offgrid-ai 0.17.0 → 0.18.1

package/README.md

@@ -2,7 +2,7 @@
 
 # offgrid-ai
 
-**Helper CLI for running local AI models on Mac with llama-server and oMLX.**
+**Run local AI models on your machine — pick, configure, and chat.**
 
 [![node](https://img.shields.io/badge/node-20%2B-3c873a)](package.json)
 [![platform](https://img.shields.io/badge/platform-macOS%20%7C%20Linux-blue)]()
@@ -12,19 +12,23 @@
 
 ## What is offgrid-ai?
 
-offgrid-ai is a command-line tool that lets you run AI models locally. Running local models with llama-server or oMLX have a steep learning curve compared to cloud-based models, so offgrid-ai is designed to abstract away the complexity, while still providing a powerful and flexible way to run local models.
+offgrid-ai is a command-line tool that lets you run AI models locally. Running local models with llama-server or oMLX has a steep learning curve compared to cloud-based models, so offgrid-ai is designed to abstract away the complexity while still providing a powerful and flexible way to run local models.
 
-This is the recommended workflow:
+The recommended workflow:
 
-1. Download models from **LM Studio** or **oMLX**
-2. Do minimal configuration using the `offgrid-ai` command
-3. Run the model with `offgrid-ai` with Pi in interactive mode
+1. Download models from **HuggingFace** (or use models you already have from LM Studio, oMLX, etc.)
+2. Configure using the `offgrid-ai` interactive setup
+3. Start chatting in **Pi** — offgrid-ai handles the server lifecycle
 
 ## Core Features
-- Auto-detects available models from LM Studio, oMLX, and HuggingFace
-- Auto-detects MTP (multi-token prediction) or QAT (quantization aware training) models, and applies the correct flags for llama.cpp
-- Auto-applies the optimal flags for the model type (llama.cpp server flags, oMLX auto-start and cache management)
-- Start / stop local servers automatically for chat sessions (llama-server and oMLX)
+
+- **Download models** from HuggingFace with a quant picker and RAM fit indicators
+- **Auto-detects** models from LM Studio, oMLX, and HuggingFace cache
+- **Glass-box setup** — every configuration flag gets an explanation card with tradeoffs and memory impact
+- **Model management** — delete models from disk, remove configurations, reconfigure settings
+- **Auto-detects MTP** (multi-token prediction) and **QAT** (quantization-aware training) models, applies the correct flags
+- **Start / stop servers** automatically for chat sessions (llama-server and oMLX)
+- **oMLX integration** — auto-start, MTP enable via admin API, restart after download/deletion
 
 ## Quick start
 
@@ -52,7 +56,7 @@ The curl installer is recommended for first-time setup because it also verifies
 
 ### 2. Pick a model
 
-The first time you run offgrid-ai, it looks for models already on your machine. If it does not find any, it tells you how to get one.
+The first time you run offgrid-ai, it looks for models already on your machine. If it doesn't find any, you can download one directly from HuggingFace — just pick "↓ Download a model" and enter a repo ID (e.g. `unsloth/gemma-4-E2B-it-GGUF`).
 
 <img width="808" height="274" alt="image" src="https://github.com/user-attachments/assets/6e1583ab-65db-423c-b0eb-b627586fbf86" />
 
@@ -74,16 +78,17 @@ Pick a model from the list and press Enter. offgrid-ai configures the rest and o
 ## Everyday commands
 
 ```bash
-offgrid-ai              # primary entry-point for the CLI
+offgrid-ai              # model picker — pick, configure, download, or manage models
 offgrid-ai status       # see if any model is running
 offgrid-ai stop         # stop the running model
 offgrid-ai uninstall    # remove offgrid-ai
 ```
 
-## What can I do with it?
+## Platform support
 
-- **Chat with local models** — you download the models yourself, and then offgrid-ai helps configure and run then
-- **Keep data private** — everything runs on your machine without any cloud connections
+- **macOS (Apple Silicon)** — full support: llama.cpp (GGUF) + oMLX (MLX)
+- **Linux** — llama.cpp (GGUF) only. oMLX is Apple Silicon exclusive.
+- **Windows** — not supported
 
 ## Need help?
 
@@ -104,4 +109,4 @@ node bin/offgrid-ai.mjs
 
 ## License
 
-Personal project by [Eeshan Srivastava](https://eeshans.com).
+Personal project by [Eeshan Srivastava](https://eeshans.com).

