What Is Ollama? Is It Free, How to Install It, How Much RAM You Need, Does It Work in Turkish?
You install Ollama, download a model and run it on your own computer; on Windows and macOS you chat in the app window, on Linux and Docker from the command line. Models running locally are unlimited and free, and what you type stays on your computer. It is free and open source (MIT); for large models that don't fit on your computer there is a paid cloud option (Ollama Cloud). Speed and the model size you can use depend on your computer's memory; Ollama does not state Turkish answer quality, which depends on the model you choose.
- What it does
- Runs AI models on your own computer
- Replies in Turkish
- Depends on the model; not stated officially
- Turkish interface
- Not stated officially
- Where
- Windows, macOS, Linux, Docker
- Data
- Stays on your computer when local
- Free plan
- Software is free; cloud is paid
- Price
- Local $0 · Pro $20/mo
What is Ollama and what is it for?
Yes. On Windows and macOS, Ollama has its own app window: you install it, pick a model, type to chat and drag and drop files; you don't need to type commands. If your computer is weak or you don't want to download models, you can use Ollama Cloud in the same app (an account is required, limited on the free plan). The command line is only needed for Linux, Docker and advanced work.
Ollama is not an AI model but the program that runs models. First you install Ollama, then you pick a model from the model library and download it; the model runs on your computer. So it is not a site you sign into with an account like ChatGPT or Claude, but an "AI engine" on your own computer.
Who it helps: business owners who don't want to send what they write and their clients' documents over the internet, developers who write code and want to connect an AI model to their own software, and anyone who wants to learn and experiment with AI. Who it doesn't help: people without a powerful computer who also don't want to open a cloud account; speed and quality depend on your computer's memory and the model you choose (see the Hardware and Models sections).
Sources: Ollama README, the download page and the "Ollama's new app" blog post (July 30, 2025). The app's file drag-and-drop and chat features were announced for macOS and Windows; an app window for Linux is not described in the documentation.
How do you download and install Ollama? Windows, macOS, Linux, Docker
The official address is ollama.com/download. The commands below are taken verbatim from the official documentation.
- Install. Run the command below in PowerShell, or download and open OllamaSetup.exe from ollama.com/download. No administrator rights are needed; it installs to your user folder and needs at least 4 GB of free space.
- Requirements. Windows 10 22H2 or newer (Home or Pro). If you have an NVIDIA card, driver 551.61 or later (the hardware page says 550 or later; see Hardware). For AMD Radeon, the ROCm v7 or Vulkan driver.
- Use it. Once installed, Ollama runs in the background; the ollama command works in cmd or PowerShell. The API address is http://localhost:11434.
› irm https://ollama.com/install.ps1 | iexIf you want to keep models on another drive, set the OLLAMA_MODELS environment variable for your user account. If Ollama is running, quit it from the taskbar and open it again.
- Install. Open Ollama.dmg and drag the app into the Applications folder (the preferred route), or run the command below in Terminal.
- Requirements. macOS Sonoma (14) or newer. On Apple M-series chips both the CPU and the GPU are used; on Intel (x86) Macs, only the CPU.
- First launch. The app checks whether the ollama command is in your PATH; if not, it asks for permission to create a link in /usr/local/bin.
› curl -fsSL https://ollama.com/install.sh | sh› curl -fsSL https://ollama.com/install.sh | sh- Manual install: extract the package under /usr with the second command below. An AMD GPU needs an extra ROCm package, and ARM64 needs a separate package.
- Start the server with ollama serve; in another terminal, confirm it is running with ollama -v. To keep it running permanently, the official documentation describes a systemd service.
- To update, run the install script again.
› curl -fsSL https://ollama.com/download/ollama-linux-amd64.tar.zst | sudo tar x -C /usr› docker run -d -v ollama:/root/.ollama -p 11434:11434 --name ollama ollama/ollama› docker run -d --gpus=all -v ollama:/root/.ollama -p 11434:11434 --name ollama ollama/ollama› docker run -d --device /dev/kfd --device /dev/dri -v ollama:/root/.ollama -p 11434:11434 --name ollama ollama/ollama:rocm› docker exec -it ollama ollama run llama3.2!Docker Desktop on macOS has no GPU acceleration (official FAQ). Only the CPU is used, and large models become very slow.
