign stays fixed, and changing it takes both signatures:
  - [Terms of Service](/terms-of-service). How the platform may be used, who owns what, and who is liable for what.
  - [Data Processing Addendum](/data-processing-addendum). How we handle your customers' personal data for you, and who else handles it.
  - [Security Addendum](/security-addendum). How we protect the platform and your data, and what happens if something goes wrong.
  - [Service Level Agreement](/service-level-agreement). Uptime, response times, and what you get back when we fall short.
- **Security.** ISO 27001 certified. Our SOC 2 Type II audit is under way. An independent penetration test every year, AI included.
- **Data.** Each customer gets its own environment. Your customers' data never trains any model, ours or our AI providers'. For their personal data, we act only on your instructions, as your processor under GDPR.
- **Hosting.** Microsoft Azure, in the US (Virginia), or in the EU (Frankfurt) for customers in the EU.
- **Uptime.** 99.0% a month. If we miss it, credits go to your account without you asking.
- **Control.** Your team approves every playbook before it goes live. Contact rules are enforced by the system, not left to the AI, and there is a kill switch.
- **Price.** Public: credits, three plans, no platform fees on top. Your order form is what counts. [Pricing](/pricing)
- **Privacy.** What we collect about website visitors and the people at our customers, and why: [Privacy Policy](/privacy-policy).
- **Who you sign with.** Mindsmiths, Inc., a Delaware corporation, or Mindsmiths d.o.o. if you're in the EU.

Below, **For your review team** has one section each for IT, security, data, messaging, legal and procurement, and each can be forwarded on its own.

## What it fixes

Most customers hear from a company when it wants something: an offer, a renewal, a review. They learn the pattern. They open less every year and check the price before they buy.

Every tool a company owns is built around one number. The app pushes the offer. The campaign tool sends the promotion. The loyalty programme rewards the purchase. The service AI closes the ticket. Each one does its job well. None of them works on whether the customer would choose you again without being asked.

The AI going into those tools makes each one faster at its own job. It's book smart: it knows your catalogue, your policies and your help pages. It doesn't know when a message will cost you more than it earns.

Meanwhile, personal agents are starting to sit between people and the companies they deal with: summing up messages, comparing offers, deciding what is worth a look. They're book smart too, and they don't work for you. What gets through to a person is what that person already trusts.

That gap shows up in places every company already pays for:

- **Acquisition.** Most visitors leave without a word, and the cost of each lead goes up every year.
- **The first months.** Early cancellations, returns and support calls from customers who never got the product working.
- **Retention.** Customers go quiet long before they leave, and the first sign is often the cancellation.
- **Staff.** It takes months before a new hire sells like the best person on the team, and when the best person leaves, what they knew leaves with them.
- **Decisions.** What customers actually think reaches the people who decide weeks late, and smoothed over.

None of this shows up as one line in a budget, so it rarely gets fixed. It gets paid for every year instead, in ad spend, discounts and hiring.

## How it's different

Most AI that talks to customers does one of two jobs. Marketing tools send what the company wants to sell, and they're finished when the customer buys. Customer service AI answers what people ask, and it's finished when the customer hangs up. You keep both.

Ours does a third job. It starts when nobody is calling: the months in between, when a customer isn't buying or complaining. It works out what would actually help each person, and when. Sometimes that's a heads-up before something runs out. Sometimes it's weeks of help before any offer. Most days it leaves them alone. That's street smarts.

### Where it sits: the reliance layer

- **Your CRM, billing and data platform** keep the record. It reads from them and writes back what it did. Nothing is migrated.
- **Your decisioning engine** picks the best offer from your catalogue for the people most likely to take it. Ours decides something earlier: whether to reach this person at all, with what, and when. Often that isn't an offer. When it is, the offer is still yours.
- **Your campaign tool** sends what you plan to send. Contact rules, such as how often someone may hear from you, when, and who is left out, are hard rules in the system, not instructions to the AI. When your systems tell it who already heard from you this week, it holds back.
- **Your service AI** answers when customers ask. Ours is the one that notices before they ask.

It adds a job to your stack, not a system to replace. If you stop, nothing else has to change.

### Why can't a general model do this?

It can write the messages, and they'll read well. We build on the same models, from OpenAI, Anthropic and Google. What a model can't give you is knowing which message, to whom, and when, or whether to send one at all. You can't read your way to that, and the models have already read everything. It comes from what real people did next, over months: from your best people, from our playbooks, built on what we learned works with earlier customers, and from testing every change on your own customers against a group that doesn't get it. What helps stays. What doesn't gets dropped. That's the part we bring.

