feat: harbor pull smart routing, volumes CLI, and down resilience

- harbor pull delegates to run_models_pull for HF metadata-based
  routing (GGUF -> llamacpp, other -> hf download) instead of
  hardcoding llamacpp for all HuggingFace repos
- harbor down stops host services (MLX, DMR) before compose down
  so they are cleaned up even when Docker is unavailable
- Add harbor volumes CLI and help text
- Rewrite MLX CLI functions for mlx-lm, remove mlx-serve references
- Extract config search/list to configSearch routine
- Update README and install docs for DMR/MLX backends

Author: av

README.md

@@ -71,6 +71,10 @@ harbor up ollama
 harbor up llamacpp
 harbor up vllm
 
+# Host-native Metal/GPU inference on macOS
+harbor up dmr
+harbor up mlx
+
 # Set and remember args for llama.cpp
 harbor llamacpp args -ngl 32
 ```
@@ -97,11 +101,14 @@ Supported host tools include `claude`, `codex`, `copilot`, `droid`, `hermes`, `m
 
 ####  Cutting Edge Inference
 
-Harbor supports most of the major inference engines as well as a few of the lesser-known ones.
+Harbor supports most of the major inference engines as well as a few of the lesser-known ones. On macOS, [Docker Model Runner](./docs/2.2.22-Backend-Docker-Model-Runner.md) and [MLX](./docs/2.2.23-Backend-MLX.md) run inference on the host with Metal acceleration — no containers needed for the compute.
 
 ```bash
 # We sincerely hope you'll never try to run all of them at once
 harbor up vllm llamacpp tgi litellm tabbyapi aphrodite sglang ktransformers mistralrs airllm
+
+# Host-native inference (macOS Metal / Linux GPU)
+harbor up dmr mlx
 ```
 
 #### Tool Use
@@ -247,8 +254,8 @@ harbor eject searxng llamacpp > docker-compose.harbor.yml
 [KoboldCpp](https://github.com/av/harbor/wiki/2.2.16-Backend:-KoboldCpp) ⦁︎ [KTransformers](https://github.com/av/harbor/wiki/2.2.13-Backend:-KTransformers) ⦁︎ [Lemonade](https://github.com/av/harbor/wiki/2.2.19-Backend-Lemonade) ⦁︎ [llama.cpp](https://github.com/av/harbor/wiki/2.2.2-Backend:-llama.cpp)
 [lmdeploy](https://github.com/av/harbor/wiki/2.2.10-Backend:-lmdeploy) ⦁︎ [mistral.rs](https://github.com/av/harbor/wiki/2.2.6-Backend:-mistral.rs) ⦁︎ [Modular MAX](https://github.com/av/harbor/wiki/2.2.17-Backend-Modular-MAX) ⦁︎ [Needle](https://github.com/av/harbor/wiki/2.2.20-Backend-Needle)
 [Nexa SDK](https://github.com/av/harbor/wiki/2.2.15-Backend:-Nexa-SDK) ⦁︎ [Ollama](https://github.com/av/harbor/wiki/2.2.1-Backend:-Ollama) ⦁︎ [openedai-speech](https://github.com/av/harbor/wiki/2.2.7-Backend:-openedai-speech) ⦁︎ [Parler](https://github.com/av/harbor/wiki/2.2.8-Backend:-Parler)
-[SGLang](https://github.com/av/harbor/wiki/2.2.12-Backend:-SGLang) ⦁︎ [Speaches](https://github.com/av/harbor/wiki/2.2.14-Backend:-Speaches) ⦁︎ [TabbyAPI](https://github.com/av/harbor/wiki/2.2.4-Backend:-TabbyAPI) ⦁︎ [Text Generation Inference](https://github.com/av/harbor/wiki/2.2.9-Backend:-text-generation-inference)
-[vLLM](https://github.com/av/harbor/wiki/2.2.3-Backend:-vLLM)
+[Docker Model Runner](https://github.com/av/harbor/wiki/2.2.22-Backend-Docker-Model-Runner) ⦁︎ [MLX](https://github.com/av/harbor/wiki/2.2.23-Backend-MLX) ⦁︎ [SGLang](https://github.com/av/harbor/wiki/2.2.12-Backend:-SGLang) ⦁︎ [Speaches](https://github.com/av/harbor/wiki/2.2.14-Backend:-Speaches)
+[TabbyAPI](https://github.com/av/harbor/wiki/2.2.4-Backend:-TabbyAPI) ⦁︎ [Text Generation Inference](https://github.com/av/harbor/wiki/2.2.9-Backend:-text-generation-inference) ⦁︎ [vLLM](https://github.com/av/harbor/wiki/2.2.3-Backend:-vLLM)
 
 ##### Satellites
 [Activepieces](https://github.com/av/harbor/wiki/2.3.64-Satellite-Activepieces) ⦁︎ [Agent Zero](https://github.com/av/harbor/wiki/2.3.47-Satellite-Agent-Zero) ⦁︎ [aichat](https://github.com/av/harbor/wiki/2.3.14-Satellite:-aichat) ⦁︎ [Aider](https://github.com/av/harbor/wiki/2.3.13-Satellite:-aider)
@@ -296,6 +303,10 @@ harbor up speaches
 # Open Webui is automatically connected to them.
 harbor up llamacpp tgi litellm vllm tabbyapi aphrodite sglang ktransformers
 
+# Host-native backends for macOS (Metal acceleration, no containers)
+harbor up dmr
+harbor up mlx
+
 # Run different Frontends
 harbor up librechat chatui bionicgpt hollama
 
@@ -309,6 +320,11 @@ harbor up comfyui
 # Use custom models for supported backends
 harbor llamacpp model https://huggingface.co/user/repo/model.gguf
 
+# Manage models for host-native backends
+harbor dmr pull ai/mistral
+harbor dmr ls
+harbor mlx pull mlx-community/Qwen3.5-4B-4bit
+
 # Access service CLIs without installing them
 # Caches are shared between services where possible
 harbor hf scan-cache

docs/1.0.-Installing-Harbor.md

@@ -186,7 +186,7 @@ Follow the [Harbor CLI installation methods](#harbor-cli-installation-methods) b
 - Administrator access
 
 > [!NOTE]
-> **ARM64 Limitation:** Some services don't ship ARM64 images (Apple Silicon). GPU pass-through is not supported on macOS.
+> **ARM64 Limitation:** Some services don't ship ARM64 images (Apple Silicon). GPU pass-through is not supported on macOS. For Metal-accelerated inference, use host-native backends like [Docker Model Runner](./2.2.22-Backend-Docker-Model-Runner.md) (`harbor up dmr`) or [MLX](./2.2.23-Backend-MLX.md) (`harbor up mlx`) — they run on the host and are exposed to Harbor containers via a proxy.
 
 #### Step 1: Install Docker Desktop (10 minutes)