No account is needed for local use. For cloud models you open an Ollama account (sign in in the app, or run ollama signin in the terminal). Account requirement: the Terms of Service (May 2026) require you to be at least 18; the Privacy Policy, however, mentions parental consent for children under 13. Because the two official texts conflict, 18 is taken as the requirement here. Whether phone verification is required is not stated in the official documentation.
Current version: v0.40.4 (October 12, 2026). On Windows and macOS, Ollama downloads updates by itself; you apply them with "Restart to update" in the menu. On Linux you run the install script again.
Sources: ollama.com/download, the docs.ollama.com Windows, macOS, Linux, Docker and FAQ pages, GitHub releases (October 14, 2026). If you don't want Ollama to start automatically when your computer boots, turn it off in Task Manager > Startup apps on Windows, or System Settings > Login Items on macOS.
How much RAM and which GPU do you need for Ollama?
The official documentation does not give a general "you need this much RAM" table; the yardstick is the model's file size. If the model fits in your graphics card's memory (VRAM), it runs fast. If it doesn't, Ollama also uses your computer's RAM, but answers slow down. On Macs, the CPU and GPU share the same memory (unified memory).
The small example model on the quickstart page downloads about 7.2 GB, and 8 GB of VRAM (unified memory on a Mac) is recommended. If VRAM is short, system RAM is used and answers slow down. The larger the context window, the more memory is needed.
The program itself needs at least 4 GB (Windows). Models can take tens or even hundreds of GB on top of that. By default models download to the .ollama\models folder in your user folder; OLLAMA_MODELS moves them to another drive.
4k context if GPU memory is under 24 GiB, 32k between 24 and 48 GiB, 256k at 48 GiB and above. For web search, agent and coding tools the official recommendation is at least 64000. Increasing context needs more memory. (The older FAQ page gives the default as 4096; the newer context-length page is used here.)
NVIDIA: compute capability 5.0 or higher, driver 550 or later (570 for 5.0–6.2). AMD Radeon: ROCm v7 (on Windows, RX 7600 through 7900 XTX and PRO W7500 and above); more cards through Vulkan. Apple: Metal. The full list of supported cards is at docs.ollama.com/gpu.
The PROCESSOR column of the ollama ps command: "100% GPU" means entirely on the graphics card, "100% CPU" entirely in RAM, "48%/52% CPU/GPU" split. If you see split or CPU, pick a smaller model.
The model is unloaded from memory 5 minutes after the last request; you can change this with OLLAMA_KEEP_ALIVE, or unload it immediately with ollama stop.
If your computer is weak there are two options: pick a smaller model or use a cloud model (Pricing section). In the cloud the model runs on Ollama's server, so your computer's hardware doesn't matter.
Sources: docs.ollama.com Quickstart, Hardware support, Context length, FAQ (October 14, 2026). The Windows install page says 551.61 for the NVIDIA driver, the hardware page says 550; check your own driver version.
How do you use Ollama?
- Open the Ollama app (Windows and macOS). In the setup questions you can sign in for a cloud model or pick a local model.
- Pick a model; on first use the model downloads to your computer (a few GB; see the Models section).
- Type in English or Turkish in the chat box. You can drag and drop a PDF or text file into the window; for long documents, enlarge the context slider in settings (it needs more memory).
- If you want to send an image, the model you pick must support image input (the model page says "Text, Image").
If you like the command line, the same thing works like this (official CLI documentation):
› ollama run gemma4:e2b› ollama ls
ollama rm gemma4:e2b
ollama stop gemma4:e2b› ollama psIf you always want to give a model the same instruction, you write a text file named Modelfile: the first line is the model to start from, the SYSTEM line is the fixed instruction. Then the ollama create -f Modelfile command creates a new local model under your own name (there is an example in the official CLI documentation).
!Local models also make mistakes and can make things up. Check numbers, dates and quotes against their source; quality can drop noticeably, especially with small models.
Where do you use Ollama: app, command line, cloud, API
Ollama is used in four ways. For most people the first two are enough.
All of them use the same local server: the app, the command line and other programs connect to the same Ollama through http://localhost:11434.
Using Claude Code, Codex and other tools with Ollama, and the API
Ollama can connect to coding agent tools as the backend model: you keep the tool and pick the model from Ollama. The "ollama launch" command installs and configures the tool itself (if it's not installed, it offers to install it).
› ollama launch claude› ollama launch codex
ollama launch opencode- Cost: Free with a local model. If you choose a cloud model you need credits or a paid plan. The 0.40.4 release notes say ollama launch can now be used with credits without moving to a paid plan.
