run-llmslisted

Set up and run local LLMs with Harbor. Use when the user wants to run models locally, install Harbor, start Open WebUI, llama.cpp, Ollama, vLLM, Docker Model Runner, MLX, or oMLX, pull GGUF or HuggingFace models, add SearXNG web search, Speaches TTS/STT, or Open Terminal code execution, launch Codex/Claude/Grok/OpenCode against a Harbor backend, or troubleshoot GPU, VRAM, and service startup.
av/skills · ★ 13 · AI & Automation · score 80

Install: claude install-skill av/skills

# Run LLMs locally with Harbor Harbor is a containerized LLM toolkit. Default `harbor up` starts Open WebUI (`webui`) and llama.cpp (`llamacpp`). This skill is the operational playbook for installing Harbor, choosing a backend, pulling models, wiring search/voice/code tools, and pointing coding agents at the same stack. For the full CLI catalog, `harbor skills get harbor`. For Boost module authoring, use the `boost-modules` skill. ## Agent rules - `harbor logs` follows forever and will hang a non-interactive shell. Read logs with `docker logs harbor.<service>` (add `--tail 200` if needed). - Never edit `.env` by hand. Use `harbor config get/set` and `harbor env`. - Do not set `llamacpp.model` / `llamacpp.model.specifier` unless the user explicitly wants a single pinned GGUF. Router mode (empty specifier) discovers pulled GGUFs automatically. - `harbor launch` options (`--backend`, `--model`, `--web`, `--workflow`, `--config`, `--service`) go **before** the tool name. Everything after the tool name is the tool's own argv. - After `harbor update`, run `harbor config update` if new keys are missing from the local config. ## Read only what you need | File | When | |------|------| | `references/models.md` | Pull, list, remove, routing, HuggingFace token/cache | | `references/backends.md` | llamacpp, ollama, vLLM, DMR, MLX, oMLX, ikllamacpp | | `references/stack.md` | webui, searxng, speaches, openterminal, tunnels | | `references/launch.md` | Codex, Claude Code, Grok, OpenCo