offgrid-ai 0.16.0 → 0.17.0

package/README.md

@@ -2,7 +2,7 @@
 
 # offgrid-ai
 
-**Helper CLI for running local AI models on Mac with llama-server, mlx-vlm, and oMLX.**
+**Helper CLI for running local AI models on Mac with llama-server and oMLX.**
 
 [![node](https://img.shields.io/badge/node-20%2B-3c873a)](package.json)
 [![platform](https://img.shields.io/badge/platform-macOS%20%7C%20Linux-blue)]()
@@ -12,7 +12,7 @@
 
 ## What is offgrid-ai?
 
-offgrid-ai is a command-line tool that lets you run AI models locally. Running local models with llama-server, mlx-vlm, 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 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.
 
 This is the recommended workflow:
 
@@ -23,8 +23,8 @@ This is the recommended workflow:
 ## 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, mlx-vlm APC/thinking/context flags)
-- Start / stop local servers automatically for chat sessions (llama-server and mlx-vlm)
+- 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)
 
 ## Quick start
 
@@ -77,14 +77,12 @@ Pick a model from the list and press Enter. offgrid-ai configures the rest and o
 offgrid-ai              # primary entry-point for the CLI
 offgrid-ai status       # see if any model is running
 offgrid-ai stop         # stop the running model
-offgrid-ai benchmark    # run a benchmark paired with my local llm benchmark runner
 offgrid-ai uninstall    # remove offgrid-ai
 ```
 
 ## What can I do with it?
 
 - **Chat with local models** — you download the models yourself, and then offgrid-ai helps configure and run then
-- **Run benchmarks** — compare how different models perform on creative or data-science tasks. Pairs with my other [local llm benchmark runner](https://github.com/eeshansrivastava89/local-llm-visual-benchmark)
 - **Keep data private** — everything runs on your machine without any cloud connections
 
 ## Need help?

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "offgrid-ai",
-  "version": "0.16.0",
-  "description": "Privacy-first CLI for running local LLMs — discover, configure, run, benchmark",
+  "version": "0.17.0",
+  "description": "Privacy-first CLI for running local LLMs — discover, configure, run, and chat",
   "author": "Eeshan Srivastava (https://eeshans.com)",
   "type": "module",
   "bin": {
@@ -11,7 +11,6 @@
     "bin/*.mjs",
     "src/*.mjs",
     "src/commands/*.mjs",
-    "src/benchmark/*.mjs",
     "resources/*.py",
     "resources/recommendations.json",
     "install.sh"
@@ -34,7 +33,7 @@
     "start": "node bin/offgrid-ai.mjs",
     "test": "node --test test/*.mjs",
     "test:integration": "OFFGRID_INTEGRATION=1 node --test test/integration/*.mjs",
-    "lint": "eslint src/*.mjs src/commands/*.mjs src/benchmark/*.mjs scripts/*.mjs bin/*.mjs",
+    "lint": "eslint src/*.mjs src/commands/*.mjs scripts/*.mjs bin/*.mjs",
     "check:privacy": "node scripts/privacy-gate.mjs",
     "release:check": "bash scripts/release-check.sh",
     "release:check:fast": "bash scripts/release-check.sh --skip-install --skip-manual",
@@ -42,9 +41,6 @@
     "pretest": "npm run lint"
   },
   "dependencies": {
-    "@earendil-works/pi-agent-core": "^0.80.3",
-    "@earendil-works/pi-ai": "^0.80.3",
-    "@earendil-works/pi-coding-agent": "^0.80.3",
     "@inquirer/prompts": "^8.5.2",
     "picocolors": "^1.1.0"
   },
@@ -61,9 +57,5 @@
     "@eslint/js": "^10.0.1",
     "eslint": "^10.4.1",
     "globals": "^17.6.0"
-  },
-  "allowScripts": {
-    "@google/genai": true,
-    "protobufjs": true
   }
 }

package/src/autodetect.mjs

@@ -55,7 +55,7 @@ export function computeFlags(capabilities, modelPath, mmprojPath, draftModelPath
   const isLowMem = quant && /[Qq]4[_0]/i.test(quant);
 
   const flags = {
-    ...defaultFlagsForBackend(mtp ? "llama-cpp-mtp" : "llama-cpp"),
+    ...defaultFlagsForBackend("llama-cpp"),
     ctxSize: capabilities.ctxSize,
     flashAttention: "on",
     cacheTypeK: isLowMem ? "f16" : "bf16",

package/src/backends.mjs