### How to judge anyone selling you this

Whoever you talk to, including us, these questions separate a project that proves something from one that only looks busy:

1. **Is the success number fixed before launch, on a number you already track, against a group that doesn't get it?** If results are reported only after the fact, they will always look good.
2. **Who approves what it says, and how fast can you switch it off?** The answer should be your team, and immediately.
3. **What does it do when someone is upset, overwhelmed or in crisis?** A system that only knows how to sell will keep selling.
4. **Does it run alongside the tools you already have?** You shouldn't have to replace anything to find out whether it works.

## What it does

The same street smarts work wherever your company meets a person: on your website, in onboarding, through your staff in the store and on the phone, and in the months in between. Your AI gets them, and so do your people. Each use case below is measured on a number you already track, and each is a place to start. One contract and one meter cover all of them: credits are pooled across your projects, so each business unit can run its own use case against its own number. Your team runs it, and running it doesn't take an engineer.

### For your customers

- **Finding you (acquisition).** Visitors get a real answer, with the reasons, before anyone asks for their number. *Moves:* leads from the same traffic, cost per lead. *Usually owned by:* marketing or growth. *Setup:* one line of code on your website; live within weeks.
- **Choosing well (conversion).** It recommends what fits and says why, including what they can skip. *Moves:* conversion, returns, early cancellations. *Usually owned by:* sales or e-commerce.
- **Getting started (onboarding).** One step at a time, when they need it, until the product works for them. *Moves:* early churn, support calls. *Usually owned by:* customer experience. *Setup:* connects to the systems where your customer data lives, so it takes longer.
- **Staying (retention).** A heads-up before something goes wrong, and quiet the rest of the time, so they still read you when it counts. *Moves:* churn, response rate, usage. *Usually owned by:* customer experience or retention. *Setup:* connects to your customer systems, so it takes longer.
- **Coming back (win-back).** For people who went quiet: weeks of something useful first, the offer last. *Moves:* accepted offers, cost per reactivation. *Usually owned by:* marketing or CRM.

### For your staff

- **Daily coaching.** A few minutes a day of practice on the conversations people avoid, not the easy ones. *Moves:* conversion per person, time to ramp, attrition. *Usually owned by:* sales leadership or HR. *Setup:* quick; live within weeks.
- **Field sales.** Every morning: where to go, and which accounts are slipping before the quarter shows it. *Moves:* target attainment, accounts lost. *Usually owned by:* sales leadership.

### For your managers

- **The weekly picture.** What customers said this week in their own words, including what nobody wants to hear, and an AI to ask the next question. *Moves:* days from a customer signal to a decision. *Usually owned by:* leadership.

### How it works

1. **Notices.** It reads what is happening for each person from the systems a company already runs, such as billing, support and usage.
2. **Decides.** It works out whether to speak up now or stay quiet. Often the answer is to stay quiet.
3. **Helps.** It explains, reminds or answers on the channel the person already uses. When a person would be better placed to help, it hands over.
4. **Checks.** It looks at what happened next. What helped is kept, and what didn't is dropped.

## Customers

### Talk to a customer

Our customers take reference calls, in the US and in Europe. Jordan arranges them once we know which use case you're considering. Case studies, with what changed and how it was measured, are published separately as each customer approves them.

### Who's behind us

Our seed round was led by Feels Good Capital, a venture firm that invests only in companies with a positive social impact. They made impact one of our main KPIs. The protocols under **When it gets hard** are how we hold to it.

## Working with us

### How you start

- It starts with one audience and one group of people, measured against a control group that doesn't get it.
- Your team approves every playbook before it goes live.
- We agree the success number before launch, on a number you already track.
- How long setup takes depends on what it touches. Helping visitors on your website takes one line of code, and short daily lessons for staff are just as quick: both can be live in about two weeks. Keeping customers longer means connecting to the systems where your customer data lives, so those projects take longer.
- It then runs against the control group until the result is clear. For how long, and on what terms, ask our assistant on this page or write to [hello@mindsmiths.com](mailto:hello@mindsmiths.com).

### What you need on your side

Someone who owns the number, someone who knows your end users well enough to approve what the AI says, and access to the systems the use case touches. Some customers run it with one person, others with whole teams; it depends on the use case and what it connects to. On committed plans, we support your team through the first 30 days. Custom integrations beyond our standard connectors, and moving data over, aren't part of any plan.

