Ollama releases 0.18.1 with OpenClaw web search plugin and headless launch mode

• Ollama's 0.18.1 announcement adds two OpenClaw plugins: web search for current results and web fetch for extracting readable page content, with the explicit limit that fetch "does not execute JavaScript." release details
• The same 0.18.1 announcement adds a non-interactive ollama launch flow, positioning it for container jobs, CI pipelines, and scripted eval or review tasks instead of an interactive setup wizard. headless mode
• Ollama's Nemotron post also made NVIDIA's Nemotron 3 Nano 4B available through ollama run, and pitched it with Pi as "a great fit" for agents on constrained hardware. model availability
• The linked supporting docs thread fills in the implementation context: Pi is Ollama's minimal agent runtime with plugin support, while the model page describes Nemotron 3 Nano 4B as a small agent-oriented member of NVIDIA's Nemotron family. Pi docs

Ollama 0.18.1's main engineering change is that OpenClaw now gets first-party web search and web fetch plugins. In Ollama's release post, the company says models running through Ollama, whether local or cloud, can "search the web for the latest content and news," and OpenClaw can fetch a page and "extract readable content for processing." The important implementation boundary is also in that same post: the fetch path does not run JavaScript, so this is retrieval for static or server-rendered content, not full browser automation.

The release also adds a headless path for ollama launch. Ollama's headless mode frames it around unattended runs in Docker, CI/CD, and scripts, including evals, prompt validation, code review, and security-check jobs. The example command passes --yes to skip interactive confirmation and then forwards a prompt directly into the launched integration, which is the practical change for pipeline use.

Alongside the release, Ollama added nemotron-3-nano:4b as a runnable model and explicitly paired it with Pi via ollama launch pi --model nemotron-3-nano:4b. In Ollama's launch example, Pi is described as "the minimal agent runtime that powers OpenClaw," which ties the new headless launch flow to a smaller-footprint agent stack.

The Pi documentation describes manual and quick-start configuration around providers, models, and plugin support, which makes it the lightweight orchestration layer here. The linked model page adds why Ollama is emphasizing this pairing: Nemotron 3 Nano 4B is positioned as an efficient, agent-focused model variant, with the release post calling it suitable for "constrained hardware." Together, that gives Ollama a clearer unattended path: lightweight runtime, small model, and a launch command that can now run without a human in the loop.