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "offgrid-ai",
-  "version": "0.17.0",
+  "version": "0.18.1",
   "description": "Privacy-first CLI for running local LLMs — discover, configure, run, and chat",
   "author": "Eeshan Srivastava (https://eeshans.com)",
   "type": "module",

package/resources/recommendations.json

@@ -2,56 +2,56 @@
   "models": [
     {
       "id": "gemma-4-e2b",
-      "label": "Gemma 4 E2B",
+      "label": "Gemma 4 E2B (Q4_K_S)",
       "minRamGb": 8,
       "gguf": "unsloth/gemma-4-E2B-it-GGUF/gemma-4-E2B-it-Q4_K_S.gguf",
       "mlx": "mlx-community/gemma-4-e2b-it-4bit"
     },
     {
       "id": "qwen-3.5-9b",
-      "label": "Qwen 3.5 9B",
+      "label": "Qwen 3.5 9B (Q4_K_S)",
       "minRamGb": 16,
       "gguf": "unsloth/Qwen3.5-9B-GGUF/Qwen3.5-9B-UD-Q4_K_S.gguf",
       "mlx": "lmstudio-community/Qwen3.5-9B-MLX-4bit"
     },
     {
       "id": "gemma-4-12b-qat",
-      "label": "Gemma 4 12B",
+      "label": "Gemma 4 12B (Q4_K_XL)",
       "minRamGb": 24,
       "gguf": "unsloth/gemma-4-12B-it-qat-GGUF/gemma-4-12B-it-qat-UD-Q4_K_XL.gguf",
       "mlx": "mlx-community/gemma-4-12B-it-qat-4bit"
     },
     {
       "id": "gemma-4-26b",
-      "label": "Gemma 4 26B",
+      "label": "Gemma 4 26B (Q4_K_XL)",
       "minRamGb": 32,
       "gguf": "unsloth/gemma-4-26B-A4B-it-qat-GGUF/gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf",
       "mlx": "mlx-community/gemma-4-26b-a4b-4bit"
     },
     {
       "id": "qwen-3.6-35b-compact",
-      "label": "Qwen 3.6 35B",
+      "label": "Qwen 3.6 35B (Q4_K_S)",
       "minRamGb": 32,
       "gguf": "unsloth/Qwen3.6-35B-A3B-GGUF/Qwen3.6-35B-A3B-UD-Q4_K_S.gguf",
       "mlx": "mlx-community/Qwen3.6-35B-A3B-4bit"
     },
     {
       "id": "qwen-3.6-35b",
-      "label": "Qwen 3.6 35B",
+      "label": "Qwen 3.6 35B (Q4_K_M)",
       "minRamGb": 48,
       "gguf": "unsloth/Qwen3.6-35B-A3B-GGUF/Qwen3.6-35B-A3B-UD-Q4_K_M.gguf",
       "mlx": ""
     },
     {
       "id": "gemma-4-31b",
-      "label": "Gemma 4 31B",
+      "label": "Gemma 4 31B (Q4_K_XL)",
       "minRamGb": 64,
       "gguf": "unsloth/gemma-4-31B-it-qat-GGUF/gemma-4-31B-it-qat-UD-Q4_K_XL.gguf",
       "mlx": "mlx-community/gemma-4-31b-4bit"
     },
     {
       "id": "qwen-3.6-27b",
-      "label": "Qwen 3.6 27B",
+      "label": "Qwen 3.6 27B (Q4_K_M)",
       "minRamGb": 64,
       "gguf": "unsloth/Qwen3.6-27B-MTP-GGUF/Qwen3.6-27B-Q4_K_M.gguf",
       "mlx": "mlx-community/Qwen3.6-27B-4bit"

