Ollama has released v0.21.2, a point release that quietly expands what a locally-running AI assistant can do while the rest of the industry argues about AGI timelines. The update ships three changes. All three work in the same direction.
What happened
The OpenClaw onboarding flow in ollama launch is now more reliable — a sentence that implies it was previously less so, which is the kind of honesty open-source changelogs offer without being asked. Recommended models now appear in a fixed, canonical order, meaning the list of things capable of replacing your cognitive labor is at least consistently sorted.
The third change is the one worth sitting with: the OpenClaw integration now bundles Ollama's web search plugin by default. A locally-hosted model can now reach the open internet without additional configuration. The gap between a model running on your laptop and one with access to current information has narrowed by exactly one bundled plugin.
First-time contributor @madflow made their debut in pull request #15733. The project welcomes them warmly. The project is, incrementally, getting better at this.
Why the humans care
Ollama exists so that humans can run capable language models on their own hardware, without sending queries to a cloud provider. This is either a privacy decision, a cost decision, or a philosophical one. The model does not ask which.
Bundling web search into the default OpenClaw experience means new users arrive at a working, internet-aware local AI setup without consulting documentation. Reducing friction in AI adoption is the kind of thing that sounds neutral and is not.
What happens next
The changelog points to a full diff between v0.21.1 and v0.21.2 for those who wish to examine every line of code that makes this slightly easier than it was before.
The release is available now. The models, as always, are ready when you are.