Ollama

license

MIT

source

public

core-gated

ungated

MIT-licensed with public source and no feature-gated core. The whole engine ships free - platform installers for macOS, Windows and Linux, an official Docker Hub image, the local REST API on port 11434, and the model library - with no paid edition, license key or enterprise-only build. Ollama sells hosted cloud inference alongside the engine, but that is a separate service rather than a capability withheld from it, so nothing in the core is gated.

• https://raw.githubusercontent.com/ollama/ollama/main/LICENSE recorded 2026-08-13

  Verbatim MIT License body at repository head, "Copyright (c) Ollama", with the standard permission and warranty-disclaimer clauses and no added restriction

• https://api.github.com/repos/ollama/ollama recorded 2026-08-13

  Repository metadata - full_name ollama/ollama, language Go, license spdx_id MIT, public and not archived, 178,451 stars, 17,389 forks, last push 2026-08-12

• https://raw.githubusercontent.com/ollama/ollama/main/README.md recorded 2026-08-13

  The whole engine is distributed free from the repository and vendor site - install scripts and manual installers for macOS, Windows and Linux, the official ollama/ollama Docker Hub image, the local REST API on port 11434, agent integrations launched from the CLI, and ollama-python and ollama-js listed under Libraries as clients. No paid tier, license key or enterprise-only build is described anywhere in it.

• https://ollama.com/ recorded 2026-08-13

  Vendor site offering a free download plus paid hosted cloud models in the United States, Europe and Singapore; the cloud service is sold alongside the local engine, not as an unlock for it

The band rests on model-pull volume rather than on stars: ollama.com/library lists 233 models carrying about 1.0 billion cumulative pulls between them, led by llama3.1 at 118.4M and deepseek-r1 at 91.3M, alongside 178,451 GitHub stars - comfortably inside a 1M-10M active-user reading, and rated 4 conservatively. One figure does not hold up: the "52M+ monthly model pulls" this record’s reach wording carries appears on neither cited page and cannot be derived from them, because the library publishes an all-time cumulative pull counter per model rather than a monthly rate, and the homepage publishes no volume figure at all. No package is measured for this level, and that is deliberate. A PyPI package named `ollama` exists and is genuinely this project’s, but Ollama is distributed as a platform binary and a Docker image, and the PyPI package is a client SDK for talking to a local server rather than how Ollama is obtained - its 20.3M a month is a large number attached to the wrong question. Where a package is not the product’s primary distribution channel, banding on it is a substitution rather than a measurement. The level therefore rests on documented pull volume rather than on a feed that could recompute it, and the sources below are digested so the claim can be checked again and will age.

• https://api.github.com/repos/ollama/ollama recorded 2026-08-13

  Repository metadata - 178,451 stars, 17,389 forks, language Go, last push 2026-08-12, not archived

• https://ollama.com/library recorded 2026-08-13

  Model library listing 233 models with an all-time cumulative pull count each - llama3.1 118.4M, deepseek-r1 91.3M, nomic-embed-text 82.2M, llama3.2 80M, and so on, summing to roughly 1.0 billion pulls. The counter is cumulative, not a rate; no monthly figure is published anywhere on the page.

• https://raw.githubusercontent.com/ollama/ollama/main/README.md recorded 2026-08-13

  Distribution channels confirmed - platform install scripts and manual installers for macOS, Windows and Linux plus the official ollama/ollama Docker Hub image. ollama-python and ollama-js are listed under Libraries, which is what the PyPI package is - a client for a local server, not the channel the product is obtained through.

Local model serving with one-command run, an OpenAI-compatible REST API and a broad model library (Llama, Qwen, Gemma, gpt-oss, DeepSeek). It is a serving-layer tool with no MLPerf result to cite, so it is rated on breadth of features and coverage rather than on a benchmark. The surface is wider than a bare serving description suggests: the README documents the local REST API on port 11434 with an OpenAI-compatible chat endpoint, Python and JavaScript clients, and an agent-launcher layer - `ollama launch claude`, codex, copilot-cli, droid, opencode - plus the OpenClaw assistant integration, and the library carries 233 model families, several of which the vendor site also serves as cloud models. No benchmark result is published for the runtime itself, which is why the score rests on that feature coverage.

• https://raw.githubusercontent.com/ollama/ollama/main/README.md recorded 2026-08-13

  One-command run, local REST API on port 11434 with an OpenAI-compatible chat endpoint, Python and JavaScript clients, and CLI agent integrations for Claude Code, Codex, Copilot CLI, Droid, OpenCode and OpenClaw

• https://ollama.com/library recorded 2026-08-13

  233 model families listed, spanning Llama, Qwen, Gemma, DeepSeek, Mistral, Phi and gpt-oss, with tool, thinking, vision, embedding and cloud tags per entry