- Context: For agent and coding tools the official recommendation is at least 64000 tokens of context (Hardware section).
- Web search: Ollama's web search requires an Ollama account and signing in; how many searches the free plan allows is not stated in the official documentation. See your usage at ollama.com/settings/usage.
Using it in your own software: The local server listens at http://localhost:11434, also understands OpenAI- and Anthropic-style endpoints, and needs no API key locally. For cloud models you create an API key at ollama.com/settings/keys and the request goes to https://ollama.com.
!Your API key is like a password: don't put it in your code, in the browser or in a file you share; the official documentation also says to keep the key on the server.
Which models does Ollama have? Differences and which to choose
Ollama has no model of its own: it runs the open models at ollama.com/library. The library has many models and new ones are added often; the table below shows a few well-known examples with the description, file size and context length from Ollama's own model pages. File size is a hint for how much you'll download and how much memory you need.
| Model | What it's for, per the model page | File size | Context |
|---|---|---|---|
| gemma4 (e2b, e4b) | Google DeepMind's model; small versions for edge devices and laptops; text and image input | e2b 4.6–7.5 GB · e4b 6.6–9.5 GB | 128K |
| gemma4 (12b, 26b, 31b) | Larger versions of the same family; reasoning, code and agent work; text and image input | 12b 7.7–8.0 GB · 26b 16–19 GB · 31b 19–20 GB | 256K |
| gpt-oss:20b | OpenAI's open-weight model; reasoning and agent work; per the page it can run on systems with 16 GB of memory | 14 GB | 128K |
| qwen3 (8b, 30b) | The latest generation of the Qwen series; the page lists support for 100+ languages and dialects | 8b 5.2 GB · 30b 19 GB | 40K · 256K |
| llama3.2 (1b, 3b) | Meta's small models; the official language list on the page does not include Turkish | 1b 1.3 GB · 3b 2.0 GB | 128K |
| mistral:7b | Mistral AI's 7B model (version 0.3) | 4.4 GB | 32K |
- On a weak computer or one with under 8 GB of memory: small models whose pages list 1–3 GB (llama3.2 1b/3b) or a cloud model
- Mid-range computer (8–16 GB of memory): small but current models like gemma4 e2b/e4b; per its page, gpt-oss:20b can run with 16 GB of memory
- Powerful GPU (24 GB and up): larger versions like gemma4 26b/31b
- If you'll feed it images, the model page must list "Text, Image" input
- If you need a large model that fits none of these: Ollama Cloud (Pricing section)
Ollama does not say which model is good in Turkish. Model pages only give their own language information (e.g. 8 official languages for Llama 3.2, Turkish not included; "100+ languages" for Qwen3). If Turkish matters, test the model you choose with your own text.
In the app you pick a model's cloud version by adding :cloud to the end of its name (e.g. gemma4:cloud). In the API you use the name in the ollama.com/api/tags list. Cloud models are not downloaded; retiring cloud models are announced in advance on the Usage page in settings, and downloaded local models are not affected.
The small models on the DeepSeek-R1 page are not R1 itself but separate models derived from other companies' models (Qwen- and Llama-based). They are not the same as DeepSeek in its app; this is covered separately in the DeepSeek guide.
Models in the library are updated often; this section was checked on October 14, 2026. See the current list at ollama.com/library.
Limits of running DeepSeek models in Ollama, and DeepSeek's own app
Is Ollama free? How much are Ollama Cloud and Pro?
The Ollama software is free and open source (MIT license). The official pricing page: "Running models on your own hardware is always unlimited", meaning models you run on your computer have no limit and no cost. The only paid part is the optional Ollama Cloud: running large models on Ollama's servers.
- Unlimited local models
- Starter credits, starter models
- Credits can be bought for all cloud models
- Concurrent requests: 1
- Access to larger "pro" models (Pro and up)
- Several models at once (Pro and up)
- $60 of usage credit per month
- Access to larger pro models
- Run several models at once
- Concurrent requests: 3
- $300 of usage credit per month
- Early access to the newest models
- Concurrent requests: 10
- Unlimited users
- $1,000 of shared credit per month
- Central billing and management
- Priority support
View as a table (plans)
| Plan | Monthly price | Included credit | Concurrent requests |
|---|---|---|---|
| Free | $0 | Starter credit (amount not stated) | 1 |
| Pro | $20 ($200 yearly) | $60 | 3 |
| Max | $100 | $300 | 10 |
| Team (early access) | $500 | $1,000 (shared) | 10 |
| Enterprise | Custom | Custom | Custom |
- Usage is metered per token at each model's own price; every model's dollar price per 1 million tokens is in the table on ollama.com/pricing. Some models have a lower price outside 12:00–18:00 UTC on weekdays and on weekends.
