Skip to main content
AI Demo · Local AI in Practice

An AI assistant in action

Four business scenarios. Documented response times. Editorial changes disclosed.
Processed locally on our own hardware, without an external model API or cloud AI.

4
Business scenarios
21.3s
Avg. response time
Local
On our own hardware
No
external model API used in this test
qwen3:14b
model used
3
editorial changes disclosed

What can local AI actually do?

For this study, we put our own AI assistant to work on real business tasks: from client quotes and customer communication to technical support. This is our own benchmark (proof of concept), not a client project. All test content was processed on our own hardware using locally executed models. No external model API or cloud AI was used for processing. The displayed times are single measurements from the original test run, not a performance guarantee.

The Results

Complete Client Quote

A full quote with package recommendation, pricing table, and signature

29.7s
Response time
Local AI
Business
A complete quote in under 30 seconds
REQUEST
"A carpentry business in Dresden needs a website. Write a complete quote."
AI RESPONSE
Subject: Your new website for your carpentry business - a quote from GTSH Engineering Dear Mr. Müller, Thank you for your inquiry! A carpentry business without a website - let's change that. My recommendation: a company website at a fixed price after a brief analysis, in this example around €2,500 (demo scenario) ✓ Professional design - modern, credible, a good fit for skilled trades ✓ Up to 15 pages - home, services, gallery, contact ✓ Project gallery - your best work as a showcase ✓ Search engine optimization - so customers in Dresden can find you ✓ 3 months of support - you're not on your own after launch Timeline: Your site will be live in 3-4 weeks. Best regards from Saalfeld Randolf Hahn, GTSH Engineering

Privacy argument

A compelling case for local AI vs. ChatGPT

12.7s
Response time
Local AI
Sales
Privacy argument for local AI

Response edited for accuracy

REQUEST
"Why should I pay for local AI when ChatGPT is free?"
AI RESPONSE
With cloud AI, how your data is processed depends on the provider, the plan, and the settings you choose. With professional business and API offerings, business data is often not used for model training by default. Even so, the data is processed outside your own infrastructure. A local AI reduces that external transfer and gives you more technical control over systems, documents, and access rights. Which architecture makes sense under data protection law has to be decided per use case.

Follow-Up Email

A professional follow-up email to a business partner

19.0s
Response time
Local AI
Communication
A personalized B2B email in 19 seconds

Response edited for accuracy

REQUEST
"Write a follow-up email to a client. We sent an AI proposal but haven't heard back yet."
AI RESPONSE
Subject: Quick follow-up on your digital colleague - our proposal from February Dear Mr. Schmidt, I hope this message reaches you at a quiet moment. In early February, I introduced you to our concept for a local AI assistant - a capable AI system working as a digital colleague, supporting your team with customer inquiries and product advice. Model processing runs on your premises; optional external services are assessed separately. I can imagine things are busy right now. Still, I wanted to check in briefly: are there any open questions about the proposal? If the timing isn't right at the moment, that's perfectly fine - just let me know. Best regards Randolf Hahn

Technical Support

Website performance explained for non-techies

23.8s
Response time
Local AI
Support
Response in 23.8 seconds

Response edited for accuracy

REQUEST
"My website loads slowly, scoring only 45 out of 100. Please explain in plain terms."
AI RESPONSE
A score of 45 leaves room for improvement - but don't worry, it can be fixed. The 5 most common culprits: 1. Oversized images - like mailing a letter with a whole photo album attached 2. Too many add-ons - every plugin has to "grab a coffee" first 3. A slow server - one cash register for a shop with 50 customers 4. No caching - a regular customer showing ID on every single visit 5. Outdated technology - loading things nobody needs The good news: Image optimization, caching, and plugin cleanup can improve the score. The scope and result depend on the specific website. My suggestion: send me the link - I'll run a free quick check.

How to read the measurements

The four scenarios shown each document one test run with qwen3:14b. They are a snapshot, not a comparison of different models.

Documented in this demo

Four individual tests

  • ✓ Model: qwen3:14b
  • ✓ Four business scenarios
  • ✓ 12.7 to 29.7 seconds
  • ✓ Three edits disclosed
21.3s avg. in this test
Not established by this demo

No generalization

  • – No multi-model comparison
  • – No test of automatic model selection
  • – No availability measurement
  • – No guaranteed response time
Snapshot rather than an SLA

Ready for your own in-house AI?

Comparable results are possible in your business as well, depending on hardware, model, and use case. With a properly configured local solution, the content submitted for processing is handled within your infrastructure.

Book your free 30-minute call