@@ -7,7 +7,6 @@ import { scanOmlxModelSizes, lookupOmlxModelInfo } from "./mlx-discovery.mjs";
 
 export const LOCAL_HOST = "127.0.0.1";
 export const LLAMA_CPP_PORT = 8080;
-export const LLAMA_CPP_MTP_PORT = 8081;
 export const OMLX_PORT = 8000;
 
 export function baseUrlFor({ host = LOCAL_HOST, port, path = "/v1" }) {
@@ -30,17 +29,6 @@ export const BACKENDS = {
     needsCommandFile: true,
     scanModels: async () => (await scanGgufModels()).models,
   },
-  "llama-cpp-mtp": {
-    id: "llama-cpp-mtp",
-    label: "llama.cpp MTP",
-    type: "local-server",
-    providerId: "llama-cpp-mtp",
-    defaultHost: LOCAL_HOST,
-    defaultPort: LLAMA_CPP_MTP_PORT,
-    defaultBaseUrl: baseUrlFor({ port: LLAMA_CPP_MTP_PORT }),
-    needsCommandFile: true,
-    scanModels: async () => (await scanGgufModels()).models,
-  },
   "omlx": {
     id: "omlx",
     label: "oMLX",

package/src/cli.mjs

@@ -5,7 +5,6 @@ import { modelsCommand } from "./commands/models.mjs";
 import { runCommand } from "./commands/run.mjs";
 import { statusCommand } from "./commands/status.mjs";
 import { stopCommand } from "./commands/stop.mjs";
-import { benchmarkCommand } from "./commands/benchmark.mjs";
 import { uninstallCommand } from "./commands/uninstall.mjs";
 
 async function offerUpdate(argv) {
@@ -45,7 +44,6 @@ export async function run(argv) {
   if (command === "run") return runCommand(argv.slice(1));
   if (command === "status") return statusCommand();
   if (command === "stop") return stopCommand(argv.slice(1));
-  if (command === "benchmark") return benchmarkCommand();
   if (command === "uninstall" || command === "--uninstall") return uninstallCommand(argv.slice(1));
   if (command === "--verbose") return mainFlow();
 
@@ -69,10 +67,9 @@ function printHelp() {
     ["Start", pc.bold("offgrid-ai")],
     ["Status", "offgrid-ai status"],
     ["Stop", "offgrid-ai stop"],
-    ["Benchmark", "offgrid-ai benchmark"],
     ["Uninstall", "offgrid-ai uninstall"],
     ["Version", "offgrid-ai version"],
   ]), { formatBorder: pc.cyan }));
-  console.log("\n" + renderCard("How it works", "Run offgrid-ai, choose a local model, and start chatting in Pi.\n\nFirst run walks you through missing tools. After that, offgrid-ai remembers your model setup.\n\nFor benchmarks, run offgrid-ai benchmark to prepare a visual or data-science benchmark run.", { formatBorder: pc.magenta }));
+  console.log("\n" + renderCard("How it works", "Run offgrid-ai, choose a local model, and start chatting in Pi.\n\nFirst run walks you through missing tools. After that, offgrid-ai remembers your model setup.", { formatBorder: pc.magenta }));
   console.log("\n" + pc.dim("Tip: use --verbose only when you want detailed install output."));
 }

package/src/commands/main.mjs

@@ -4,9 +4,10 @@ import { scanGgufModels } from "../scan.mjs";
 import { loadProfiles } from "../profiles.mjs";
 import { hasPi } from "../harness-pi.mjs";
 import { offerManagedLlamaRuntimeUpdate } from "../runtime.mjs";
-import { hasLmStudioInstalled, hasOmlxInstalled, scanManagedModels } from "../managed.mjs";
+import { offerManagedOmlxUpdate, hasOmlx } from "../omlx-runtime.mjs";
+import { hasLmStudioInstalled, scanManagedModels } from "../managed.mjs";
 import { recommendedModel } from "../recommendations.mjs";
-import { pc, startInteractive, createPrompt } from "../ui.mjs";
+import { pc, startInteractive, createPrompt, renderCard } from "../ui.mjs";
 import { onboardFlow } from "./onboard.mjs";
 import { modelCommandCenter } from "./models.mjs";
 import { statusCommand } from "./status.mjs";
@@ -18,6 +19,7 @@ export async function mainFlow() {
     const runtimePrompt = createPrompt();
     try {
       await offerManagedLlamaRuntimeUpdate(runtimePrompt);
+      await offerManagedOmlxUpdate(runtimePrompt);
     } finally {
       runtimePrompt.close();
     }
@@ -56,14 +58,36 @@ export async function mainFlow() {
   if (!process.stdin.isTTY) return await statusCommand();
 
