We can't paste contracts or client data into ChatGPT
The model runs on a server you control. Prompts and documents stay inside your infrastructure, and we connect no external APIs.
A private AI model on your own server: answers from your documents, data stays with you
We deploy an open language model (LLM) on your server or a dedicated rented one and connect it to your documents. Contracts, patient records, financial reports and source code never go to a third-party API.
Quote and schedule after your brief
Who it's for
- Law, healthcare and finance firms that cannot send documents to outside AI services
- Companies with a large internal knowledge base, contracts or technical documentation
- Engineering teams that want an AI assistant over their own code without leaking it
- EU and UK businesses that need to keep data inside their own infrastructure for GDPR reasons
Problems we solve
Staff are quietly using public chatbots anyway
We give them a convenient internal chat with access to your documents. When people have their own tool, the urge to use someone else's fades.
Enterprise AI vendors in the US and EU charge like a small department
We build on open models and a proven stack, so the price stays well below a classic agency. We prepare a quote after your brief.
It's unclear whether this will deliver value or become an expensive toy
We start with a pilot on one use case and one metric, such as time to find an answer in documents. You decide on scaling after seeing results.
What's included
- Task and data review: what the model must know, who gets access, which legal limits apply
- Open model selection (Llama, Mistral, Gemma and others) for your language, task and hardware budget
- Hardware sizing: how much GPU memory you need, and whether buying or renting a dedicated server makes more sense
- Deployment with vLLM or Ollama in Docker, reachable only from your network
- RAG (retrieval-augmented generation: the model first finds relevant passages, then answers and cites them)
- Document ingestion for PDF, Word, spreadsheets and scans with OCR, stored in a Qdrant or pgvector vector database
- A web chat interface (for example Open WebUI) with logins and permission levels
- A test set built from your real questions to measure answer quality
- Admin guide and handover of all access and files to you
- Connection to AI agents, your CRM or workflows when needed
Choose the scope
Use a package to describe what you need. We agree on the quote and schedule after reviewing your brief and requirements.
01
Pilot
Teams that want to test the idea on one use case, such as search across internal documents
- One model from your shortlist, deployed on one server
- RAG over one document set (up to a few hundred files)
- Web chat for a small group of users
- Test question set and quality report
- Full access handover
02
Production rollout
Companies where the whole team or department needs the model daily
- Several data sources: file storage, email, CRM, knowledge base
- Roles, permissions and a request log
- Quantization (compressing the model so it runs faster on smaller hardware) and speed tuning
- Integration with your CRM or messenger
- Staff training and documentation
03
Fine-tuning and agents
Companies with specialized domain language and a large body of clean data
- Fine-tuning on your prepared examples
- Several AI agents that take actions: draft documents, fill fields, create tasks
- Quality evaluation before and after fine-tuning
- Backup and model update plan
- Preparation for an internal security audit
How we work
Brief and requirements
You describe the task and current workflow. We clarify requirements, choose a first stage and prepare a quote.
Pilot with a metric
Together we pick one use case and a measurable goal, such as minutes spent finding an answer in policies. We record the baseline first.
Server and model
We size the hardware, deploy the model and connect your documents. All accounts and the server are registered to you under the agreed terms.
Quality testing
We run the model on your real questions and fix chunking, prompts and retrieval. Weak spots are shown openly.
Handover and support
We train your admin and hand over documentation and access. You can keep us on for support or run it yourselves.
Scope and limits
- We don't promise the level of the largest cloud models: open models on one server are weaker, especially at complex reasoning
- We don't guarantee zero errors: models can make things up, so answers come with source links and critical decisions stay with a person
- Server and GPU costs are not included: they are a separate bill we calculate up front, and the hardware is yours
FAQ
Does our data really stay on our server?
Yes, when the model runs on your server with no external APIs. We block outside access, disable telemetry where possible and document the data flow. If you need an audit, we can prepare a document for your security team.
What hardware do we need?
It depends on the model. Smaller models run on a single 24 GB GPU, while larger ones need several or data-center cards. After the brief we compare options: buying, renting a dedicated server, or hosting in your own data room.
How is a private model worse than ChatGPT?
For document search, summaries, drafts and classification, open models are often good enough. For hard reasoning or rare languages, large cloud models are usually stronger. We show the gap on your own questions before you scale.
Do we need to fine-tune the model?
Usually not. In most cases RAG is enough, since the model pulls answers from your documents. Fine-tuning makes sense for a specific style or narrow terminology, and when you have hundreds of good examples.
What do we need to provide?
Access to your documents and someone who knows what the correct answers are. You also need an admin or IT contact for the server and network. The cleaner the documents, the better the answers.
What does it actually cost?
We prepare a quote after reviewing your brief and requirements. A first stage can cover: one model from your shortlist, deployed on one server; RAG over one document set (up to a few hundred files). Scope, payment and schedule are agreed before work starts.
Start with your task
Describe your business, the outcome you need and your current workflow. We will clarify the requirements and prepare a quote after your brief.