Blog · Published · Updated
How to run a local AI model for a trial
Short answer
You run a first trial of a local AI model on a computer you can administer: you choose one task and permitted material, check the model's terms, run it in an application such as LM Studio and record the result for assessment. Downloading the weights means using an existing model, with no training.
How we workDetails ↓
- 01Call30 minutes with an engineer, free
- 02PrototypeA working system on your process and data, free
- 03ImplementationTarget and fixed price in the contract; 50% at signing
- 04Acceptance50% once the agreed target is met
- analysis with an engineer and a free prototype
- €0
- team in ChatGPT or Claude, up to 10 people
- from €1,150
- automation, agent or application for one process
- from €3,000
We belong to the OpenAI and Anthropic partner programmes and learn through their official courses. What this means ↓
Author: Syntalith · 3 min read
Statistics
Local AI models: statistics 2026
2.0 mo.
payback of a €3,000 local application at 400 documents a month: €2,327 of work today, €796 with checking and the GPU server. Scenario.
502.14 s
According to a Syntalith experiment (2026), Qwen3.8-27B on a modified vLLM fixed a bug in Go code in 502.14 s, with a score of 100/100 by Syntalith's criteria.
Source: Syntalith, Qwen3.8-27B measurements on an RTX 3090 (2026) ↗
1
According to a Syntalith experiment (2026), all measurements ran on one RTX 3090 card with one active request, so they do not describe throughput for many users.
Source: Syntalith, Qwen3.8-27B measurements on an RTX 3090 (2026) ↗
48.8%
According to Eurostat (2025), 48.8% of EU companies that considered AI but did not use it named concerns about data protection and privacy.
11 of 13
According to the NoLiMa study (2025), at 32,000 tokens of context 11 of 13 models fall below half of the accuracy they reach with a short context.
Figures checked: 22 October 2026
A personal trial differs from a service for a team: on your own computer you check whether the model can handle the task, while a team also needs accounts, backups, access and maintenance. LM Studio describes working offline once a model is downloaded, which lets you check the task without sending the material anywhere.
- 01Taskone, with permitted material
- 02Modelthe licence and terms checked before downloading
- 03Runningan application for local models, working offline
- 04Resultrecorded next to the expected answer
1. Choose the task and the permitted material
To start, prepare a short, fictional document and a question whose answer you know. Then you can check whether the set-up works and whether the result keeps the meaning. Do not start with the whole company archive, which you cannot assess quickly.
Write down the operating system, the available memory and the type of graphics processor. The requirements depend on the model, the format, the length of the material and the application. We give no single minimum memory for all models and no list of hardware that guarantees a given speed.
2. Check the model's terms and download the files
Choose an official source for the application and a model with a licence description. The right to download a file does not settle every kind of commercial use or redistribution. Keep the name, version and terms of the chosen model, and leave the assessment of the terms to the person responsible for this purchase.
LM Studio is an example tool. Its documentation describes chatting with downloaded models and documents without the internet. Searching for and downloading models, engines and updates needs a connection. Documentation, checked on 9 October 2026. This describes what the product can do; this article does not test its installation.
3. Run the model and keep the trial result
In the application, choose a downloaded model in a supported format. Start with a short question, then add the prepared document. If the tool does not start the model or cuts the answer off, check the message and the maker's requirements before changing settings.
In the fictional material, a sentence says: „the date has not been confirmed”. A correct summary should keep that gap. Repeat the trial with a question about information that is not in the document. Record the result, the settings and the waiting time. A more detailed set is in the Polish language test for a local LLM.
4. Keep a personal trial apart from a team service
A chat that works on a laptop does not confirm support for several users, backups or separation of documents. Before you share the service with others, agree user identity, access, conversation records and responsibility for updates. Do not expose a local server to the internet only to make the first demo easier.
| Trial result | Next decision |
|---|---|
| The model runs, but the answers are wrong | Assess the model and how the task is prepared |
| The quality is enough, but the answer takes too long | Measure on the target configuration |
| One person works correctly | A concurrent load trial before buying for a team |
| The data needs wider control | A design of the whole flow, including logs and backups |
Syntalith can agree a private model rollout with acceptance criteria for the chosen task. The price depends on the environment, the data and the integrations. Before you buy a machine, read the comparison of a workstation and a server. A description of the work and the configuration is enough for the call; access to confidential files is not needed to choose the trial.
What it costs
- Today
- 400 documents × 10 min = 66.7 h (€2,327)
- With the application, an employee checks the result in 25% of the time
- €582
- Hardware in the company, counted in the scenario like renting a GEX45 server
- €214
- Saving
- €1,531 a month
Payback of €3,000
2.0 months
The volume, the time and the checking share are the scenario's assumptions, which we measure in the trial. €34.90 an hour is the EU average hourly labour cost for 2025 according to Eurostat. Price of the GEX45 server with 24 GB of GPU memory from Hetzner's website on 22 October 2026, excluding the one-off setup fee of €209.
- Where the application takes documents from and where it saves results
- What stays in logs, backups and the question history
- Who has access to questions, answers and outputs
- Who updates the model, the application and the document set
Where to start
In the free review we choose one task and 20 to 100 documents the team has already worked on, and within 2 working days you receive a written summary with a price. We deploy the local model as part of private LLMs, and the application around it as part of AI apps.
Worked example · Scenario
How we count the payback: a first local model
We run the same count for your team in the free review.
€4,188
a month for the work on 600 questions
EU cloud version payback: 0.8 mo.
Our assumptions are 12 minutes to find an answer in the documentation, and checking in 15% of the old time. €34.90 an hour is the EU average hourly labour cost for 2025 according to Eurostat, for companies with 10 or more employees. Model cost: Claude Sonnet 5.5, about 6,000 + 500 tokens per answer in the EU cloud version, at the ECB rate of 1.1205 USD per euro on 22 October 2026. We count the payback for an app with a model in the EU cloud (from €3,000 net). We price the model on your server after a call, and the cost of people's work stays the same.
See the full calculation →
