ollama mit Homebrew, apk, dnf, MacPorts, Nix, pacman, zypper, chocolatey, scoop, winget installieren

Projektlinks

• Homepage
• Repository
• Upstream-Dokumentation
• Paketmanager-Seite

Befehle und Aliase

• ollama

Projektgeschichte

The ollama/ollama repository was created on 2023-06-26, during the fast expansion of open-weight LLM releases after LLaMA-family models made local inference a mainstream developer activity. Written primarily in Go and licensed under MIT, the project made the single command pattern, such as ollama run model-name, central to its identity.

Ollama expanded from a terminal-first model runner into a broader local-AI substrate. Official docs expose a localhost API at port 11434, document generation, chat, embeddings, model pulling, pushing, and listing, and point developers to official Python and JavaScript libraries.

Adoptionsgeschichte

Ollama's adoption grew with developers' desire to run models without sending prompts, code, or data to hosted APIs. Its model library made open models discoverable by name and tag, while the CLI hid much of the friction around quantized model files, runtime setup, and GPU/CPU execution details.

By 2026-07-01, GitHub repository metadata reported 175,249 stars and 16,789 forks, making Ollama one of the most visible packages in the local-AI tooling wave. The official model library displays pull counts for individual models, and external developer docs such as GitLab's local-model guidance use Ollama as a practical way to serve supported LLMs for development.

Wie es verwendet wird

Developers use Ollama to pull and run models such as Llama, Gemma, Qwen, Mistral, DeepSeek, and embedding models; expose them to apps through the REST API; prototype agents and chat tools; and keep code or private documents on local hardware. The API is simple enough to call with curl and broad enough to support libraries, web UIs, IDE integrations, and agent tools.

The package also matters for model packaging. An Ollama model is not only weights: it can include a name, tag, quantization choice, prompt template, parameters, and Modelfile instructions. That makes the package feel closer to Docker-style model distribution than to a bare inference binary.

Warum Paket-Nerds sich dafür interessieren

Ollama is significant because it moved local LLMs from specialist inference stacks into normal package-manager territory. Homebrew, Linux package managers, Windows package managers, Docker, language clients, and an HTTP API all point at the same basic workflow: install one executable, pull a named model, and serve or run it.

Zeitleiste

• 2023-06-26: The ollama/ollama repository was created on GitHub.
• 2024-01-23: Ollama announced initial Python and JavaScript libraries.
• 2024-09-25: Meta announced Llama 3.2 models, including vision models and small edge-oriented text models.
• 2024-11: Ollama documented Llama 3.2 Vision support in Ollama 0.4, including 11B and 90B variants.
• 2026-07-01: GitHub repository metadata reported 175,249 stars, 16,789 forks, Go as the primary language, and MIT licensing.

Related projects

• Related projects and ecosystems include llama.cpp, Docker, Open WebUI, LangChain-style application frameworks, Hugging Face model distribution, Ollama's Python and JavaScript libraries, and open model families from Meta, Google, Alibaba/Qwen, Mistral, DeepSeek, and others.

Quellen

• https://ai.meta.com/blog/llama-3-2-connect-2024-vision-edge-mobile-devices/
• https://api.github.com/repos/ollama/ollama
• https://docs.gitlab.com/development/ai_features/local_models/
• https://docs.ollama.com/api/introduction
• https://formulae.brew.sh/formula/ollama
• https://github.com/ollama/ollama
• https://github.com/ollama/ollama/blob/main/docs/api.md
• https://ollama.com/
• https://ollama.com/blog/llama3.2-vision
• https://ollama.com/blog/python-javascript-libraries
• https://ollama.com/library