   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("");
   return await modelCommandCenter({ profiles, ggufModels, managedModels, drafters });
 }
 
+function printStatusHeader({ llamaBinary, managedModels, piInstalled, omlxInstalled, profiles }) {
+  const omlxServerUp = managedModels.some((m) => m.backendId === "omlx" && m.status === "ok");
+  const parts = [
+    llamaBinary ? pc.green("llama.cpp ✓") : pc.red("llama.cpp ✗"),
+  ];
+  if (omlxInstalled) {
+    parts.push(omlxServerUp ? pc.green("oMLX ✓ server up") : pc.yellow("oMLX ✓ server down"));
+  } else {
+    parts.push(pc.red("oMLX ✗"));
+  }
+  parts.push(piInstalled ? pc.green("Pi ✓") : pc.red("Pi ✗"));
+  if (profiles.length > 0) parts.push(pc.dim(`${profiles.length} model${profiles.length === 1 ? "" : "s"}`));
+  console.log(renderCard("offgrid-ai", parts.join(pc.dim("  ·  ")), { formatBorder: pc.cyan }));
+  console.log("");
+}
+
 async function printNoModelsHelp(llamaBinary) {
   console.log(pc.yellow("No models found."));
   console.log(pc.dim("You need to download a model to use offgrid-ai.\n"));
 
-  const omlxInstalled = await hasOmlxInstalled();
+  const omlxInstalled = await hasOmlx();
   const lmStudioInstalled = hasLmStudioInstalled();
   const hasBackends = llamaBinary || omlxInstalled || lmStudioInstalled;
   if (!hasBackends) {

package/src/commands/models.mjs

@@ -1,12 +1,16 @@
-import { ensureDirs } from "../config.mjs";
+import { ensureDirs, getModelScanDirs, addModelScanDir, removeModelScanDir, DEFAULT_MODEL_DIRS, findLlamaServer, HF_HUB_DIR } from "../config.mjs";
+import { existsSync } from "node:fs";
+import { homedir } from "node:os";
+import { join } from "node:path";
 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 } from "../harness-pi.mjs";
-import { configureLocalProfile } from "../profile-setup.mjs";
-import { pc, startInteractive, createPrompt, modelSelect } from "../ui.mjs";
+import { syncPiConfig, removeFromPiConfig, hasPi } from "../harness-pi.mjs";
+import { hasOmlx } from "../omlx-runtime.mjs";
+import { configureLocalProfile, configureManagedProfile } from "../profile-setup.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, printWorkspaceHeader, printBenchmarkLine, printProfileDetails } from "../model-presenters.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");
@@ -31,7 +35,19 @@ export async function modelCommandCenter(initialCatalog) {
     return;
   }
 
-  const catalog = initialCatalog.newModels ? initialCatalog : await loadModelCatalog();
+  let catalog = initialCatalog.newModels ? initialCatalog : await loadModelCatalog();
+
+  while (true) {
+    const result = await showModelPicker(catalog);
+    if (result === "rescan") {
+      catalog = await loadModelCatalog();
+      continue;
+    }
+    return;
+  }
+}
+
+async function showModelPicker(catalog) {
   const normalized = normalizeCatalog(catalog);
   const allItems = buildCatalogItems(normalized);
   if (allItems.length === 0) {
@@ -50,9 +66,6 @@ export async function modelCommandCenter(initialCatalog) {
       if (!(await modelAvailableOnServer(profile))) modelMissingIds.add(profile.id);
     }
   }
-  printWorkspaceHeader(normalized, runningProfilesNow, modelMissingIds);
-  await printBenchmarkLine();
-
   const nameWidth = modelNameWidth(allItems);
 
   const statusFor = (item) => {
@@ -82,16 +95,10 @@ export async function modelCommandCenter(initialCatalog) {
   }
 
   const groups = [];
-  const backendColors = {
-    "llama-cpp": pc.cyan,
-    "llama-cpp-mtp": pc.blue,
-    omlx: pc.magenta,
-  };
   for (const { backendId, sourceId, items } of byBackend.values()) {
     const backendLabel = backendFor(backendId)?.label ?? backendId;
     const sourceLabel = formatSourceLabel(sourceId);
-    const color = backendColors[backendId] ?? pc.dim;
-    const sep = `Inference: ${pc.bold(color(backendLabel))} ${pc.dim("|")} Source: ${sourceLabel} (${items.length})`;
+    const sep = `  ${pc.dim(backendLabel + " · " + sourceLabel + " (" + items.length + ")")}`;