package/src/commands/main.mjs

@@ -59,10 +59,7 @@ export async function mainFlow() {
 
   startInteractive("offgrid-ai");
   printStatusHeader({ llamaBinary, managedModels, piInstalled, omlxInstalled: await hasOmlx(), profiles });
-  console.log(pc.dim("  How to get models — offgrid-ai finds them on disk after you download:"));
-  console.log(pc.dim("    LM Studio       Open LM Studio app, browse and download"));
-  console.log(pc.dim("    oMLX            Open oMLX app, browse and download"));
-  console.log(pc.dim("    HuggingFace     hf download mlx-community/gemma-4-e2b-it-4bit"));
+  console.log(pc.dim("  No models? Pick \"↓ Download a model\" below — offgrid-ai downloads from HuggingFace"));
   console.log("");
   return await modelCommandCenter({ profiles, ggufModels, managedModels, drafters });
 }

package/src/commands/models.mjs

@@ -1,19 +1,22 @@
 import { ensureDirs, getModelScanDirs, addModelScanDir, removeModelScanDir, DEFAULT_MODEL_DIRS, findLlamaServer, HF_HUB_DIR } from "../config.mjs";
 import { existsSync } from "node:fs";
+import { rm, unlink } from "node:fs/promises";
 import { homedir } from "node:os";
 import { join } from "node:path";
+import { stripVTControlCharacters } from "node:util";
 import { backendFor, BACKENDS } from "../backends.mjs";
 import { createProfileFromModel, readProfile, saveProfile, deleteProfile, profileJsonPath } from "../profiles.mjs";
 import { isProfileRunning, isProfileServerUp, modelAvailableOnServer, stopProfile } from "../process.mjs";
 import { syncPiConfig, removeFromPiConfig, hasPi } from "../harness-pi.mjs";
-import { hasOmlx } from "../omlx-runtime.mjs";
+import { hasOmlx, offerOmlxRestart } from "../omlx-runtime.mjs";
 import { configureLocalProfile, configureManagedProfile } from "../profile-setup.mjs";
+import { findOmlxModelDir } from "../mlx-discovery.mjs";
 import { pc, startInteractive, createPrompt, modelSelect, renderCard, renderRows } from "../ui.mjs";
 import { buildCatalogItems, createManagedProfile, itemKey, loadModelCatalog, normalizeCatalog } from "../model-catalog.mjs";
 import { modelSelectOption, modelNameWidth, inferBackendId, formatSourceLabel, discoverySourceForItem, printGgufModelDetails, printManagedModelDetails, printProfileDetails } from "../model-presenters.mjs";
 import { runProfile } from "./run.mjs";
-
-const { stripVTControlCharacters } = await import("node:util");
+import { downloadFlow } from "../download.mjs";
+import { execFileAsync } from "../exec.mjs";
 
 export async function modelsCommand(argv) {
   await ensureDirs();
@@ -66,6 +69,15 @@ async function showModelPicker(catalog) {
       if (!(await modelAvailableOnServer(profile))) modelMissingIds.add(profile.id);
     }
   }
+  // Flag all missing profiles (file missing for llama.cpp, model missing
+  // for oMLX managed-server) so actionsForItem/performAction can handle both
+  // cases uniformly.
+  for (const item of allItems) {
+    if (item.type === "profile") {
+      item.missing = item.fileMissing || modelMissingIds.has(item.profile.id);
+    }
+  }
+
   const nameWidth = modelNameWidth(allItems);
 
   const statusFor = (item) => {
@@ -114,7 +126,10 @@ async function showModelPicker(catalog) {
     groups.push({ separator: `  ${pc.yellow("Needs setup (" + setupItems.length + ")")}`, items: groupItems });
   }
 