### What it costs

One meter. You pay in credits, a unit like the tokens AI models charge in, for what the AI actually does. Every use case and every channel draws from the same pool, and there are no platform, maintenance or infrastructure fees on top.

| Plan | Price per credit | Term | Annual minimum |
|---|---|---|---|
| On-Demand | $0.002570 | none | none, pay for what you use |
| Annual Commit (30% off) | $0.001799 | 12 months | $21,588 |
| Strategic Commit (50% off) | $0.001285 | 36 months | $15,420 |

- **What a credit buys.** From about 1.5 credits for a proactive message (under $0.01) to about 510 for a four-turn conversation with the AI (about $1.31). Your dashboard always has the current figures.
- **How credits work.** Committed plans get 1,000,000 credits a month. Unused credits expire at the end of each month, and use above the commitment is billed at the On-Demand rate. Credits are pooled across your projects.
- **What's included.** The dashboard on every plan. On committed plans, onboarding support for the first 30 days and an account manager of your own. Channel costs, such as SMS and WhatsApp fees, stay with you.
- **Billing.** On-Demand is invoiced monthly, for what you used. Committed plans pay the annual minimum a year in advance. Prices don't include sales tax or VAT. Customers in the EU sign with Mindsmiths d.o.o. and pay in euros.
- **Renewing and stopping.** A committed plan runs its full term and renews unless one of us gives 30 days' notice before it ends. On-Demand can be stopped by either side with 30 days' notice.

These are our standard plans; what you sign in your order form is what counts. The rest, including what each activity uses, paying late and switching plans, is on [Pricing](/pricing).

### Once you have a result

You see the number against the control group. Then you decide what's next: more people, another use case, or a different approach. When you can stop depends on your plan (see **What it costs**). For anything else, ask our assistant on this page or write to [hello@mindsmiths.com](mailto:hello@mindsmiths.com).

If you stop, your data goes with you. You have 30 days to export it, free. Then we delete it from our systems within 30 days and from backups within 90, and confirm it in writing.

## When it gets hard

### Who decides what it says

Your team approves how the AI behaves and every playbook before it goes live. Consent and contact limits are enforced by the system, not left to the AI. And there is a kill switch.

### It reads the relationship, not just the message

The AI recognises where each relationship stands, and each situation has its own protocol:

- **Someone who prefers to be left alone.** It stops reaching out and stays available when they ask.
- **Someone leaning on it for what belongs with family, friends or a professional.** It cools things down and draws a boundary.
- **A charged topic.** Politics and social controversies: it stays out of them.
- **Someone in crisis.** Self-harm, abuse, danger: it hands over to a real person.
- **Someone who says stop.** No more messages until they agree again.

## Where we're not a good fit

We'd rather tell you now than in month three.

- **Things people just want done.** Paying a bill, resetting a password, tracking a parcel: a good service flow is enough. We're for decisions people care about getting right: a plan, a loan, a first road bike, a trip.
- **Small customer bases.** Our results come from comparing people who get it with people who don't. With too few customers, there aren't enough of either to show a result.
- **Help with a company you're a customer of.** If a company that works with us sent you here, write to them; they run their own assistant.
- **Pressure.** Countdown timers, manufactured urgency, blasting the whole list: we don't build it.
- **Decisions that need a person.** A loan, an insurance policy, a job, a diagnosis, one person's price: the AI doesn't decide these on its own, and our terms don't allow it. It can explain, remind and recommend; a person decides. Anything the EU AI Act bans or treats as high-risk is out too.
- **Some industries.** Our terms rule out tobacco, distilled spirits, weapons and ammunition, casinos, internet gambling and pornography.

## For your review team

One section per reviewer, so each can be forwarded on its own.

