Ollama provides a local runtime for large language models, distributable as a Docker image (ollama/ollama) or via a shell install script (curl -fsSL https://ollama.com/install.sh \| sh). Models are managed through the CLI; for example, ollama run gemma4 starts the Gemma 4 model, while ollama launch claude or ollama launch openclaw invoke supported integrations such as Claude Code, Codex, Copilot CLI, DeepSeek Harness, Droid, OpenCode, and the OpenClaw assistant for messaging platforms. A REST API listens on http://localhost:11434/api/chat, accepting JSON payloads that specify the model, messages, and streaming flag, as shown in the sample curl request to gemma4. Language‑specific bindings are available via pip (Python), npm (Node.js), and analogous packages for .NET, Rust, Go, Java, Swift, Ruby, and others, enabling programmatic chat completions. The ecosystem includes numerous self‑hosted UIs (Open WebUI, Onyx, LibreChat, Lobe Chat, NextChat, Perplexica, big‑AGI, Lollms WebUI, ChatOllama, Bionic GPT, Hollama, Chatbox, etc.) and SDKs for vector databases (chromem‑go), observability (OpenLIT, Lunary, Langfuse, HoneyHive, MLflow Tracing), and agent frameworks. Underlying model execution relies on the llama.cpp project initiated by Georgi Gerganov, and the project invites contributions through pull requests.