feat: auto-bootstrap missing DMR and MLX host components on harbor up

- dmr_install_components: auto-installs docker-model-plugin via apt/dnf
  on Linux, or enables Model Runner via Docker Desktop on macOS
- dmr_install_runner: installs the DMR runtime when available
- harbor up dmr now bootstraps gracefully instead of erroring out
- harbor up mlx auto-installs mlx-serve via uv on macOS when missing
- Updated all three backend docs to describe the automatic setup

Author: av

.facts

@@ -83,6 +83,9 @@
 - label: cli > routines > routines/utils.ts has explicit TypeScript annotations for shared utility exports so deno check routines/models.ts is not blocked by implicit-any diagnostics
   command: deno check routines/models.ts
   tags: [spec, routines-utils, implemented]
+- label: services > dmr/mlx > harbor up dmr and harbor up mlx automatically bootstrap missing host-managed backend components before starting their Harbor proxy services
+  command: rg -q 'dmr_install_components' harbor.sh && rg -q 'mlx_install_components' harbor.sh && rg -q 'automatic.*install|automatically install|auto-installs' docs/2.2.22-Backend-Docker-Model-Runner.md docs/2.2.23-Backend-MLX.md
+  tags: [spec, auto-host-bootstrap, implemented]
 # project
 - Harbor is a containerized LLM toolkit distributed as a Docker Compose project with a CLI and Tauri desktop app
 - label: the project is licensed under Apache 2.0

docs/2.2.22-Backend-Docker-Model-Runner.md

@@ -13,7 +13,7 @@ DMR is especially useful on Apple Silicon because Metal inference runs on the ma
 harbor up dmr
 ```
 
-When `HARBOR_DMR_MANAGE_HOST=true`, Harbor checks for the Docker Model Runner CLI, enables the TCP endpoint when configured, and pulls `HARBOR_DMR_MODEL` before starting the proxy.
+When `HARBOR_DMR_MANAGE_HOST=true`, Harbor automatically installs or enables missing Docker Model Runner host components before starting the proxy. On Linux this uses the official `docker-model-plugin` package through `apt` or `dnf` when the `docker model` CLI is missing; on Docker Desktop systems Harbor uses Docker Desktop's Model Runner enablement command when it is available. Harbor then installs the runner runtime when supported, enables the TCP endpoint when configured, and pulls `HARBOR_DMR_MODEL`.
 
 Start a frontend against DMR:
 
@@ -96,7 +96,7 @@ harbor logs dmr
 
 ##### Docker Model Runner CLI is missing
 
-Install or update Docker Desktop and verify:
+`harbor up dmr` attempts automatic setup when host management is enabled. If setup cannot proceed on your platform, install or update Docker Desktop, or install Docker's `docker-model-plugin` package on Linux, then verify:
 
 ```bash
 docker model --help

docs/2.2.23-Backend-MLX.md

@@ -9,19 +9,13 @@ This service is intended for Apple Silicon systems. On other platforms the proxy
 
 #### Starting
 
-Install `mlx-serve` on the host first:
-
-```bash
-uv tool install mlx-serve
-```
-
-Then start the Harbor service:
+Start the Harbor service:
 
 ```bash
 harbor up mlx
 ```
 
-When `HARBOR_MLX_MANAGE_HOST=true`, Harbor renders `services/mlx/models.yaml` from the template, starts `mlx-serve` from `services/mlx/`, and starts the Harbor proxy.
+When `HARBOR_MLX_MANAGE_HOST=true`, Harbor automatically installs missing MLX host components before starting the proxy. If `mlx-serve` is missing on macOS, Harbor installs it with `uv tool install mlx-serve`; if `uv` is missing and Homebrew is available, Harbor installs `uv` first. Harbor then renders `services/mlx/models.yaml` from the template, starts `mlx-serve` from `services/mlx/`, and starts the Harbor proxy.
 