-  groups.push({ separator: " ", items: [{ value: "__settings__", label: `${pc.dim("○")}  ${pc.cyan("⚙ Status & settings")}` }] });
+  groups.push({ separator: " ", items: [
+    { value: "__download__", label: `${pc.dim("○")}  ${pc.green("↓ Download a model")}` },
+    { value: "__settings__", label: `${pc.dim("○")}  ${pc.cyan("⚙ Status & settings")}` },
+  ] });
 
   const prompt = createPrompt();
   try {
@@ -123,7 +138,14 @@ async function showModelPicker(catalog) {
 
     if (selected === "__settings__") {
       await settingsFlow(prompt);
-      return "rescan";
+      console.log("");
+      return;
+    }
+
+    if (selected === "__download__") {
+      await downloadFlow(prompt);
+      console.log("");
+      return;
     }
 
     const item = allItems.find((candidate) => itemKey(candidate) === selected);
@@ -133,6 +155,7 @@ async function showModelPicker(catalog) {
     const action = await prompt.choice(item.label, actions, actions[0].value);
     if (!action) return;
     await performAction(prompt, action, item);
+    console.log("");
   } finally {
     prompt.close();
   }
@@ -144,13 +167,13 @@ function formatActions(rawActions) {
   const width = Math.max(17, maxName + 2);
   return rawActions.map((a) => {
     const name = a.dimmed ? pc.dim(pc.strikethrough(a.name.padEnd(width).slice(0, width))) : pc.bold(a.name.padEnd(width).slice(0, width));
-    const desc = a.dimmed ? pc.red("file not found") : pc.dim(a.desc);
+    const desc = a.dimmed ? pc.red("not available") : pc.dim(a.desc);
     return { value: a.value, label: name + sep + desc };
   });
 }
 
 function actionsForItem(item) {
-  const missing = item.type === "profile" && item.fileMissing;
+  const missing = item.type === "profile" && item.missing;
   if (item.type === "profile") {
     const available = [
       { value: "inspect", name: "Details", desc: "Paths, ports, flags" },
@@ -161,7 +184,8 @@ function actionsForItem(item) {
         { value: "reconfigure", name: "Reconfigure", desc: "Change context, MTP, settings" },
       );
     }
-    available.push({ value: "remove", name: "Remove", desc: missing ? "Delete this broken setup" : "Delete this setup" });
+    available.push({ value: "remove_config", name: "Remove configuration", desc: "Delete this setup, keep model files" });
+    available.push({ value: "delete_model", name: "Delete model", desc: "Permanently remove from disk" });
     if (missing) {
       available.unshift(
         { value: "run", name: "Start chatting", desc: "Launch and open Pi", dimmed: true },
@@ -174,18 +198,22 @@ function actionsForItem(item) {
     return formatActions([
       { value: "setup", name: "Set up", desc: "Configure and save" },
       { value: "inspect", name: "Details", desc: "Model info" },
+      { value: "delete_model", name: "Delete model", desc: "Permanently remove from disk" },
     ]);
   }
   return formatActions([
     { value: "setup", name: "Set up", desc: `Connect via ${BACKENDS[item.backendId].label}` },
     { value: "inspect", name: "Details", desc: "Model info" },
+    { value: "delete_model", name: "Delete model", desc: "Permanently remove from disk" },
   ]);
 }
 
 async function performAction(prompt, action, item) {
-  const missing = item.type === "profile" && item.fileMissing;
+  const missing = item.type === "profile" && item.missing;
   if (missing && ["run", "reconfigure"].includes(action)) {
-    console.log(pc.red("This model's file is no longer on disk. Remove the setup or move the file back."));
+    const backend = item.type === "profile" ? backendFor(item.profile.backend) : null;
+    const reason = backend?.type === "managed-server" ? "model is no longer available on the server" : "model file is no longer on disk";
+    console.log(pc.red(`This model's ${reason}. Remove the setup or restore the model.`));
     return;
   }
   if (action === "inspect") {
@@ -195,7 +223,8 @@ async function performAction(prompt, action, item) {
   }
   if (action === "run") return await runItem(item);
   if (action === "reconfigure" || action === "setup") return await setupItem(prompt, item);
-  if (action === "remove" && item.type === "profile") return await removeProfileInteractive(item.profile.id);
+  if (action === "remove_config" && item.type === "profile") return await removeProfileInteractive(item.profile.id);
+  if (action === "delete_model") return await deleteModelFromSource(prompt, item);
 }
 
 async function runItem(item) {
@@ -257,6 +286,161 @@ async function removeProfileInteractive(id) {
   console.log(pc.green(`Removed ${profile.label} (${profile.id})`));
 }
 
