AI for business —
offline where data is sensitive
AI on your terms — offline where it matters, faster where it counts. Sensitive recordings and documents are processed locally — nothing is sent to the cloud. Repetitive team work gets faster with a configured Claude Code. Three services, one approach — technology matched to the problem, not the other way around.
Two problems AI can solve — if you deploy it right.
Problem one: data. Every file uploaded to ChatGPT, every recording sent to a cloud transcription service — leaves your network. Who has access? How long do they keep it? You don’t know. For law firms, that’s attorney-client privilege. For clinics — patient data. For manufacturing — specs on someone else’s server.
Problem two: time. Your team spends hours on work that AI could do in minutes. Transcriptions, translations, documentation, boilerplate, reports. You know AI can help — but configuring it is not a one-prompt job.
Both problems are solvable. Sensitive data? We process it locally — Whisper and language models on your hardware, no cloud. Repetitive work? We configure Claude Code for your processes, train your team, and make sure the tool actually works.
We match the tool to the problem. Offline where data is sensitive. Cloud where productivity matters. No dogma.
We didn’t start with a client. We started with our own servers.
Before deploying local AI for our first client, we presented the topic of sovereign Polish language models to NASK and the public sector. We mapped the open-source model ecosystem — from autonomy to computer vision — used by Polish development teams. In parallel, we tested Llama and other open-source models on our own hardware.
Since 2023 we have been deploying locally: RAG — questions answered from your own knowledge base — transcription and chat. We built our own servers for on-premise model serving and knowledge bases that map internal company processes. We wrote document anonymization that recognizes Polish identifiers: PESEL, NIP, addresses. We also built offline transcription tools as volunteers — for SocialDroneUA.
We deploy commercially with full awareness of the limits: we say where a local model is enough — and where it isn’t.
“Local serving” isn’t a slogan. Physically it means: data never leaves your hardware — you can pull the network cable and the system keeps working. On that hardware run transcription, document analysis, questions to your own knowledge base (RAG), anonymization, classification and reports.
Three services. One approach — AI matched to the problem.
Offline transcription
Whisper on your hardware
Audio and video to text without sending anything to the cloud. Court depositions, board meetings, HR conversations, medical consultations, interviews. Speaker detection, timestamps, batch processing. Recording goes in — text comes out. Data stays with you.
17 h audio → 3.5 h processing · 100% offline · No per-minute limits
Ready-to-use AI workstation
Local LLMs on a preconfigured Mac
A preconfigured Mac with local language models — Llama, Mistral, Qwen. Document analysis, report generation, text classification, Q&A over your own knowledge base (RAG). Instead of a week of configuration — unbox, power on, and you have working AI. No OpenAI account, no tokens. One-time purchase.
Plug & play · Llama / Mistral / Qwen · 100% offline · One-time purchase
Claude Code for teams
AI assistant in the terminal
Your developers are probably already using ChatGPT for code — on personal accounts, without oversight. We configure Claude Code for your team: company account, CLAUDE.md tailored to your processes, memory, workflows, training. Code review, documentation, tests, data processing — tedious work, 2‑3× faster. Honestly: Claude Code uses Anthropic’s API — it’s a cloud tool. Not for processing sensitive patient data. For accelerating your team’s work — yes.
Company account · Configuration + training · Anthropic API (cloud)
Before anything reaches cloud AI — we clean it locally
We detect Polish identifiers — PESEL, NIP (tax ID), names, addresses — and remove them from documents automatically. The whole process runs fully offline, on your hardware. A document leaves your network only once it contains no personal data.
It’s also a service for companies not yet ready for a full local deployment: we anonymize locally, and you keep processing the cleaned documents in cloud tools — reducing the risk instead of exporting it.
PESEL · NIP · addresses · 100% offline
Companies like yours already use this
Law firm
200 hours of deposition and hearing recordings per year. Before: manual transcription or cloud tools risking attorney-client privilege. After: Whisper transcribes locally, within the firm’s network. No fragment of any recording leaves the building.
Manufacturing company
Technical documentation in 4 languages, product specs, quality reports. Before: manual translations or Google Translate (with content leakage). After: local LLM translates, classifies, and summarizes — without sending specs outside.
Software house (8 people)
Every developer loses 2 h daily to boilerplate, tests, documentation. 8 people × 2 h = 80 h per week. After Claude Code deployment: the same work in half the time. Configuration per project, team training, post-launch support.
Medical clinic
Patient visit recordings — for documentation, not for the cloud. Whisper transcribes offline. Medical data stays within the clinic’s network. GDPR by design, no additional safeguards needed on the subcontractor’s side — because there is no subcontractor.
Honestly — is this for you?
This is for you if:
- You have recordings or documents that cannot leave your network — legal, medical, HR, manufacturing
- Your team wastes hours on repetitive tasks that AI could do in minutes
- You want AI matched to your problem — not “everything in the cloud” or “everything offline”, just what makes sense
- You’re looking for someone to configure it, train your team, and leave a working tool — you don’t want to fight with Python yourself
This is NOT for you if:
- You need a customer-facing chatbot on your website — we don’t build those and don’t plan to
- You’re looking for AI that will replace a doctor, lawyer, or consultant — it doesn’t work that way
- Your data isn’t sensitive and you just need ChatGPT — then ChatGPT Plus at $20/mo is cheaper and simpler, honestly
- You want a zero-cost solution — free cloud tools exist and are sufficient for many businesses
From conversation to a working tool
Conversation
You tell us what you process and what hurts. We assess whether local AI makes sense in your case. If it doesn’t — we’ll say so. No strings attached, no “free audit” pitch.
Tool selection
Whisper, LLM workstation, Claude Code — or a combination. We match tools to your processes and data, not to our price list. If ChatGPT is enough — we’ll tell you.
Deployment & training
Installation, configuration, testing on your data. We train your team for daily use. We don’t leave you with a PDF manual.
Support
If needed — maintenance, development, model updates. No long-term contracts. If not needed — the tool is yours.
What we DON'T promise
Common concerns
How much does it cost?
Depends on scope. Whisper, AI workstation, Claude Code — three different scales. Quote within 24 h of our conversation. Free, no strings.
Wait — Claude Code is cloud?
Yes. Whisper and local LLMs are 100% offline. Claude Code uses Anthropic’s API (cloud). Not for sensitive data. For team productivity — it’s excellent. We match the tool to the problem.
How does local AI compare to the cloud?
For general tasks — worse. For specific use cases — transcription, classification, data extraction — well enough. The difference: your data doesn’t leave the building.
Does it work on my hardware?
Whisper and LLMs need computing power — we offer pre-tested workstations. Claude Code works on any computer with a terminal.
Is local transcription as accurate?
No. Whisper Large locally achieves ~92–95% accuracy for Polish audio. For most business use cases, that’s sufficient. For 99%+ — it’s a first-pass tool, not a stenographer.
Is this GDPR-compliant?
Whisper and local LLMs — yes, data stays in your infrastructure. Claude Code — data goes to Anthropic. We match the tool to context.
Is Claude Code too complex for my team?
We train on-site. Configure CLAUDE.md for your processes. After 2–3 days the team works independently.
Do you anonymize documents before cloud AI?
Yes. PESEL, NIP, names, addresses — removed locally before a document leaves your network. Fully offline. Then cloud tools are safer to use.
Describe the problem.
We'll honestly tell you if and how we can solve it.