 Start a frontend against MLX:
 
@@ -119,7 +113,7 @@ harbor logs mlx
 
 ##### mlx-serve is not installed
 
-Install it on the macOS host:
+`harbor up mlx` automatically installs `mlx-serve` on macOS when host management is enabled. If automatic setup cannot proceed, install `uv` and `mlx-serve` manually:
 
 ```bash
 uv tool install mlx-serve

docs/8.5-OpenAI-Compatible-Local-LLM-Backends.md

@@ -75,6 +75,8 @@ harbor up dmr
 
 On Apple Silicon, DMR is the preferred Docker-managed path for host-native Metal inference. Harbor starts a proxy container at `http://dmr:8080/v1`; Docker Model Runner itself stays on the host.
 
+When host management is enabled, `harbor up dmr` also attempts to bootstrap missing Docker Model Runner components before the proxy starts.
+
 ### MLX
 
 [MLX](./2.2.23-Backend-MLX.md) is useful when you specifically want Apple's MLX runtime on an Apple Silicon Mac. Harbor manages `mlx-serve` on the host and exposes it to the Compose network through a proxy container.
@@ -86,6 +88,8 @@ harbor launch --backend mlx --model mlx-qwen2.5-3b codex
 
 MLX acceleration does not run inside Harbor's Linux containers. The Harbor service owns lifecycle, config, docs, and integration; Metal inference runs on macOS.
 
+When host management is enabled, `harbor up mlx` automatically installs `mlx-serve` on macOS if it is missing.
+
 ## Choose or Swap Backends
 
 The default Harbor profile starts Ollama and Open WebUI. You can still start another backend directly:
@@ -188,7 +192,7 @@ Use Docker Model Runner when:
 Use MLX when:
 
 - You are on an Apple Silicon Mac and want direct MLX-backed inference.
-- You are comfortable installing `mlx-serve` on the host.
+- You want Harbor to manage `mlx-serve` startup and install it when it is missing.
 - You want Harbor-managed config and proxying for a host-native MLX process.
 
 For many users, the practical workflow is simple: start with Ollama, add llama.cpp when you want direct GGUF serving, use vLLM when serving throughput or model format support matters more, and use DMR or MLX on Apple Silicon when Metal acceleration is the goal.

harbor.sh

@@ -5033,8 +5033,85 @@ sed_replacement_escape() {
     printf '%s' "$1" | sed 's/[&|]/\\&/g'
 }
 
+docker_cli_subcommand_available() {
+    local subcommand="$1"
+    command -v docker >/dev/null 2>&1 && docker "$subcommand" --help 2>&1 | grep -q "Usage:  docker $subcommand"
+}
+
+docker_model_subcommand_available() {
+    local subcommand="$1"
+    docker_cli_subcommand_available model && docker model "$subcommand" --help 2>&1 | grep -q "docker model $subcommand"
+}
+
 docker_model_available() {
-    command -v docker >/dev/null 2>&1 && docker model --help >/dev/null 2>&1
+    docker_cli_subcommand_available model
+}
+
+run_privileged_install_command() {
+    if [ "$(id -u)" -eq 0 ]; then
+        "$@"
+    else
+        if ! command -v sudo >/dev/null 2>&1; then
+            log_error "sudo is required to install host packages automatically."
+            return 1
+        fi
+        sudo "$@"
+    fi
+}
+
+dmr_install_linux_plugin() {
+    if command -v apt-get >/dev/null 2>&1; then
+        log_info "Installing Docker Model Runner plugin with apt."
+        run_privileged_install_command apt-get update
+        run_privileged_install_command apt-get install -y docker-model-plugin
+        return
+    fi
+
+    if command -v dnf >/dev/null 2>&1; then
+        log_info "Installing Docker Model Runner plugin with dnf."
+        run_privileged_install_command dnf install -y docker-model-plugin
+        return