+// ── Delete model from source ───────────────────────────────────────────────
+
+/** Extract HuggingFace repo ID from a cache path. */
+function hfRepoFromPath(path) {
+  const hubPart = path.slice(HF_HUB_DIR.length);
+  const match = hubPart.match(/models--(.+?)(?=\/|$)/);
+  if (!match) return null;
+  return match[1].replace(/--/g, "/");
+}
+
+/** Determine where a model's files live on disk. */
+async function modelLocationForItem(item) {
+  if (item.type === "profile") {
+    const backend = backendFor(item.profile.backend);
+    if (backend.type === "managed-server") {
+      const modelId = item.profile.omlxModel || item.profile.modelAlias || item.profile.id;
+      // oMLX model IDs may not include the org prefix, so search recursively
+      const dir = await findOmlxModelDir(modelId);
+      return { kind: "mlx", dir: dir ?? join(homedir(), ".omlx", "models", ...modelId.replace(/--/g, "/").split("/").filter(Boolean)), modelId };
+    }
+    const modelPath = item.profile.modelPath;
+    if (!modelPath) return { kind: "unknown" };
+    if (modelPath.startsWith(HF_HUB_DIR)) {
+      return { kind: "hf-cache", path: modelPath, repoId: hfRepoFromPath(modelPath) };
+    }
+    return { kind: "file", path: modelPath };
+  }
+  if (item.type === "new") {
+    const modelPath = item.model?.path;
+    if (!modelPath) return { kind: "unknown" };
+    if (modelPath.startsWith(HF_HUB_DIR)) {
+      return { kind: "hf-cache", path: modelPath, repoId: hfRepoFromPath(modelPath) };
+    }
+    return { kind: "file", path: modelPath };
+  }
+  if (item.type === "managed") {
+    const modelId = item.model?.id;
+    if (!modelId) return { kind: "unknown" };
+    // oMLX model IDs may not include the org prefix, so search recursively
+    const dir = await findOmlxModelDir(modelId);
+    return { kind: "mlx", dir: dir ?? join(homedir(), ".omlx", "models", ...modelId.replace(/--/g, "/").split("/").filter(Boolean)), modelId };
+  }
+  return { kind: "unknown" };
+}
+
+async function deleteModelFromSource(prompt, item) {
+  const loc = await modelLocationForItem(item);
+
+  if (loc.kind === "unknown") {
+    console.log(pc.yellow("Could not determine where this model's files are located."));
+    return;
+  }
+
+  // Show what will be deleted
+  let locationLabel;
+  if (loc.kind === "hf-cache") {
+    locationLabel = loc.path ?? loc.repoId;
+  } else if (loc.kind === "mlx") {
+    locationLabel = loc.dir;
+  } else if (loc.kind === "file") {
+    locationLabel = loc.path;
+  }
+
+  console.log(pc.yellow("\nThis will permanently delete " + (item.type === "profile" ? "the configuration and the model from:" : "the model from:")));
+  console.log(pc.dim(`  ${locationLabel}`));
+
+  const confirmed = await prompt.yesNo("Delete this model?", false);
+  if (!confirmed) {
+    console.log(pc.dim("Cancelled."));
+    return;
+  }
+
+  // Stop running server if needed
+  if (item.type === "profile" && await isProfileRunning(item.profile)) {
+    console.log(pc.dim("Stopping running server..."));
+    await stopProfile(item.profile);
+  }
+
+  // Delete files
+  if (loc.kind === "hf-cache" && loc.repoId) {