- Credit included in a plan resets on the same day each month and does not roll over to the next month. Extra credit you buy expires 1 year after it is added (Terms of Service, section 6).
- On paid plans you get an email warning when you reach 90% of your included usage; you can see your usage at ollama.com/settings/usage.
- Every plan, Free included, can buy extra credit; the plan's included credit is spent first, then the balance you bought.
Prices are in dollars; there is no official Turkey page or local-currency price. Payments are processed by Stripe and you are responsible for taxes (Terms of Service, section 6); VAT (KDV) details for Turkey are not stated in the official documentation, so go by the amount shown at checkout. The subscription renews automatically.
Prices checked on October 14, 2026 (ollama.com/pricing, ollama.com/terms). Ollama may change prices and plan contents; the Ollama blog (August 31, 2026) announced that the Pro, Max and Team plans moved to per-token pricing.
Official pricing page last checked on .
Does Ollama work in Turkish? Is there a Turkish model?
| Replies in TurkishUnclear | Ollama doesn't state this itself: answer quality depends on the model you choose. Ollama's model pages show no support information that names Turkish; the Llama 3.2 page lists 8 official languages (no Turkish), and the Qwen3 page says "100+ languages and dialects" without listing them. Test with your own text. |
|---|---|
| App interfaceUnclear | The desktop app's interface language is not stated in the official documentation. |
| Dictation and spoken repliesUnclear | The official documentation does not describe voice input/output in the Ollama app or command line. Some small models' pages mention audio input support, but how to use it in the Ollama interface is not stated. |
| Turkish documentation and supportUnclear | The documentation (docs.ollama.com) and the pricing page are in English. The official sources show no Turkish documentation or Turkish support channel. |
If the answer comes back in the wrong language, add "Reply in Turkish" at the start of your request. Turkish can be noticeably weaker in small models; if you run into problems, try a larger model or a cloud model.
Reported by users Users looking for a local Turkish model mostly write on forums that "Turkish is weak in small models"; the official sources say nothing on this.
Is Ollama safe? Where does my data stay, and what happens in the cloud?
Privacy Policy (March 2026): "Ollama runs locally. We don't see your prompts or data when you run locally." So with a local model, what you type stays on your computer and does not go to Ollama. You need the internet to download model files.
Prompts and responses are processed temporarily to provide the service, are not stored or logged, and are not used for model training (Ollama FAQ, Cloud page, pricing page). Basic account information and limited usage data that carries no content (e.g. number of requests, app version) are collected; data is not sold. Models are hosted mainly in the US, with possible routing to Europe for capacity; data may be transferred to and processed in the US.
If you only want to work locally, cloud features can be turned off: write disable_ollama_cloud: true in the ~/.ollama/server.json file or set the OLLAMA_NO_CLOUD=1 environment variable, then restart Ollama. Cloud models and web search will not work.
By default the server listens only on 127.0.0.1:11434, so no one else can connect except from the same computer. If you open it to the network with OLLAMA_HOST, anyone on the network can connect; if you do this, use a firewall or a proxy server. For browser access the default allowed origins are 127.0.0.1 and 0.0.0.0; additional origins are added with OLLAMA_ORIGINS.
The GitHub security advisory database has several entries affecting past Ollama versions (example: a memory-read vulnerability in the model file loader in May 2026, fixed in 0.17.1). On Windows and macOS updates download automatically; on Linux, run the install script again. This is not a "safe" seal.
Ollama's security and the behavior of the model you download are separate: check who published the model file. Models in the Ollama library are published by their makers or by the community.
The Privacy Policy says you can delete your account at any time; the official sources don't show the deletion steps, and the support address hello@ollama.com appears in the Terms of Service. Account information is kept as long as the account stays open.
!This page is not a security certification. If you'll work with customer or patient data, run models locally and turn off cloud features; for personal data law (such as KVKK, Turkey's Personal Data Protection Law), consult a qualified professional. This is not legal advice.
Are Ollama and the models I run fit for commercial use?
