Assistants over your documents
- BeforePolicies, contracts and manuals searched by hand, or by asking around.
- AfterAsk in plain words. Get the answer with links to the exact passages.
Home / Services / AI integration
AI integration
Nobody had to ask you.The answer came with its source.
We build AI assistants, copilots and agents that answer from your company's own documents. They work inside your CRM, helpdesk and ERP, and we test them on your real questions before anyone relies on them.
Free blueprint · in your inbox within a few hours · no call needed
- Senior team, you keep the code
- EU data, GDPR-first
- Fixed-price start
Sound familiar?
The answer exists. It's in a PDF from two years ago, a shared drive and one colleague's memory. So everyone asks that colleague.
Many AI pilots stall at the demo. It sounds clever, but nobody can check the answers. Our assistants answer from your documents and show where each answer came from. They live in the tools your team already opens.
Show me examples for
Pick one. The examples on this page will use it.
What we build01 02 03 04 05 06
AI that works from your knowledge, inside your tools.
Each one has one clear job, a person who owns it and a test set that shows how often it gets the answer right.
Copilots in your CRM and helpdesk
- BeforeReplies and account summaries written from scratch.
- AfterA first draft waits in the ticket or the deal. Your people edit it and send it.
Agents that take the next step
- BeforeSomeone reads, decides and clicks through three systems.
- AfterThe agent prepares the action. A person approves it with one click.
Ticket and email triage
- BeforeEvery message read just to decide who gets it.
- AfterTagged, prioritised and sent to the right person, with a suggested answer.
Reading documents into your ERP
- BeforeFields copied from contracts, orders and forms.
- AfterRead, checked against your rules and written into the right record. Then read back to confirm.
AI inside your own product
- BeforeA request for “something with AI” and no way to judge it.
- AfterA scoped feature. Test results, running costs and limits are known before launch.
It works inside what you already use: CRM, helpdesk, ERP, email, shared drives, wikis, Slack and Teams.
Customer-facing AI
Need it to talk to your customers?
Chatbots and voice agents for your website, WhatsApp and phone line. They answer day and night from your own documents and book visits in your calendar. Anything unusual goes to a person, with a summary.
See chat & voice agents
One question, answered from your own documents
A question someone asks every week.
It usually costs a colleague twenty minutes of searching. Or it interrupts you.
Searched in your own sources.
Policies, contracts and the CRM. Only the files this person may already open.
Outdated files are left out.
The 2021 pricing deck is skipped. The current policy is used.
An answer in plain words, with its sources.
Every sentence links back to the passage it came from, so anyone can check it.
Not sure? It says so and asks a person.
A missing credit check goes to finance as a task. The assistant does not guess when money is involved.
Illustrative3 sourcescited in one answer
Company assistantAnswered
Question from salesCan we give a new distributor 60-day payment terms?
Sources searched- 1Credit policy v4.pdfShared drive · FinanceUsed
- 2Distributor terms 2026.docxShared drive · SalesUsed
- 3Account notesCRMUsed
- 4Pricing deck 2021.pptxSupersededSkipped
- 5Board minutesNo access for this userSkipped
Yes, up to 60 days once a credit check is signed 1. New distributors start on 30 days unless the account owner agrees otherwise 2. This account has no credit check on file yet 3.
To a personCredit check missing: task created for finance.
Data, privacy and the AI ActEU hosting Permissions Sources People decide Tested AI Act
Written down before anything is built.
In the Sprint we agree in writing what the AI may see, where the data goes and who checks its work.
Data stays in the EU where possible
EU-region model hosting by default. A self-hosted model when the data needs it. A written map of which data goes where.
It sees what the user may see
The assistant only retrieves what the person asking can already open in the source system.
Every answer can be checked
Answers link to the passages they used. When nothing supports an answer, it says so.
A person approves what matters
Agents prepare actions. Anything involving money, customers or legal risk waits for a person to approve it.