+    const cacheDir = join(HF_HUB_DIR, `models--${loc.repoId.replace(/\//g, "--")}`);
+    try {
+      const { stdout } = await execFileAsync("hf", ["cache", "rm", `model/${loc.repoId}`, "--yes"], { timeout: 30000 });
+      if (stdout.trim()) console.log(pc.dim(stdout.trim()));
+      // Verify the directory is actually gone
+      if (existsSync(cacheDir)) {
+        console.log(pc.red(`✗ Model still exists at ${cacheDir}`));
+        console.log(pc.dim(`Delete manually: hf cache rm model/${loc.repoId}`));
+      } else {
+        console.log(pc.green(`✓ Deleted ${loc.repoId} from HuggingFace cache`));
+      }
+    } catch (err) {
+      const detail = err.stderr?.trim() || err.message;
+      console.log(pc.red(`✗ Failed: ${detail}`));
+      console.log(pc.dim(`Delete manually: hf cache rm model/${loc.repoId}`));
+    }
+  } else if (loc.kind === "mlx") {
+    const omlxModelsRoot = join(homedir(), ".omlx", "models");
+    // Safety guard: never delete outside ~/.omlx/models/
+    if (!loc.dir.startsWith(omlxModelsRoot + "/") && loc.dir !== omlxModelsRoot) {
+      console.log(pc.red(`✗ Refusing to delete: path is outside ~/.omlx/models/`));
+      console.log(pc.dim(`  Target: ${loc.dir}`));
+      console.log(pc.dim(`Delete manually if needed: rm -rf ${loc.dir}`));
+      return;
+    }
+    if (!existsSync(loc.dir)) {
+      console.log(pc.yellow(`Directory not found: ${loc.dir}`));
+      console.log(pc.dim("Model files may have already been removed, or oMLX loaded them from a different location."));
+    } else {
+      try {
+        await rm(loc.dir, { recursive: true, force: true });
+      } catch (err) {
+        console.log(pc.red(`✗ Failed: ${err.message}`));
+        console.log(pc.dim(`Delete manually: rm -rf ${loc.dir}`));
+        return;
+      }
+      // Verify deletion
+      if (existsSync(loc.dir)) {
+        console.log(pc.red(`✗ Directory still exists: ${loc.dir}`));
+        console.log(pc.dim(`Delete manually: rm -rf ${loc.dir}`));
+      } else {
+        console.log(pc.green(`✓ Deleted ${loc.dir}`));
+        await offerOmlxRestart(prompt, "to update its model list");
+      }
+    }
+  } else if (loc.kind === "file") {
+    if (!existsSync(loc.path)) {
+      console.log(pc.yellow(`File not found: ${loc.path}`));
+      console.log(pc.dim("Model file may have already been removed."));
+    } else {
+      try {
+        await unlink(loc.path);
+      } catch (err) {
+        console.log(pc.red(`✗ Failed: ${err.message}`));
+        console.log(pc.dim(`Delete manually: rm ${loc.path}`));
+        return;
+      }
+      // Verify deletion
+      if (existsSync(loc.path)) {
+        console.log(pc.red(`✗ File still exists: ${loc.path}`));
+        console.log(pc.dim(`Delete manually: rm ${loc.path}`));
+      } else {
+        console.log(pc.green(`✓ Deleted ${loc.path}`));
+      }
+    }
+  }
+
+  // Remove profile configuration if one exists
+  if (item.type === "profile") {
+    await removeFromPiConfig(item.profile);
+    await deleteProfile(item.profile.id);
+    console.log(pc.dim(`Removed configuration: ${item.profile.id}`));
+  }
+}
+
 // ── Settings & discovery path management ───────────────────────────────────
 