### IT and architecture

- **What it connects to.** It reads from the systems you already run (billing, support, usage, CRM) and writes back actions. No data migration; it runs alongside your current tools.
- **Channels.** On your website, by WhatsApp, RCS, SMS and email, and in your app: the same assistant that runs on your website can sit inside it. Where it fits, that gives your customers a better experience and costs less than sending messages.
- **On your website.** One line of JavaScript.
- **Hosting.** Microsoft Azure. If you sign with Mindsmiths, Inc., your data is in the US (East US, Virginia). Customers in the EU sign with Mindsmiths d.o.o., and their data stays in the EU (Germany West Central, Frankfurt). Backups stay in the same region. Our team in Zagreb runs and supports the platform.
- **Uptime.** 99.0% a month, measured by how many real requests fail. Planned maintenance doesn't count, and neither do outages at cloud, AI model and channel providers that are outside our control. If we miss it, credits go to your account without you asking: 10% of that month's credits below 99.0%, 25% below 98.0%, 50% below 95.0%. They're credits, not cash, and they last 12 months. Below 95.0% in any month, or below 99.0% in three months out of six, you can leave and get your prepaid fees back.
- **Scheduled messages.** Every month, 99% start no more than 15 minutes after they're due. If not, you get 10% of that month's credits back.
- **Incidents.** If your customers are affected, we respond within 2 hours, aim to fix it within 3, and update you every 2 hours until it's fixed. If only your team is affected, 4 hours to respond and 8 to fix.
- **Backups.** At least daily. We aim to lose no more than an hour of data, and to recover from a major incident within 3 hours. Planned downtime is announced a week ahead; urgent security fixes can't always wait.
- The details are in the [Service Level Agreement](/service-level-agreement).

### Security

- ISO 27001 certified. Our SOC 2 Type II audit is under way and on schedule.
- Data encrypted at rest (AES-256) and in transit (TLS 1.2 or higher).
- Each customer gets its own environment and its own databases, never shared with another customer. That includes what the AI reads and remembers.
- Everyone on our side who can reach production signs in with multi-factor authentication. Access is need-to-know, reviewed every year, and removed within 48 hours when someone leaves or changes role. Access is logged, and the logs are kept for 12 months.
- An independent penetration test every year, AI included. You can see the summary and what we did about each finding. Critical and high vulnerabilities are fixed within 30 days of a patch, medium ones within 90.
- Defences against prompt injection and jailbreaks, tested every year. Your data never goes into development or testing.
- If your customers' personal data is breached, you hear from us within 72 hours.
- Insurance: $1M per claim and $2M in total for technology errors and omissions, $2M per claim and $3M in total for cyber. Certificates on request.
- The details are in the [Security Addendum](/security-addendum).

### Data and privacy

- **Your customers' data stays put.** Conversations, behaviour and personal details stay in your own isolated environment. They are never pooled across companies and never train any model, ours or our AI providers'; our contracts with the providers rule it out.
- **Your playbooks stay yours.** The playbooks, rules and results your team builds are yours. They are never shown to or reused for another customer, including one in your industry.
- **What we learn from.** Only your team's chats with Smith, the AI assistant your builders use, and only anonymised, so nothing traces back to you. Never your customers' data, your playbooks, your rules or your results. Your administrator can switch Smith off at any time.
- **Our role.** For your customers' data, we act only on your instructions (processor under GDPR), under our [Data Processing Addendum](/data-processing-addendum). Customers in the EU sign it with Mindsmiths d.o.o.
- **Who else sees it.** Microsoft Azure, where it's hosted; our team in Zagreb, who run and support the platform; and OpenAI, Anthropic and Google, whose language models we use. We tell you 30 days before adding anyone. If you object on data protection grounds and we can't sort it out, you can leave and get unused prepaid fees back.
- **Deletion.** When the contract ends, you have 30 days to export everything. Then we delete it within 30 days, and from backups within 90, and confirm it in writing. What the AI reads and writes is kept for 24 months at most.
- **Audits.** We share our ISO 27001 certificate and audit reports. If a specific concern remains, you can inspect on site, once a year.
- **US financial services.** If you're a financial institution under the Gramm-Leach-Bliley Act, the Data Processing Addendum adds the safeguards it requires and commits us to work with your regulators.
- **Control groups.** Comparing people who get the AI with people who don't runs on your customers, under your consent and contact rules.

### Messaging and consent

We build and run assistants for the companies we work with, including over SMS and messaging apps. The same rules apply whether a message comes from us or from a company on our platform.

- **Opt in first.** We only message people who have agreed to be contacted on that channel. No bought lists.
- **Clear sender.** Every conversation says which company it comes from, right at the start.
- **Easy to stop.** To stop receiving messages at any time, reply STOP to the message or write to hello@mindsmiths.com and we'll opt you out.
- **Only what's useful.** Service updates, account and billing information, onboarding, support, and educational content you've opted into. How often you hear from us depends on your account activity.
- **No cost from us.** Message and data rates from your carrier may apply.
- **Your number.** Mobile numbers and consent records are used only to deliver this service. We don't sell personal information, and we don't share opt-in data with third parties for their marketing.

### Legal

- Every document in the contract, with what each covers: see **At a glance** at the top.