There are two separate licenses: the program's (Ollama) and the model's. The program's license is MIT. Each model has its own separate license, stated on the model page or by the maker.
- Using the Ollama program in commercial work, modifying it and including it in your own product (MIT; provided you keep the license text and copyright notice)
- Inputs and outputs you produce on the cloud service remain yours; Ollama does not use them for model training (Terms of Service)
- Using open models whose license allows it in a commercial product (for example, Apache 2.0-licensed ones)
- Assuming you can use every model commercially: some models (e.g. Llama-based ones) are bound by their own license terms
- Using Ollama's name and brand without permission (Terms of Service)
- Using the service for illegal or harmful work
- Examples from Ollama's model pages: Apache 2.0 for gpt-oss, the Apache license for mistral 7B; the DeepSeek-R1 page lists Apache 2.0 for the Qwen-based small models and Llama's own licenses for the Llama-based ones.
- No license text appears on the Gemma 4, Qwen3 and Llama 3.2 pages: read the maker's license before commercial use.
- You must check the accuracy of content you'll publish yourself; the Terms of Service require independent verification for critical decisions.
Sources: GitHub (MIT), Ollama Terms of Service (May 2026), model pages (October 14, 2026). This is not legal advice.
Common Ollama problems and fixes
| What happens | What to do |
|---|---|
| The model is very slow | Run ollama ps. If the PROCESSOR column shows "100% CPU" or a split (e.g. 48%/52% CPU/GPU), the model didn't fit in GPU memory. Pick a smaller model or lower the context length. |
| The answer gets cut off, agent work breaks | Increase the context length: the slider in the app settings, or OLLAMA_CONTEXT_LENGTH=64000 ollama serve. It needs more memory. |
| Disk is full; where are the models? | On Windows C:\Users\<name>\.ollama\models, on macOS ~/.ollama/models. To change the location, the OLLAMA_MODELS environment variable; to free up space, ollama rm <model>. |
| Ollama starts automatically when the computer boots | On Windows: Task Manager > Startup apps > ollama > Disable. On macOS: System Settings > Login Items. |
| The cloud model doesn't work | Sign in with ollama signin. Some models require a paid plan or credits (Pricing section). In the app/command, the cloud name ends with :cloud. |
| The model won't download; I'm behind a proxy | Set the HTTPS_PROXY environment variable; don't set HTTP_PROXY, as it can break the connection. |
| An AMD Radeon RX 6000 card isn't used on Windows | With current Windows AMD drivers, ROCm v7 may not see some RX 6000 cards; Vulkan is on by default and is the recommended fix. With mixed integrated/discrete graphics, select the discrete card with GGML_VK_VISIBLE_DEVICES. |
| Where do I find the logs? | On Windows, %LOCALAPPDATA%\Ollama via Win+R; on macOS ~/.ollama/logs; on Linux journalctl -u ollama. |
Sources: Ollama FAQ, Troubleshooting, Hardware support, Windows and Context length pages (October 14, 2026). Reported by users Some users complain about slow downloads that drop midway; the official sources give no explanation for this.
How to uninstall Ollama, delete a model and cancel a subscription
› ollama ls
ollama rm gemma4:e2bThe deleted model file is removed from disk. The same model can be downloaded again later.
- Uninstall Ollama from the list in Windows Settings > Apps (Add or remove programs).
- If you moved models to another folder with OLLAMA_MODELS, the uninstaller doesn't delete them; delete the folder yourself.
The official documentation gives a manual removal list: delete Ollama.app, the /usr/local/bin/ollama link, the Ollama folders under ~/Library and the ~/.ollama folder (the models are here). The commands are at docs.ollama.com/macos.
The official documentation says, in order: stop and disable the systemd service, delete the service file, delete the ollama libraries and executable, remove the ollama user/group and the /usr/share/ollama folder (the models are here). The commands are at docs.ollama.com/linux.
Pro, Max and Team subscriptions renew automatically. The Terms of Service (section 6) say to cancel in your account settings or by getting in touch; refund terms are not stated in the official sources. Extra credit you buy expires 1 year after it is added. You have the right to delete your account, but the steps are not stated in the official sources; you can write to Ollama at hello@ollama.com.
Ollama or LM Studio? Ollama or ChatGPT?
Reported by users On Hacker News, users say Ollama makes adding models with "ollama pull" and switching models without restarting very easy; some report moving to llama.cpp or LM Studio for more up-to-date model support. These are user opinions. The rows for other tools summarize those tools' own guides.