     const groupItems = items.map((item) => {
       const opt = modelSelectOption(item, { runningProfilesNow, modelMissingIds, nameWidth, compact: true });
       return { value: opt.value, label: opt.label, description: opt.description };
@@ -104,13 +111,21 @@ export async function modelCommandCenter(initialCatalog) {
       const opt = modelSelectOption(item, { runningProfilesNow, modelMissingIds, nameWidth, compact: true });
       return { value: opt.value, label: opt.label, description: opt.description };
     });
-    groups.push({ separator: `  ${pc.bold(pc.yellow(`Needs setup (${setupItems.length})`))}`, items: groupItems });
+    groups.push({ separator: `  ${pc.yellow("Needs setup (" + setupItems.length + ")")}`, items: groupItems });
   }
 
+  groups.push({ separator: " ", items: [{ value: "__settings__", label: `${pc.dim("○")}  ${pc.cyan("⚙ Status & settings")}` }] });
+
   const prompt = createPrompt();
   try {
     const selected = await modelSelect("Select a model", groups, { pageSize: 20 });
     if (!selected) return;
+
+    if (selected === "__settings__") {
+      await settingsFlow(prompt);
+      return "rescan";
+    }
+
     const item = allItems.find((candidate) => itemKey(candidate) === selected);
     if (!item) return;
 
@@ -145,10 +160,6 @@ function actionsForItem(item) {
         { value: "run", name: "Start chatting", desc: "Launch and open Pi" },
         { value: "reconfigure", name: "Reconfigure", desc: "Change context, MTP, settings" },
       );
-      const backend = backendFor(item.profile.backend);
-      if (backend.type === "local-server" || backend.type === "managed-server") {
-        available.push({ value: "benchmark", name: "Benchmark", desc: "Prepare a benchmark run" });
-      }
     }
     available.push({ value: "remove", name: "Remove", desc: missing ? "Delete this broken setup" : "Delete this setup" });
     if (missing) {
@@ -156,10 +167,6 @@ function actionsForItem(item) {
         { value: "run", name: "Start chatting", desc: "Launch and open Pi", dimmed: true },
         { value: "reconfigure", name: "Reconfigure", desc: "Change context, MTP, settings", dimmed: true },
       );
-      const backend = backendFor(item.profile.backend);
-      if (backend.type === "local-server" || backend.type === "managed-server") {
-        available.push({ value: "benchmark", name: "Benchmark", desc: "Prepare a benchmark run", dimmed: true });
-      }
     }
     return formatActions(available);
   }
@@ -177,7 +184,7 @@ function actionsForItem(item) {
 
 async function performAction(prompt, action, item) {
   const missing = item.type === "profile" && item.fileMissing;
-  if (missing && ["run", "reconfigure", "benchmark"].includes(action)) {
+  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."));
     return;
   }
@@ -186,16 +193,8 @@ async function performAction(prompt, action, item) {
     if (item.type === "managed") return printManagedModelDetails(item.model, BACKENDS[item.backendId]);
     return printGgufModelDetails(item.model, item.drafter);
   }
-  if (action === "benchmark") {
-    if (item.type === "profile") {
-      const { benchmarkForProfile } = await import("../benchmark.mjs");
-      return await benchmarkForProfile(await readProfile(item.profile.id));
-    }
-    const { benchmarkFlow } = await import("../benchmark.mjs");
-    return await benchmarkFlow();
-  }
   if (action === "run") return await runItem(item);
-  if (action === "reconfigure" || action === "setup") return await setupItem(prompt, item, action);
+  if (action === "reconfigure" || action === "setup") return await setupItem(prompt, item);
   if (action === "remove" && item.type === "profile") return await removeProfileInteractive(item.profile.id);
 }
 
@@ -207,26 +206,28 @@ function printProfileSaved(id) {
   console.log(pc.dim(`  Profile: ${profileJsonPath(id)}`));
 }
 
-async function setupItem(prompt, item, action) {
+async function setupItem(prompt, item) {