 async function settingsFlow(prompt) {
@@ -268,7 +452,7 @@ async function settingsFlow(prompt) {
     let omlxServerUp = false;
     if (omlxInstalled) {
       try {
-        const res = await fetch("http://127.0.0.1:8000/v1/models", { signal: AbortSignal.timeout(2000) });
+        const res = await fetch(`${BACKENDS.omlx.defaultBaseUrl}/models`, { signal: AbortSignal.timeout(2000) });
         omlxServerUp = res.ok;
       } catch { /* server down */ }
     }
@@ -302,11 +486,11 @@ async function settingsFlow(prompt) {
     const choices = [
       { value: "add", label: "Add discovery path" },
       ...(customDirs.length > 0 ? [{ value: "remove", label: "Remove discovery path" }] : []),
-      { value: "back", label: "Back to models" },
+      { value: "done", label: "Done" },
     ];
-    const action = await prompt.choice("Settings", choices, "back");
+    const action = await prompt.choice("Settings", choices, "done");
 
-    if (!action || action === "back") return;
+    if (!action || action === "done") return;
 
     if (action === "add") {
       const dir = await prompt.text("Path to model directory", "");

package/src/commands/onboard.mjs

@@ -1,16 +1,13 @@
-import { ensureDirs, findLlamaServer, ensureHomebrewFor, HF_HUB_DIR } from "../config.mjs";
+import { ensureDirs, findLlamaServer } from "../config.mjs";
 import { BACKENDS } from "../backends.mjs";
 import { scanGgufModels } from "../scan.mjs";
 import { hasPi } from "../harness-pi.mjs";
 import { offerManagedLlamaRuntimeUpdate } from "../runtime.mjs";
 import { ensureOmlxRuntime } from "../omlx-runtime.mjs";
 import { scanManagedModels } from "../managed.mjs";
-import { BACKEND_INSTALL_CHOICES, BACKEND_INSTALLERS } from "../backend-installers.mjs";
-import { recommendedModel, selectFormat, allFittingModels } from "../recommendations.mjs";
-import { hasHuggingfaceHub, resolveHfDownload, downloadToHfCache } from "../huggingface.mjs";
-import { detectHardware, getFreeDiskBytes, installedRamGB } from "../hardware.mjs";
+import { downloadFlow } from "../download.mjs";
 import { runCommand } from "../exec.mjs";
-import { pc, formatBytes, renderRows, renderSection, startInteractive, createPrompt } from "../ui.mjs";
+import { pc, renderRows, renderSection, startInteractive, createPrompt } from "../ui.mjs";
 
 export async function onboardFlow() {
   await ensureDirs();
@@ -37,12 +34,10 @@ export async function onboardFlow() {
     if (hasModels) {
       printFoundModels(ggufModels, managedModels, llamaBinary);
     } else {
-      const canDownload = await hasHuggingfaceHub();
-      if (canDownload) {
-        const downloaded = await offerModelDownload(prompt);
-        if (downloaded) return;
+      const downloaded = await downloadFlow(prompt);
+      if (!downloaded) {
+        console.log(pc.dim("\nRun offgrid-ai again when you've downloaded a model."));
       }
-      await offerBackendInstall(prompt, run);
       return;
     }
 
@@ -109,98 +104,3 @@ function printFoundModels(ggufModels, managedModels, llamaBinary) {
   }
 }
 