### Procurement

- Who you'd sign with: Mindsmiths, Inc., a Delaware corporation, or Mindsmiths d.o.o. if you're based in the EU (see **Company details**).
- Prices, plans and payment terms: see **What it costs**, and [Pricing](/pricing).
- Insurance: see **Security**. Certificates on request.

## About

### Where we are

Mindsmiths, Inc. is a US company, incorporated in Delaware, with people in San Francisco and Miami. We work from home offices, so there's no street address to visit. Our engineering team works in Zagreb, Croatia, at our subsidiary, Mindsmiths d.o.o.

### Who leads what

- **Mislav Malenica,** founder and CEO. San Francisco.
- **Jordan Yates,** VP of Sales. Miami.
- **Tomica Pustišek,** COO: pricing, contracts and operations.
- **Juraj Malenica,** Head of Engineering. Zagreb.
- **Martina Silov,** new opportunities and media.

### How it started

In the spring of 2020, Croatia's healthcare hotlines were getting four times more calls than they could handle. Our founder, Mislav Malenica, had spent 20 years as an AI scientist, building medical diagnostic algorithms used on millions of people. So the team set out to build an AI assistant with the Croatian government to take some of the load.

The plan was a diagnostic AI, and it fell apart almost at once. There was no data to train on and no medical guidelines to follow yet. Even the country's best epidemiologists couldn't tell flu from COVID without a test. So the assistant, Andrija, couldn't diagnose anything. It couldn't be book smart, because there were no books yet. What it could do was show up every day: answer what it could, say plainly what it didn't know, and ask how you and your family were doing.

More than 50,000 people used it in the first 15 minutes. Within a week, the load on the call centers had dropped by more than 25%. When the team called people to ask why, they kept hearing the same thing: with Andrija, the situation felt serious but under control.

Soon after, companies in telecom, healthcare, fintech and higher education asked for the same thing for the people they serve.

## Join us

Our engineering team is in Zagreb, Croatia. What we build goes live with national telecom operators, fintechs and an airport, in Europe and the US, usually soon after it ships.

The hard part isn't getting a model to write a good message. It's deciding, for one person on one day, whether to send anything at all, and checking later whether that was the right call.

You own a problem and ship it, end to end. If you'd like a detailed plan handed to you each week, it won't suit you. If you'd like to see your work in front of real customers within weeks, it will.

### Open positions

- **Software Engineer.** Junior or experienced. Design, build, test and ship with a lot of freedom. We use Python, Java and Vue.js most. Write to [juraj@mindsmiths.com](mailto:juraj@mindsmiths.com).
- **Product.** Work close to customers and shape what gets built: UX, product ownership and project management. Write to [juraj@mindsmiths.com](mailto:juraj@mindsmiths.com).

### Before you write

- Engineering and product roles are on site in Zagreb, at Zavrtnica 17.
- We hire students and full-time employees. Both work full time.
- We don't have machine learning or data science roles.
- There's no form, and you don't need a CV.


## Contact

**To see it running on one of your use cases:** write to Jordan Yates, VP of Sales, at [jordan@mindsmiths.com](mailto:jordan@mindsmiths.com), or ask our assistant on this page. The first call takes 30 minutes. Bring the use case you'd start with and the number you'd want it to move.

- **Pricing, contracts and reselling.** Tomica, COO. [tomica@mindsmiths.com](mailto:tomica@mindsmiths.com)
- **New opportunities and media.** Martina Silov. [martina@mindsmiths.com](mailto:martina@mindsmiths.com)
- **Technical questions and security reviews.** Juraj Malenica, Head of Engineering. [juraj@mindsmiths.com](mailto:juraj@mindsmiths.com)
- **Your data.** See, correct or delete the data we hold about you. [privacy@mindsmiths.com](mailto:privacy@mindsmiths.com)
- **Security, incidents and service levels.** [security@mindsmiths.com](mailto:security@mindsmiths.com)

For anything else, write to the person who handles it, or to [hello@mindsmiths.com](mailto:hello@mindsmiths.com) and we'll pass it on.

## Company details

- **Mindsmiths, Inc.** A Delaware corporation. 1000 N West Street, Suite 1200, Wilmington, DE 19801, U.S.A.
- **Mindsmiths d.o.o.** Our subsidiary. It builds, runs and supports the platform, and customers in the EU sign with it. Zavrtnica 17, HR-10000 Zagreb, Croatia. Company reg. no. 081287362, Commercial Court Zagreb. Director: Mislav Malenica.
- **Legal notices.** [hello@mindsmiths.com](mailto:hello@mindsmiths.com)

© 2026 Mindsmiths, Inc.