   if (item.type === "profile") {
     const configured = await configureLocalProfile(prompt, await readProfile(item.profile.id));
     if (!configured) return;
-    await saveProfile(configured, { writeCommand: true });
+    await saveProfile(configured);
     await syncPiConfig(configured);
     printProfileSaved(configured.id);
     return;
   }
   if (item.type === "managed") {
     const profile = createManagedProfile(item.model, item.backendId);
-    await saveProfile(profile);
-    await syncPiConfig(profile);
-    printProfileSaved(profile.id);
+    const configured = await configureManagedProfile(prompt, profile);
+    if (!configured) return;
+    await saveProfile(configured);
+    await syncPiConfig(configured);
+    printProfileSaved(configured.id);
     return;
   }
   const profile = await createProfileFromModel(item.model, null, item.drafter?.path);
   const configured = await configureLocalProfile(prompt, profile);
   if (!configured) return;
-  await saveProfile(configured, { writeCommand: action === "reconfigure" });
+  await saveProfile(configured);
   await syncPiConfig(configured);
   printProfileSaved(configured.id);
 }
@@ -254,4 +255,77 @@ async function removeProfileInteractive(id) {
   await removeFromPiConfig(profile);
   await deleteProfile(id);
   console.log(pc.green(`Removed ${profile.label} (${profile.id})`));
+}
+
+// ── Settings & discovery path management ───────────────────────────────────
+
+async function settingsFlow(prompt) {
+  while (true) {
+    const llamaBinary = await findLlamaServer();
+    const omlxInstalled = await hasOmlx();
+    const piInstalled = await hasPi();
+
+    let omlxServerUp = false;
+    if (omlxInstalled) {
+      try {
+        const res = await fetch("http://127.0.0.1:8000/v1/models", { signal: AbortSignal.timeout(2000) });
+        omlxServerUp = res.ok;
+      } catch { /* server down */ }
+    }
+
+    console.log("");
+    console.log(renderCard("Runtime status", renderRows([
+      ["llama.cpp", llamaBinary ? pc.green("✓ ") + pc.dim(llamaBinary) : pc.red("✗ not found")],
+      ["oMLX", omlxInstalled ? (omlxServerUp ? pc.green("✓ server up") : pc.yellow("✓ installed · server down")) : pc.red("✗ not found")],
+      ["Pi", piInstalled ? pc.green("✓ installed") : pc.red("✗ not found")],
+    ]), { formatBorder: pc.cyan }));
+
+    const scanDirs = await getModelScanDirs();
+    const defaultSet = new Set(DEFAULT_MODEL_DIRS);
+    const pathLabels = new Map([
+      [join(homedir(), ".lmstudio", "models"), "LM Studio downloads"],
+      [join(homedir(), ".omlx", "models"), "oMLX downloads"],
+      [HF_HUB_DIR, "HuggingFace CLI downloads"],
+    ]);
+    const pathRows = scanDirs.map((dir) => {
+      const exists = existsSync(dir);
+      const isBuiltin = defaultSet.has(dir);
+      const desc = pathLabels.get(dir);
+      const label = `${exists ? pc.green("✓") : pc.red("✗")}  ${dir}`;
+      const tags = [desc, isBuiltin ? "built-in" : "custom"].filter(Boolean).join(pc.dim(" · "));
+      return [label, pc.dim(tags)];
+    });
+    console.log("");
+    console.log(renderCard("Discovery paths", renderRows(pathRows), { formatBorder: pc.magenta }));
+
+    const customDirs = scanDirs.filter((d) => !defaultSet.has(d));
+    const choices = [
+      { value: "add", label: "Add discovery path" },
+      ...(customDirs.length > 0 ? [{ value: "remove", label: "Remove discovery path" }] : []),
+      { value: "back", label: "Back to models" },
+    ];
+    const action = await prompt.choice("Settings", choices, "back");
+
+    if (!action || action === "back") return;
+
+    if (action === "add") {
+      const dir = await prompt.text("Path to model directory", "");
+      if (!dir || !dir.trim()) continue;
+      const cleanDir = dir.trim();
+      if (!existsSync(cleanDir)) {
+        console.log(pc.red(`Directory not found: ${cleanDir}`));
+        continue;
+      }
+      await addModelScanDir(cleanDir);
