Three programs, two modes of operation
Ollama and LM Studio are different products with one thing in common: they load a language model onto your computer and run it there. Biwak does not download a model; it sends every model request via the Biwak account to a model provider. Whether Ollama or LM Studio gives you a fully sealed-off workflow also depends on connected tools, search and your configuration.
| Biwak | Ollama | LM Studio | |
|---|---|---|---|
| Where the model computes | At the model provider in the EU: on Microsoft Azure in the EU Data Zone, alternatively via Prem AI (contracting party PREM SA, Switzerland). | On your hardware. There are also cloud models that Ollama runs on its own servers. | On your hardware. |
| Without internet | New model requests need a connection to Biwak. Files, conversations and checkpoints stay on the computer. | Local models, yes. Cloud models and web search can be switched off completely in the settings. | Chat, working with documents and the local server run without internet; model search, downloads and app updates need it. |
| What happens to inputs | Question, context and the file contents used go to the model; content is not used for training. Data path. | In local operation, Ollama says it does not see the inputs. Ollama processes cloud requests, but according to its FAQ does not store or log them. | According to LM Studio, what you enter in the chat does not leave the device. |
| Hardware | macOS 14 or later on Apple Silicon or Intel, Windows 10 (1809) or later, 64-bit; the model needs no computing power on the device. | Depends on the chosen model; the larger the model, the more memory. | Mac only with Apple Silicon and macOS 14 or later, 16 GB of RAM recommended; on Windows, a processor with AVX2, 16 GB of RAM and 4 GB of video memory recommended. |
| Setup | Install the app, sign in with a Biwak account, choose a working folder. A custom model endpoint is not provided for. | Download a model, check memory requirements and model licence; set up and test document workflows separately. | Search for and load a model in the app, check memory requirements and licence. |
| Cost | The app is free; model requests run via a monthly subscription from €19 a month. Plans. | The program is under the free MIT licence, and local operation is included in the Free plan. Cloud models use up credit; Pro costs $20 and Max $100 a month. Plus hardware, electricity and maintenance. | Free for business use as well since 8 July 2025; there is an Enterprise plan for SSO sign-in and model approvals. Plus hardware, electricity, maintenance and the licence of the respective model. |
What Ollama and LM Studio do better
If data must not leave your own computer, there is no way around a local model. That is what these two programs are for:
- Computing on your own hardware: the model request does not leave the computer as long as no cloud function is switched on.
- Working offline: LM Studio documents chat, documents and the local server without internet once the model is loaded; Ollama can be restricted to local models.
- Free choice of model: you decide which model runs in which version, and you keep it even if a provider changes something.
- No billing per task: in local operation you pay for hardware and electricity, not for usage.
Consider Ollama or LM Studio if you have to run the model yourself and someone in-house handles setup, updates and troubleshooting. Consider Biwak if you are looking for managed model access and an agent that works through tasks on files in the working folder.
Test case: a local model on your hardware
- Choose a model that fits the memory
In Ollama or LM Studio, load a model version that fits into the memory of the computer it will later run on, not the fastest one in the building.
- Disconnect and set a real task
After the download, cut the connection and set a typical task with an approved sample file. Note response time and quality.
- The same task in Biwak
With a connection, in the working folder, with the same acceptance criteria. Have the same person check both results.
- Count operation too
With a local model, setup, updates and troubleshooting are part of the effort; with Biwak, the question of whether the content may leave the computer.
What hardware a local model needs, and why a blanket gigabyte figure does not prove suitability, is explained in the article Running AI locally.
How we checked Ollama and LM Studio
Biwak is one of the providers compared; this page is not a neutral review. The information on Ollama and LM Studio comes from their own documentation: Ollama's FAQ, licence and pricing page, plus LM Studio's guide to offline operation, system requirements and terms of use, read on 23 and 24 September 2026. We have not measured hardware requirements or model quality, and we did not carry out the test case above ourselves: it describes how you can compare both modes of operation on your own task. We welcome corrections at kontakt@biwak.ai.
Frequently asked questions
Does Ollama see my inputs?
Not in local operation, according to Ollama. If you use cloud models or web search, Ollama processes the requests on its own servers, but according to its FAQ does not store or log them. If you want to rule that out, switch off the cloud functions with "disable_ollama_cloud": true in ~/.ollama/server.json or the environment variable OLLAMA_NO_CLOUD=1.
Can I add Ollama as a custom model endpoint in Biwak?
No. Customers do not set up their own model endpoint or API key; model requests run via the signed-in account and the managed model access. If you want to run a model yourself, use Ollama or LM Studio directly.
Is a local model automatically the best choice?
No, it depends on your requirements. Local computation answers the question of where data goes, but not whether the result is correct or whether connected tools send something themselves. Test model quality, hardware, maintenance and the complete chain on a real task.
Sources
- Ollama: FAQ, local operation and switching off cloud functions (undated, checked on 23 September 2026)
- LM Studio: Offline Operation (undated, checked on 23 September 2026)
- Ollama: licence (MIT) in the source code repository
- Ollama: Pricing (Free with local operation, Pro, Max, Team)
- LM Studio: System Requirements
- LM Studio Blog: LM Studio is free for use at work, 8 July 2025
- Biwak: Privacy policy, section 14 (requests to a language model)