Frequently asked questions
What is Ollama and what is it for?
Ollama is a free, open-source program that runs AI language models on your own computer. You download a model from the library and chat in the app or the command line; what you type stays on your computer. If you want, you can also run large models in Ollama's cloud.
How do you download and install Ollama?
From ollama.com/download: OllamaSetup.exe or a PowerShell command for Windows, Ollama.dmg or an install command for macOS, and an install script for Linux. There is also a Docker image. It requires Windows 10 22H2 or macOS Sonoma (14) or newer.
Does Ollama work in Turkish? Is there a Turkish model?
Ollama does not state Turkish answer quality or a Turkish interface; quality depends on the model you run. Model pages give their own language information and do not name Turkish separately. Test with your own text.
Is Ollama free?
Yes, the program is free and open source (MIT); models running on your computer have no limit and no cost. Optional cloud: starter credit on the Free plan, Pro $20 a month, Max $100 a month, Team $500 a month.
Which models does Ollama run, and which should I choose?
It runs the open models at ollama.com/library. If your computer is weak, choose small 1–3 GB models or a cloud model; if you have a powerful GPU, choose larger models. Details and example sizes are in the Models section.
How much RAM and which GPU do you need for Ollama?
The official documentation gives no general RAM table; the yardstick is the model's file size. The small example model in the quickstart downloads about 7.2 GB, and 8 GB of VRAM is recommended. If VRAM isn't enough, RAM is used but it slows down. Recommended: driver 550 or later for NVIDIA, ROCm v7 or Vulkan for AMD, Apple M series for Mac.
Ollama or LM Studio?
Both are programs that run models on your computer. Ollama comes as a command line and a desktop app, free and open source (MIT). LM Studio's details are in its own guide.
What is Ollama Cloud, and is it free?
Running large models on Ollama's servers without downloading them. The Free plan has starter credit and starter models; all models require credit. Pro is $20 a month and includes $60 of credit, Max is $100 a month with $300 of credit.
What is the Ollama API key, and does it work with Claude Code?
Local use needs no API key; the server runs at localhost:11434. For cloud models you create a key at ollama.com/settings/keys. Tools like Claude Code, Codex and OpenCode can be started with an Ollama model using the ollama launch command.
Is Ollama safe, and where does my data go?
When running locally, Ollama does not see your prompts. With a cloud model, prompts are processed temporarily, not logged and not used for training; data may be transferred to and processed in the US. By default the server listens only on 127.0.0.1. Use the latest version.
How do you uninstall Ollama and delete a model?
To delete a model, ollama rm model-name. On Windows you uninstall the program from Settings > Apps; for macOS and Linux the official documentation gives step-by-step removal commands. By default models are in the .ollama folder in your user folder.
Why is Ollama running so slowly?
The model probably didn't fit in GPU memory. Check the PROCESSOR column in the ollama ps command: if you see 100% CPU or a split, pick a smaller model or lower the context length. A cloud model is also an option.
Tasks and setup kits that use Ollama
Related guides
Sources
- Ollama download page
- Ollama pricing
- Ollama Terms of Service (May 2026)
- Ollama Privacy Policy (March 2026)
- Ollama's new app (blog, July 30, 2025)
- Ollama documentation: Quickstart
- Ollama documentation: Windows
- Ollama documentation: macOS
- Ollama documentation: Linux
- Ollama documentation: Docker
- Ollama documentation: Cloud
- Ollama documentation: Hardware support
- Ollama documentation: Context length
- Ollama documentation: Command line (CLI)
- Ollama documentation: FAQ
- Ollama documentation: Troubleshooting
- Ollama documentation: Claude Desktop
- GitHub: ollama/ollama (README, releases, MIT license)
- GitHub security advisory database: Ollama
- Model page: gemma4
- Model page: gpt-oss
- Model page: qwen3
- Model page: llama3.2
- Model page: deepseek-r1
- Model page: mistral
Not clear in the official sources: The app's Turkish interface and Turkish answer quality · the web search quota on the free plan · subscription refund terms and account deletion steps · VAT (KDV) and local-currency price for Turkey · license texts for the Gemma 4, Qwen3 and Llama 3.2 models · an app window for Linux · phone verification when opening an account · the age conflict between the Terms of Service (18) and the Privacy Policy (13). Each is covered in its section, with what you can do about it.