+    fi
+
+    log_error "Docker Model Runner CLI is missing and automatic installation is only supported through apt or dnf on Linux."
+    return 1
+}
+
+dmr_desktop_enable_available() {
+    docker_cli_subcommand_available desktop && docker desktop enable model-runner --help 2>&1 | grep -q 'model-runner'
+}
+
+dmr_install_components() {
+    if docker_model_available; then
+        return 0
+    fi
+
+    log_info "Docker Model Runner CLI is missing; attempting to install or enable it."
+
+    if [[ "$(uname -s)" == "Linux" ]]; then
+        dmr_install_linux_plugin || return 1
+    elif dmr_desktop_enable_available; then
+        docker desktop enable model-runner || return 1
+    else
+        log_error "Docker Model Runner CLI is not available. Install/update Docker Desktop or install docker-model-plugin, then retry."
+        return 1
+    fi
+
+    if ! docker_model_available; then
+        log_error "Docker Model Runner CLI is still unavailable after automatic setup."
+        return 1
+    fi
+}
+
+dmr_install_runner() {
+    if docker model status >/dev/null 2>&1; then
+        return 0
+    fi
+
+    if docker_model_subcommand_available install-runner; then
+        log_info "Installing Docker Model Runner runtime."
+        docker model install-runner || return 1
+    fi
 }
 
 dmr_host_start() {
@@ -5046,10 +5123,8 @@ dmr_host_start() {
         return 0
     fi
 
-    if ! docker_model_available; then
-        log_error "Docker Model Runner CLI is not available. Install/update Docker Desktop and ensure 'docker model' works."
-        return 1
-    fi
+    dmr_install_components || return 1
+    dmr_install_runner || return 1
 
     enable_tcp=$(env_manager get dmr.enable.tcp)
     runner_port=$(env_manager get dmr.runner.port)
@@ -5106,6 +5181,52 @@ mlx_render_config() {
         "$template" > "$output"
 }
 
+mlx_add_uv_tool_bin_to_path() {
+    local tool_bin
+    tool_bin=$(uv tool dir --bin 2>/dev/null || true)
+
+    if [ -n "$tool_bin" ]; then
+        PATH="$tool_bin:$PATH"
+        export PATH
+    fi
+}
+
+mlx_install_uv() {
+    if command -v uv >/dev/null 2>&1; then
+        return 0
+    fi
+
+    if [[ "$(uname -s)" == "Darwin" ]] && command -v brew >/dev/null 2>&1; then
+        log_info "Installing uv with Homebrew."
+        brew install uv || return 1
+        return
+    fi
+
+    log_error "uv is required to install mlx-serve automatically. Install uv, then retry."
+    return 1
+}
+
+mlx_install_components() {
+    if command -v mlx-serve >/dev/null 2>&1; then
+        return 0
+    fi
+
+    if [[ "$(uname -s)" != "Darwin" ]]; then
+        log_error "mlx-serve is missing and automatic MLX host setup is only supported on macOS."
+        return 1
+    fi
+
+    mlx_install_uv || return 1
+    log_info "Installing mlx-serve with uv."
+    uv tool install mlx-serve || return 1
+    mlx_add_uv_tool_bin_to_path
+
+    if ! command -v mlx-serve >/dev/null 2>&1; then
+        log_error "mlx-serve is still unavailable after automatic setup. Ensure the uv tool bin directory is on PATH."
+        return 1
+    fi
+}
+
 mlx_host_start() {
     local manage_host auto_pull model workspace api_key
 
@@ -5115,10 +5236,7 @@ mlx_host_start() {
         return 0
     fi
 
-    if ! command -v mlx-serve >/dev/null 2>&1; then
-        log_error "mlx-serve is not installed on the host. Install it first, then retry: uv tool install mlx-serve"
-        return 1
-    fi
+    mlx_install_components || return 1
 
     mlx_render_config
     workspace=$(mlx_workspace_path)