Ollama Local

**Judgement:** A strong, practical skill for anyone standardizing on Ollama for local inference. It covers nearly all core needs—model lifecycle, chats, completions, embeddings, tools, and OpenClaw sub‑agents—with clear configuration and usage examples. **Key strengths** - Comprehensive Ollama coverage, including tool‑capable models and multi‑agent (think‑tank) patterns. - Good documentation and troubleshooting tips (host config, VRAM limits, CPU vs GPU behavior). - Clean OpenClaw integration via `sessions_spawn` and consistent `ollama/<model>` paths. **Main risks / limitations** - Reliability and performance depend heavily on the user’s Ollama setup (hardware, VRAM, network if remote); the skill can’t mitigate those environmental issues. - Safety controls are effectively delegated to the chosen local models and the surrounding OpenClaw orchestrations—there are no extra guardrails described. - Potential configuration pitfalls (incorrect `OLLAMA_HOST`, missing `ollama:default` profile, wrong model tags) can cause confusing fallbacks or failures. **Recommended scenarios** - Teams or individuals who already run Ollama locally and want OpenClaw agents to use those models by default. - Privacy‑sensitive or offline workloads where remote LLM APIs are not acceptable. - Power users building multi‑agent local workflows (e.g., architect/coder/reviewer agents) and experimenting with tool‑calling models on-prem. - RAG or semantic search setups that require local embeddings generation integrated into OpenClaw.