Your AI.
Your building. Your data.
Client files in a US cloud? Most people would rather not. Local AI runs on a computer in your office or on your own server. The data stays where it already is, and the AI comes to it. Sounds old-fashioned, but it is the most modern thing you can do with sensitive data right now.
What is local AI?
Local AI means AI models run on your own hardware instead of in a provider's cloud. Text, documents and customer data never leave your company, which makes GDPR-compliant AI much easier. Open models such as DeepSeek V4, GLM-5.3, Qwen, Llama, Mistral or Gemma can be installed with tools like Ollama or LM Studio. In a Bitkom Research survey, 48 % of German companies named data protection requirements as an obstacle to using AI, and 39 % fear their data could end up in the wrong hands. Local AI addresses both.
- No cloud upload
- GDPR-compliant
- Works with your own data
What clients say.
Thank you very much for the very successful and motivating day. The goal we discussed has been achieved, everyone is motivated for AI.
Thanks again for the great workshop. I want to get started asap.
I had my website completely redone by seeside.ai, and the result really impressed me. […] If you want a website that thinks along, automates and grows, seeside.ai is exactly right.

In your own building
Every person. One agent. One network.
Why companies run AI locally.
GDPR made easy
No data transfer to third countries, no data processing agreement with an AI provider. The data simply stays in-house.
Local AI with your own data
The AI searches your contracts, manuals and old quotes. Without anyone uploading anything anywhere.
For confidential work
Law firms, tax advisers, medical practices, engineering offices: wherever professional secrecy applies, local AI is often the only option.
Predictable costs
No per-word fees for every request. You pay for hardware and setup, then the AI works as often as you like.
What hardware does local AI need?
Less than many fear. The key figure is the memory of the graphics card (VRAM) or, on Apple computers, the shared memory. Rule of thumb for compressed models:
Small models (7 to 8 billion parameters)
about 8 GB VRAM
Sorting e-mails, summaries, simple drafts
Medium models (around 30 billion)
about 24 GB VRAM
Document search, longer texts, quotes
Large models (70 billion and up)
40 GB VRAM or more
Demanding analyses, close to cloud quality
Rule of thumb, not a guarantee: the exact figure depends on the model, its compression and how much text it has to handle at once. In the first call we find out which size your task needs.
Open-source frontier models 2026: DeepSeek V4, GLM‑5.3, Qwen and more
Open models have caught up. In the Artificial Analysis Intelligence Index of October 2026, DeepSeek V4.1 Flash and GLM-5.3 rank among the three best open-weight models. Open means: the weights are freely available, you can run the model on your own hardware, and no request leaves your company.
| Model | Size & licence | Good for | Hardware |
|---|---|---|---|
| DeepSeek V4 / V4 FlashDeepSeek | About 284 to 304 billion parameters, MIT licence, up to 1 million tokens of context | Agents, long documents, coding | Server with several GPUs |
| GLM-5.3Z.ai | About 744 billion parameters, own licence | Coding and agent tasks | Large GPU server, also rented in Germany |
| Qwen3.8-27BAlibaba | 27 billion parameters, Apache 2.0 | All-rounder for texts, e-mails, document search | One server with a strong GPU |
| Kimi K3Moonshot AI | Very large, about 1.4 TB of memory in 4-bit | Frontier performance, long tasks | Data centre, usually too big for SMEs |
| gpt-ossOpenAI | Open models from OpenAI in two sizes | All-rounder, tool use | Small version on a workstation |
| Llama 4Meta | Scout version with very long context | Long documents, images | GPU server |
Chinese AI and data protection
DeepSeek, GLM, Qwen and Kimi come from China. Anyone who uses the provider's app or online API sends their data to that provider. If the open model runs on your own server instead, no request goes to China: the weights are just a file. You get the strength of the models without the data protection problem.
Which size fits?
Honestly: the very largest models need hundreds of gigabytes of graphics memory. For most companies a medium model such as Qwen3.8-27B on one server is enough. If you want the big ones, rent a GPU server in Germany. We set up both.
Sources: geotoolbox.ai (Artificial Analysis Index, Okt. 2026), gate.ai, wavect.io, shattered.io
Honest comparison: local AI or cloud?
Speaks for local AI
- Data stays in-house
- Works without internet
- No per-request fees
- Full control over model and updates
Worth knowing
- Hardware costs money up front
- The very largest models only run in the cloud
- Someone has to maintain it (that can be us)
Our tip: sensitive data goes local, everything else may use the EU cloud. Many companies end up with a mix, and that is perfectly fine.
We are self-hosted ourselves: our workflows run on our own n8n server and image generation runs locally on our own graphics card. We know the noise of the fan and the joy when it works.
How quickly cloud dependency hurts was shown by the Microsoft outage in November 2026: e-mail, calendar and Copilot stood still worldwide for almost a day. Local AI kept running. What a full switch looks like is on our Microsoft 365 alternative for business page, how to end the subscription in our guide how to cancel Microsoft 365.
Key points at a glance
- Local AI runs AI models on your own hardware, so data never leaves the company.
- This makes GDPR-compliant AI much easier, especially with confidential data.
- Open models such as DeepSeek V4, GLM-5.3, Qwen, gpt-oss, Llama or Gemma run with Ollama or LM Studio.
- Small models need about 8 GB of graphics memory, large ones 40 GB or more.
- Seeside sets up local AI for companies, from Radolfzell on Lake Constance and remotely.
Questions about local AI
What is local AI?+
Is local AI GDPR-compliant?+
Which local AI models are suitable for companies?+
What hardware do I need to run AI locally?+
Can local AI work with our own data?+
Is a local LLM as good as ChatGPT?+
How do I install local AI?+
What does local AI for companies cost?+
Which open-source AI is the best in 2026?+
Can I run DeepSeek V4 locally?+
Is DeepSeek GDPR-compliant?+
What is GLM-5.3?+
Do you set up local AI on Lake Constance?+

Does local AI fit you?
Thomas Honold checks it with you in a free 30-minute call: task, data, hardware. Honestly, including when the cloud would be the better choice.
+49 171 3838860