+      console.log(pc.green(`Added: ${cleanDir}`));
+    }
+
+    if (action === "remove") {
+      const removeChoices = customDirs.map((d) => ({ value: d, label: d }));
+      const toRemove = await prompt.choice("Remove path", removeChoices);
+      if (!toRemove) continue;
+      await removeModelScanDir(toRemove);
+      console.log(pc.green(`Removed: ${toRemove}`));
+    }
+  }
 }

package/src/commands/onboard.mjs

@@ -1,9 +1,9 @@
-import { existsSync } from "node:fs";
-import { ensureDirs, findLlamaServer, hasHomebrew, HF_HUB_DIR } from "../config.mjs";
+import { ensureDirs, findLlamaServer, ensureHomebrewFor, HF_HUB_DIR } 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";
@@ -24,6 +24,7 @@ export async function onboardFlow() {
     console.log(pc.dim("Let's make sure you have everything you need to run local models.\n"));
 
     const llamaBinary = await ensureLlamaRuntime(prompt);
+    await ensureOmlxRuntime(prompt, run);
     if (!(await ensurePi(prompt, run))) return;
 
     const [{ models: ggufModels }, managedModels] = await Promise.all([
@@ -169,35 +170,6 @@ async function offerBackendInstall(prompt, run) {
   await installBackend(prompt, run, choice, model);
 }
 
-async function ensureHomebrewFor(prompt, run, label) {
-  if (await hasHomebrew()) return true;
-  const install = await prompt.yesNo(`Homebrew is needed to install ${label}. Install Homebrew now?`, true);
-  if (!install) {
-    console.log(pc.dim(`Install ${label} manually, or install Homebrew from https://brew.sh and run offgrid-ai again.`));
-    return false;
-  }
-  console.log(pc.cyan("Installing Homebrew..."));
-  try {
-    await run("/bin/bash", ["-c", "NONINTERACTIVE=1 /bin/bash -c \"$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)\""], "Homebrew");
-    for (const path of ["/opt/homebrew/bin", "/usr/local/bin"]) {
-      if (existsSync(path)) {
-        process.env.PATH = `${path}:${process.env.PATH}`;
-        break;
-      }
-    }
-  } catch {
-    console.log(pc.red("✗ Homebrew installation failed."));
-    console.log(pc.dim("Install it manually from https://brew.sh, then run offgrid-ai again."));
-    return false;
-  }
-  if (!(await hasHomebrew())) {
-    console.log(pc.red("Homebrew was installed but not found on PATH. Restart your terminal and run offgrid-ai again."));
-    return false;
-  }
-  console.log(pc.green("✓ Homebrew found"));
-  return true;
-}
-
 async function installBackend(prompt, run, backendId, model) {
   const installer = BACKEND_INSTALLERS[backendId];
   if (!(await ensureHomebrewFor(prompt, run, installer.label))) return;

package/src/commands/run.mjs

@@ -77,7 +77,7 @@ async function ensureLocalServer(profile, backend, options) {
       console.log(pc.yellow("Vision projector is not supported by this llama.cpp build. Retrying text-only."));
       console.log(pc.dim("Update llama.cpp later to re-enable vision for this model."));
       const textOnly = textOnlyProfile(profile);
-      await saveProfile(textOnly, { writeCommand: true });
+      await saveProfile(textOnly);
       return { handled: true, result: await runProfile(textOnly, { ...options, textOnlyRetry: true }) };
     }
     throw err;
@@ -113,6 +113,7 @@ async function launchHarness(profile, options, isManaged, withHarness, backend)
   }
 
   if (!(await hasPiModel(profile))) await syncPiConfig(profile);
+
   try {
     await launchPi(profile);
   } finally {
@@ -121,10 +122,6 @@ async function launchHarness(profile, options, isManaged, withHarness, backend)
         const result = await stopProfile(profile);
         console.log(result.stopped ? pc.green(`[stop] ${result.message}`) : pc.dim(`[stop] ${result.message}`));
       } else {
-        // Managed-server backends (oMLX): unload the model from the
-        // server's memory via its HTTP API. The server itself stays running
-        // (offgrid-ai doesn't manage it), but the model is released — same UX
-        // as local-server backends where stopProfile kills the process.
