Articles grouped by business problem, industry, and system type.
With only a DPA in place, OpenAI keeps API prompts in its logs for up to 30 days and may process them outside the EU. What ZDR, MAM and an EU-region project change, and how Anthropic, Google, Microsoft and AWS compare.
A model on the plant's server is not enough to keep a formula a trade secret. Under Article 11 of Poland's Act on Combating Unfair Competition, the steps the company takes to protect the information count too. So an AI assistant may only retrieve documents the person asking is allowed to see. With 500 operator questions a month, a €3,000 local app pays back in 1.7 months.
Evaluating a private LLM is a repeatable test of the task on a set the model has not seen: typical cases, exceptions and missing data, with criteria agreed with the process user. The result shows errors by category and the share of cases passed to a person. An application on a local model costs from €3,000 net plus GPU hardware at Syntalith. At 600 documents a month and 8 minutes each, it pays back in 1.6 months.
Jev from TypeSafe AI is a model built for tasks where software needs a structured decision, for example assigning a message to the complaints or logistics department, instead of a paragraph of text. It is available on OpenRouter, and a correct answer format has to be told apart from a correct decision.
A local model on your computer helps with notes, code and searching your own documents without a server room. You start the installation with one use and a check of the computer: graphics memory and RAM decide how large a model fits. A small model on a laptop and an assistant for a team are different projects.
Choose your own model when data must not leave the company or the load is large and steady. Choose the OpenAI, Claude or Gemini API for irregular traffic. In the scenario a GPU server with upkeep costs about €493 a month. The same amount buys about 30,700 documents a month on the Claude Sonnet 5.5 API, and twenty times more on Haiku 5.5.
Open-weight models from 2026, such as Qwen, DeepSeek, Kimi, GLM, Gemma and Mistral, differ in size, the hardware they need and how they are run. You start the review by asking where the system will run and what work it has to do, and you choose by testing on your own tasks.
You compare a private LLM with an API on the cost of a correctly handled case. A Hetzner GEX45 GPU server with 8 hours of upkeep costs about €493 a month. A case with 3,000 input tokens and 600 output tokens costs about €0.0107 on the Claude Sonnet 5.5 API, so the server wins from about 58,600 cases. Every 500 extra corrections at 3 minutes each, however, cost €873.
A private LLM lets you choose where the answer is computed, but the app has more parts: text extraction, the index, logs and backups. Before you choose EU hosting or an on-premise server, draw a data map from the document to the answer. At Syntalith an app on a local model costs from €3,000 net plus GPU hardware. At 400 questions to documents a month of 12 minutes each, it pays back in 1.6 months.
When you change a private AI application, the history export must contain the questions, the documents and which version of the result was used, because a file with answer text alone does not let the next person continue the work. You check portability with a trial export and read-back before you choose the tool and before you shut the old service down.
A local speech recognition model prepares a working text for captions in internal videos that may not go to an outside service. Before publication you still need proofreading, a split into readable segments and timing each segment to the right moment in the video. Compare a ready-made caption editor first.
A local AI app can group patients' questions to reception into topics that the manager checks and corrects. First establish whether the data set is truly anonymous, because removing names is not enough. At Syntalith an app on a local model costs from €3,000 net plus GPU hardware. At 1,200 messages a month of 2 minutes each, it pays back in 3.6 months.
When the result can wait until morning, a local AI app processes a batch of documents in the background and returns a list of descriptions for review. The employee sees which files were done, which are waiting and where reading failed. At Syntalith an app on a local model costs from €3,000 net plus GPU hardware. With 1,000 documents a month at 6 minutes each, it pays back in 1.2 months.
Local AI can read a table from a PDF when a tool such as Docling reads the layout and an employee checks the result next to the original. A local GPU server costs about €214 a month, more than a cloud model, and the documents never leave the company. With 400 tables a month at 12 minutes each, a €3,000 app pays back in 1.6 months locally and 1.4 through the API.
Several companies can use one server with a local model if each manages its own users and materials with no access to the others' data. Before connecting them, decide which decisions are shared and which each company makes alone. At Syntalith an app on a local model costs from €3,000 net plus GPU hardware. At 600 questions to documents a month of 10 minutes each, it pays back in 1.2 months.
A local app prepares a copy of a text to share. It marks passages to remove or replace, shows the proposals next to the original, and an employee approves the version before it is passed on. Removing a name is not enough when the story in the text still points to a person or a company.
A local slide search finds a slide from a description of what the employee remembers and shows the thumbnail, the presentation's name and the pages around it. The employee opens the original and decides whether the material is current and may be shown to a client. At Syntalith an app on a local model costs from €3,000 net plus GPU hardware. At 400 slide searches a month of 10 minutes each, it pays back in 2.0 months.
Local AI without a GPU administrator works when the contractor takes over monitoring the app, faults and updates, and the company keeps decisions on document access and sign-off of changes. Maintenance is described the way an employee uses it, from opening the app to the result. At Syntalith an app on a local model costs from €3,000 net plus GPU hardware. At 400 requirement summaries a month, it pays back in 1.6 months.
A local multilingual search finds a passage in an English or German document for a question asked in Polish, because it compares meaning instead of words, and shows the original alongside. At Syntalith an AI app on company documents costs from €3,000 net. With 300 questions a month at 12 minutes of searching each, it pays back in 1.7 months.
Downloading a model and permission for a specific way of using it are two separate questions. Before you order an app, describe who uses it, where the model runs and what the customer receives, then check the terms of the specific variant. One family can carry an Apache 2.0 licence, a Llama 3.1 licence and a non-commercial variant at once. An app on a local model costs from €3,000 net at Syntalith.
Private models
We select the hardware and model, deploy it and measure quality on your examples. You can start with a computer you already own.
Times are shown in your own time zone. We work with clients across time zones.
See private modelsOpenAI makes ChatGPT and Anthropic makes Claude. We belong to the partnership programs of both companies, so we work with their tools every day and learn from the creators themselves.
Claude Partner NetworkParticipation in programs does not mean that OpenAI or Anthropic endorses our services.