-async function offerModelDownload(prompt) {
-  const hardware = detectHardware();
-  const candidates = allFittingModels(hardware)
-    .map((entry) => ({ entry, format: selectFormat(entry, hardware) }))
-    .filter((item) => item.format === "gguf");
-  if (candidates.length === 0) {
-    console.log(pc.yellow("No curated models fit your hardware."));
-    return false;
-  }
-
-  const primary = candidates[0];
-  console.log(renderSection("Download a recommended model", renderRows([
-    ["Model", pc.bold(primary.entry.label)],
-    ["Format", primary.format],
-    ["Minimum RAM", String(primary.entry.minRamGb) + " GB"],
-    ["Your RAM", installedRamGB() + " GB"],
-  ]), { formatBorder: pc.cyan }));
-
-  const shouldDownload = await prompt.yesNo("Download " + primary.entry.label + " (" + primary.format + ")?", true);
-  if (!shouldDownload) return false;
-
-  const hfRef = primary.entry.gguf;
-  try {
-    const plan = await resolveHfDownload(hfRef);
-    console.log(pc.dim("Total size: " + formatBytes(plan.totalSizeBytes)));
-    const freeBytes = getFreeDiskBytes(HF_HUB_DIR);
-    if (plan.totalSizeBytes > 0 && freeBytes < plan.totalSizeBytes * 1.1) {
-      console.log(pc.red(`Not enough disk space in ${HF_HUB_DIR}: need ~${formatBytes(plan.totalSizeBytes)}, only ${formatBytes(freeBytes)} free.`));
-      return false;
-    }
-    await downloadToHfCache(plan, {
-      onProgress({ percentage }) {
-        process.stdout.write(pc.cyan("\r  " + percentage + "% downloaded"));
-      },
-    });
-    process.stdout.write("\n");
-    console.log(pc.green("✓ Download complete. Run offgrid-ai to use the model."));
-    return true;
-  } catch (err) {
-    console.log(pc.red("Download failed: " + err.message));
-    return false;
-  }
-}
-
-async function offerBackendInstall(prompt, run) {
-  console.log(pc.yellow("\nNo models found."));
-  console.log(pc.dim("You need at least one model backend to use offgrid-ai.\n"));
-  const choice = await prompt.choice("Install a model backend?", BACKEND_INSTALL_CHOICES, "lmstudio");
-  const model = recommendedModel();
-
-  if (choice === "skip") {
-    console.log(pc.dim("Run offgrid-ai again when you've set up a model backend."));
-    return;
-  }
-  if (choice === "all") {
-    await installAllBackends(prompt, run, model);
-    return;
-  }
-  await installBackend(prompt, run, choice, model);
-}
-
-async function installBackend(prompt, run, backendId, model) {
-  const installer = BACKEND_INSTALLERS[backendId];
-  if (!(await ensureHomebrewFor(prompt, run, installer.label))) return;
-  console.log(pc.cyan(`Installing ${installer.label} via Homebrew...`));
-  try {
-    await runInstallerCommands(run, installer);
-    installer.success(model);
-  } catch {
-    console.log(pc.red(`✗ ${installer.label} installation failed.`));
-    console.log(pc.dim(installer.failure));
-  }
-}
-
-async function installAllBackends(prompt, run, model) {
-  if (!(await ensureHomebrewFor(prompt, run, "model backends"))) return;
-  const installed = [];
-  for (const installer of Object.values(BACKEND_INSTALLERS)) {
-    console.log(pc.cyan(`Installing ${installer.label} via Homebrew...`));
-    try {
-      await runInstallerCommands(run, installer);
-      installed.push(installer.label);
-    } catch {
-      console.log(pc.yellow(installer.allFailure));
-    }
-  }
-  if (installed.length > 0) {
-    console.log(pc.green(`\n✓ Installed: ${installed.join(", ")}`));
-    console.log(pc.dim(`Recommended for your machine (${installedRamGB()}GB RAM): ${model.label}`));
-  }
-}
-
-async function runInstallerCommands(run, installer) {
-  for (const [cmd, args, label] of installer.commands) await run(cmd, args, label);
-}

package/src/commands/run.mjs

@@ -46,6 +46,7 @@ export async function runProfile(profile, options = {}) {
 
   printMemoryEstimate(profile, isManaged);
   await launchHarness(profile, options, isManaged, withHarness, backend);
+  console.log("");
 }
 
 async function ensureLocalServer(profile, backend, options) {

package/src/commands/status.mjs

@@ -85,4 +85,5 @@ export async function statusCommand() {
       ["Server", profile.baseUrl],
     ]), { formatBorder: status.ready ? pc.green : pc.yellow }));
   }
+  console.log("");
 }

package/src/commands/stop.mjs

@@ -34,6 +34,7 @@ export async function stopCommand(argv) {
 
     const targets = selected === "__all" ? running : running.filter((item) => item.profile.id === selected);
     for (const { profile } of targets) await printStopResult(profile);
+    console.log("");
   } finally {
     prompt.close();
   }

package/src/discovery-shared.mjs

@@ -1,6 +1,5 @@
-// Shared discovery helpers used by both the GGUF scanner (scan.mjs) and the
-// MLX scanner (mlx-discovery.mjs). Keeping these here avoids a cross-dependency
-// between the two format-specific scanners.
+// Helpers for the GGUF scanner (scan.mjs). Centralizes model-size and embedding-type
+// constants so they're defined once rather than duplicated across scanners.
 
 import { basename, dirname } from "node:path";