         const result = await unloadModelFromServer(profile);
         if (result.unloaded) {
           console.log(pc.green(`[unload] ${backend.label}: model unloaded`));

package/src/config.mjs

@@ -3,6 +3,7 @@ import { existsSync } from "node:fs";
 import { homedir } from "node:os";
 import { dirname, join } from "node:path";
 import { readFile, writeFile } from "node:fs/promises";
+import { pc } from "./ui.mjs";
 
 // ── Base directories ──────────────────────────────────────────────────────
 
@@ -25,6 +26,7 @@ export const HF_HUB_DIR = process.env.HF_HUB_CACHE
 
 export const DEFAULT_MODEL_DIRS = [
   join(homedir(), ".lmstudio", "models"),
+  join(homedir(), ".omlx", "models"),
   HF_HUB_DIR,
 ];
 
@@ -46,7 +48,6 @@ const CONFIG_PATH = join(DATA_DIR, "config.json");
 
 const DEFAULT_CONFIG = {
   modelScanDirs: [],
-  benchmarkRepoPath: null,
   binaryOverrides: {},
 };
 
@@ -77,6 +78,21 @@ export async function getModelScanDirs() {
   return [...DEFAULT_MODEL_DIRS, ...config.modelScanDirs].filter((dir, i, arr) => arr.indexOf(dir) === i);
 }
 
+export async function addModelScanDir(dir) {
+  const config = await loadConfig();
+  config.modelScanDirs ??= [];
+  if (!config.modelScanDirs.includes(dir)) {
+    config.modelScanDirs.push(dir);
+    await saveConfig(config);
+  }
+}
+
+export async function removeModelScanDir(dir) {
+  const config = await loadConfig();
+  config.modelScanDirs = (config.modelScanDirs ?? []).filter((d) => d !== dir);
+  await saveConfig(config);
+}
+
 // ── Binary discovery ──────────────────────────────────────────────────────
 
 import { execFile } from "node:child_process";
@@ -124,4 +140,49 @@ export async function hasHomebrew() {
   } catch {
     return false;
   }
+}
+
+/**
+ * Install Homebrew non-interactively and add it to PATH for this process.
+ * Returns true if Homebrew is available after installation.
+ */
+export async function installHomebrew(run) {
+  await run("/bin/bash", ["-c", 'NONINTERACTIVE=1 /bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"'], "Homebrew");
+  for (const path of ["/opt/homebrew/bin", "/usr/local/bin"]) {
+    if (existsSync(path)) {
+      process.env.PATH = `${path}:${process.env.PATH}`;
+      break;
+    }
+  }
+  return await hasHomebrew();
+}
+
+/**
+ * Ensure Homebrew is installed, prompting the user if necessary.
+ * @param {object} prompt - UI prompt interface (needs yesNo)
+ * @param {function} run - runCommand function for verbose command execution
+ * @param {string} label - what we're installing (for the prompt message)
+ * @returns {Promise<boolean>} true if Homebrew is available
+ */
+export async function ensureHomebrewFor(prompt, run, label) {
+  if (await hasHomebrew()) return true;
+  const install = await prompt.yesNo(`Homebrew is needed to install ${label}. Install Homebrew now?`, true);
+  if (!install) {
+    console.log(pc.dim(`Install ${label} manually, or install Homebrew from https://brew.sh and run offgrid-ai again.`));
+    return false;
+  }
+  console.log(pc.cyan("Installing Homebrew..."));
+  try {
+    const success = await installHomebrew(run);
+    if (!success) {
+      console.log(pc.red("Homebrew was installed but not found on PATH. Restart your terminal and run offgrid-ai again."));
+      return false;
+    }
+  } catch {
+    console.log(pc.red("✗ Homebrew installation failed."));
+    console.log(pc.dim("Install it manually from https://brew.sh, then run offgrid-ai again."));
+    return false;
+  }
+  console.log(pc.green("✓ Homebrew found"));
+  return true;
 }