Measured on your real questions
Before launch we test it on hundreds or thousands of questions your team really asked. You see the score and where it still fails.
Transparent by design
People know when they're dealing with AI. We document what the system does, which data it uses and its limits.
We're engineers, not a certification body. If a use case may count as high-risk under the EU AI Act, we flag it in the Sprint so your legal counsel can assess it.
How a project runs
Fixed price to start. A measured result before anything bigger.
You can stop after any step and keep everything we built.
1–2 weeks · fixed price
Sprint
We collect real questions and documents and pick the first use case worth building.
4 weeks · fixed scope
Pilot
One assistant or agent working in your real systems, scored on your real questions.
When the pilot pays
Integrate
We roll it out to the team, connect more sources and train the people who use it.
Monthly
Operate
We watch the answers, update sources and models and fix what drifts. You keep the code either way.
Want this for one of your processes?
Tell us the process. The free blueprint is in your inbox within a few hours.
What it costs
No hourly meter. You see the price before each step.
Every step is priced on its own, in writing. You can stop after any of them and keep what we built.
- BlueprintFree
A written plan for one process, with what the manual work costs you today.
- SprintFixed price
One to two weeks. Quoted in writing after a free call, before you commit.
- PilotFixed price, fixed scope
Four weeks, one process in your real systems, a result we measure together.
- OperateMonthly fee
Hosting, monitoring, fixes and improvements. You keep the code either way.
What moves the price
- How many systems it has to connect
- How clean and reachable your data is
- How many cases need a person to decide
- Where it has to run: EU cloud or your own servers
The blueprint shows what the manual work costs you a year, so you can weigh any quote against it before you spend anything.
Questions owners ask
Before you request a call.
What does it cost?
The blueprint is free. The sprint (1–2 weeks) and the 4-week pilot each have a fixed price, quoted in writing before you commit. After launch, a monthly fee covers hosting, monitoring and fixes. The price depends on how many systems it connects, how clean your data is and how many cases need a person, not on hours logged.
Will it make things up?
Language models can, so we design against it. The assistant answers from passages it found in your documents and shows them. When nothing supports an answer, it says so and points to a person. Before launch we test it on real questions your team asked, labelled with the right answer. You see the score and the failures before you decide.
Where is our data hosted, and is it used to train AI models?
We are an EU company based in Poland. We prefer EU-region hosting, choose providers and settings that do not use your data to train their models, and use a self-hosted model when the data calls for it. In the Sprint we write down which data goes where, before anything is built.
What about the EU AI Act?
We build to its transparency rules by default. People know when they are dealing with AI. We document what the system does, which data it uses and where its limits are. If a use case may count as high-risk, for example in hiring or credit decisions, we flag it in the Sprint so your legal counsel can assess it. We are engineers, not a certification body.
Which systems can you connect it to?
Usually the ones you already have: CRM, helpdesk, ERP, email, shared drives, wikis, Slack and Teams. If a system has an API, a reliable export or a database we may read, we can normally connect it. We confirm each connection in the Sprint, before you commit to the pilot.
How long until something runs?
The Opportunity Sprint takes 1 to 2 weeks at a fixed price. The pilot that follows takes 4 weeks and ends with one assistant or agent working in your real systems and a score we measured together on your questions.
Do we own what you build?
Yes. The code, the prompts, the test questions and the documentation are yours and live in your own accounts. Another team can take it over at any time.
For AI founders
Building an AI startup? We'll be your technical team, for cash, equity or both.
- Fractional CTO and architecture
- LLM stack and MVP build
- Scaling, and getting ready for investor tech due diligence
Free · in your inbox within a few hours
Tell us which process eats your week. Get a free automation blueprint.
- Process breakdown
- Automation options
- 4-week pilot plan
Prefer to talk? Request a 30-min call.
You talk to the senior engineer who would build it, not a salesperson. Bring one process; we tell you plainly whether it's worth automating.
Request a 30-min call30 minutes, free. We reply within a few hours with times that suit you.