Reviews · Digital · AI on your own computerWe use this
Ollama
The easy way to run an AI model on your own machine, if the machine is up to it. It costs disk space and patience, not money.
We use this since early 2026
Illustration, made with an AI image tool: a desk lamp late at night. It shows the mood, not the software.
What we did with it1 of 11
The chat inside our domain manager runs on Ollama, on our own desktop, with a model called qwen3.5:9b. We ask it things like which names renew next, and it answers from our own database.
What we measured2 of 11
What we measured, November 4, 2026
The machine
Intel i7-14700K, NVIDIA RTX 4070 with 12 GB, 64 GB of memory
The model
qwen3.5:9b, a 6.6 GB download
First question after a rest
8.2 seconds, 7.3 of them loading the model
The next question
1.7 seconds for a short paragraph
Output speed
About 75 tokens a second, both times
Two questions on one afternoon, on one strong desktop. A number to compare with, not a benchmark.
The good3 of 11
Your words stay on your machine.
With a local model nothing leaves the computer, and asking it a thousand times costs the same as asking once.
The good4 of 11
Fast enough to talk to.
Once the model is loaded, the answers come as quickly as you can read them on our desktop.
The good5 of 11
Your own programs can use it.
It answers at a local address, so our app talks to it the way it would talk to any paid AI service.
The good6 of 11
It is open source.
The software is under the MIT license.
Illustration, made with an AI image tool: a computer at home, a cat asleep beside it. It shows the mood, not the software.
The catch7 of 11
read the name
Not every model in its list is local.
Some entries run on Ollama's own servers and carry "cloud" in the name. Those send your words out. Read the name before you assume it is private.
The catch8 of 11
Our figures are from a strong desktop.
The model has to fit in memory. We have not tried it on a laptop, so we cannot tell you how that goes.
The catch9 of 11
The tool is not the brain.
Answers are only as good as the model you pick. Ours handles questions about our own data well, and we still pay for bigger AI tools for design and writing.
The catch10 of 11
Some models think out loud.
Ours reasons at length before answering unless told not to. For a short chat we switched that off, with think set to false in Ollama's own chat address.
Our verdict11 of 11
Made to buy
The easy way to run an AI model on your own machine, if the machine is up to it. It costs disk space and patience, not money.
Worth the money for the person described.
$0 to run it locallyseen November 4, 2026
Made to buymadetobuy.com
The drawing shows only what the words say: what we did with it, what we measured, what went wrong.
Where the figures come from
OllamaPricingopened November 4, 2026Running locally is in the $0 column. The cloud plans are the paid part.
Digital Real EstateOur own timing November 4, 2026Run on the machine above through Ollama's chat address, with the two questions timed by a small script.
