Separate vendor installation instructions from a dated runtime verification.
You will install Ollama, verify the CLI, choose and run a model, understand storage, update safely, remove models, and collect troubleshooting evidence.
Before you begin
Know your operating system, confirm free storage, back up important work, and use only official Ollama downloads and documentation. Begin with a modest model.
Before installation
Ollama is a local model runtime. Confirm current operating-system support at the official Ollama website, ensure you have free storage, and use an administrator account only when the installer requests it. Model requirements vary; begin with a small model rather than assuming a large one will fit.
macOS installation
Official documentation currently lists macOS Sonoma 14 or newer, with Apple M-series CPU and GPU support or x86 CPU-only operation. Download the disk image from ollama.com/download, mount it, and drag Ollama to Applications. On first start, the app can request permission to expose the CLI in Terminal. Requirements change, so recheck the official macOS page instead of treating this as a permanent promise.
Choose, download, and run a model
Browse Ollama's official model library and inspect description, size, license, and variants. Choose a modest instruct model suited to your hardware. The official command pattern is:
ollama pull MODEL_NAME
ollama run MODEL_NAMEpull downloads the selected model. run starts a session and may download it if needed. Replace the placeholder with the exact library identifier, test non-sensitive text, and type /bye to leave.
Verify, inspect, and remove
Use documented commands:
ollama --version
ollama ls
ollama ps
ollama rm MODEL_NAMEThe version confirms which CLI is found. ls lists downloaded models, ps lists loaded models, and rm removes the named model through Ollama. Read the name twice before removal. Application files and models are separate, so uninstalling an application may not remove a custom model location.
Storage and updates
Models can consume tens of gigabytes. Official Windows documentation describes the user-level OLLAMA_MODELS variable for changing storage; macOS uses ~/.ollama by default. Do not move internal files while Ollama runs unless current instructions support it. The Windows installer handles regular app updates. On Linux, the official page says rerunning the official install script updates Ollama. Note the working version and recheck platform instructions first.
Troubleshoot the first run
If the command is missing, reopen the terminal and verify installation and PATH guidance. If the service is unavailable, open the desktop app; on Linux, check the configured service or use ollama serve as documented. For download failures, check storage, connection, and model name. For memory errors or system instability, stop and choose a smaller model or quantization. Preserve the exact error, OS, version, model identifier, and hardware.
Privacy and official sources
Local models need no API key for local requests, but Ollama also offers cloud features and integrations. Confirm whether the selected model is local and whether an integration transmits content. Protect the local API from untrusted network exposure, update the app, and keep secrets out of prompts and logs. These instructions were checked on 2026-10-21 against Quickstart, Windows, Linux, macOS, and CLI documentation.
Windows and Linux installation choices
On Windows, the official installer runs Ollama in the background and makes the CLI available in Command Prompt or PowerShell; current requirements are listed on the official Windows page. On Linux, the official page provides an installation script and manual packages for specific architectures, plus optional service setup. Review instructions for your distribution and hardware before running them. This guide does not reproduce destructive uninstall or driver commands because their correct targets depend on the actual installation and should be taken from current official documentation.
Expected result and safe next step
A successful setup has a working version command, one listed model, an interactive prompt that answers a simple non-sensitive question, and a computer that remains stable. Record the Ollama version, exact model identifier, approximate storage used, and whether processing is local. Then compare three prompts with answers you know, including one where information is missing. Do not expose the local API to other networks, add cloud integrations, or process sensitive documents until you understand access, data flow, and authentication.
When Ollama is the right installation target
Decision: Use Ollama for its documented CLI/API path; prefer a desktop runtime when visual model management is the primary requirement.
Relevant options: Ollama documents Windows, macOS, Linux, CLI, and API use; LM Studio and Jan are desktop alternatives. Do not install all three for an initial fit test.
Worked example: Run version and list commands, load one modest documented model, and ask a non-sensitive question you can verify.
Failure mode: Path errors, storage limits, unsupported acceleration, or an exposed local API can undermine the install.
Validation: Confirm source, version, model, process binding, storage, and supported removal path.
Useful next step: Keep the first model small and record the exact error before changing drivers or firewall rules.
Official sources checked 2026-10-24: Ollama Windows · Ollama Linux · Ollama macOS · Ollama CLI. Product availability, support, terms, pricing, and licenses can change. These are vendor or project sources; no independent benchmark is claimed.
Verified environment record — 2026-10-27
Environment: Ubuntu 26.04.1 LTS, Linux 7.0.0-38-generic, x86_64; Intel i5-7300HQ; 7.1 GiB RAM; GeForce GTX 1050 Mobile, 2,048 MiB VRAM, driver 580.178.04.
Observed: ollama --version returned 0.35.1; ollama ls listed qwen2.5:1.5b and qwen3:4b; the local API reported 0.35.1 and completed a qwen2.5:1.5b request.
NOT_TESTED: fresh installation, update, download, removal, Windows, and macOS. See the dated test record.
Authoritative references
Reference scope checked 2026-10-27. These sources support the concepts and risk controls; product-specific behavior still requires the nearby vendor documentation and a dated test.
Your next step
Record a fresh platform-specific install yourself; do not treat this existing-install verification as an installation test.
Editorial provenance
Nyfir Studios is the publisher and editorially responsible organization. AI tools assisted with drafting and structured review for this candidate; factual claims, test labels, links, and retained evidence require human review before deployment. No named employee, credential, or unrecorded hands-on result is implied.