offgrid-ai 0.6.5 → 0.6.6

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "offgrid-ai",
-  "version": "0.6.5",
+  "version": "0.6.6",
   "description": "Privacy-first CLI for running local LLMs — discover, configure, run, benchmark",
   "author": "Eeshan Srivastava (https://eeshans.com)",
   "type": "module",

package/src/autodetect.mjs

@@ -73,14 +73,13 @@ export function computeFlags(capabilities, modelPath, mmprojPath, draftModelPath
     flags.chatTemplateKwargs = { enable_thinking: true };
   }
 
-  // Build argv. Prefer llama.cpp's Hugging Face resolver when a repo ref is
-  // known, so companion files such as Gemma 4 MTP drafters are discovered by
-  // llama.cpp instead of offgrid-ai maintaining repo-specific file rules.
-  const hfRef = capabilities.hfRepo ? `${capabilities.hfRepo}${capabilities.hfVariant ? `:${capabilities.hfVariant}` : ""}` : null;
-  const argv = hfRef ? ["-hf", hfRef] : ["--model", modelPath];
-
-  if (mmprojPath && !hfRef) argv.push("--mmproj", mmprojPath);
-  if (draftModelPath && !hfRef) argv.push("--spec-draft-model", draftModelPath);
+  // Build argv
+  const argv = [
+    "--model", modelPath,
+  ];
+
+  if (mmprojPath) argv.push("--mmproj", mmprojPath);
+  if (draftModelPath) argv.push("--spec-draft-model", draftModelPath);
 
   argv.push(
     "--host", String(flags.host),

package/src/backends.mjs

@@ -117,16 +117,18 @@ async function scanOmlxModels() {
     if (!response.ok) return [];
     const body = await response.json();
     if (!Array.isArray(body?.data)) return [];
-    return body.data.map((model) => ({
-      id: model.id,
-      label: omlxLabel(model.id),
-      aliasSuggestion: model.id,
-      sizeBytes: 0,
-      quant: null,
-      family: null,
-      backend: "omlx",
-      source: "omlx",
-    })).sort((a, b) => a.label.localeCompare(b.label));
+    return body.data
+      .filter((model) => isChatOmlxModel(model))
+      .map((model) => ({
+        id: model.id,
+        label: omlxLabel(model.id),
+        aliasSuggestion: model.id,
+        sizeBytes: 0,
+        quant: null,
+        family: null,
+        backend: "omlx",
+        source: "omlx",
+      })).sort((a, b) => a.label.localeCompare(b.label));
   } catch {
     return [];
   }
@@ -141,6 +143,14 @@ function isLocalOllamaModel(model) {
   return true;
 }
 
+function isChatOmlxModel(model) {
+  if (typeof model?.id !== "string" || !model.id.trim()) return false;
+  const type = String(model.type ?? model.model_type ?? "").toLowerCase();
+  if (["embedding", "embeddings", "reranker", "tool", "converter"].includes(type)) return false;
+  if (Object.hasOwn(model, "max_model_len") && model.max_model_len === null) return false;
+  return true;
+}
+
 function ollamaLabel(name) {
   return name.replace(/[-_]/g, " ").replace(/^gemma\b/i, "Gemma").replace(/^qwen/i, "Qwen");
 }

package/src/cli.mjs

@@ -284,7 +284,7 @@ function optionTag(text, color, width) {
 function optionStatusTag(kind) {
   const statuses = {
     running: ["RUNNING", pc.green],
-    ready: ["SET UP", pc.green],
+    ready: ["READY", pc.green],
     missing: ["FILE MISSING", pc.red],
     setup: ["NEEDS SETUP", pc.yellow],
   };
@@ -299,7 +299,6 @@ function optionSourceTag(sourceId, label) {
     ollama: pc.green,
     omlx: pc.magenta,
     gguf: pc.cyan,
-    hf: pc.cyan,
   };
   return optionTag(label, colors[sourceId] ?? pc.dim, OPTION_SOURCE_WIDTH);
 }
@@ -308,11 +307,7 @@ function optionLabel({ status, source, name, details = [] }) {
   return [status, source, pc.bold(name), ...details].filter(Boolean).join(OPTION_SEPARATOR);
 }
 
-function optionHint(parts) {
-  return parts.filter(Boolean).join(" · ");
-}
-
-function modelSelectOption(item, { runningProfilesNow, drafters }) {
+function modelSelectOption(item, { runningProfilesNow }) {
   if (item.type === "profile") {
     const { profile } = item;
     const backend = backendFor(profile.backend);
@@ -320,41 +315,27 @@ function modelSelectOption(item, { runningProfilesNow, drafters }) {
     const fileMissing = item.fileMissing;
     const status = optionStatusTag(fileMissing ? "missing" : running ? "running" : "ready");
     const source = optionSourceTag(profile.backend, backend.label);
-    const caps = profile.capabilities ?? {};
-    let mtpLabel;
-    if (profile.drafterPath) mtpLabel = pc.green("MTP on");
-    else if (drafters ? matchDrafter(profile.modelPath, drafters) : null) mtpLabel = pc.yellow("MTP available");
-    else if (caps.architecture === "gemma4") mtpLabel = pc.yellow("MTP needs drafter");
-    const ctxLabel = profile.flags?.ctxSize ? pc.dim(`${(profile.flags.ctxSize / 1000).toFixed(0)}k ctx`) : null;
-    const label = optionLabel({ status, source, name: profile.label, details: [ctxLabel, mtpLabel] });
-    return { value: itemKey(item), label, hint: optionHint([profile.modelAlias, humanCapabilitySummary(caps), profile.baseUrl]) };
+    const label = optionLabel({ status, source, name: profile.label });
+    return { value: itemKey(item), label };
   }
   if (item.type === "new") {
-    const { model, drafter } = item;
-    const caps = detectCapabilities(model.path, model.mmprojPath);
-    const mtpAvailable = caps.mtp || Boolean(drafter) || Boolean(model.hfRepo && /gemma-?4/i.test(`${model.hfRepo} ${model.label}`));
-    let mtpLabel;
-    if (mtpAvailable) mtpLabel = pc.green("MTP available");
-    else if (caps.architecture === "gemma4") mtpLabel = pc.yellow("MTP needs drafter");
+    const { model } = item;
     const label = optionLabel({
       status: optionStatusTag("setup"),
-      source: optionSourceTag(model.hfRepo ? "hf" : "gguf", model.hfRepo ? "Hugging Face" : "GGUF file"),
+      source: optionSourceTag("gguf", "GGUF file"),
       name: model.label,
-      details: [pc.dim(formatBytes(model.sizeBytes)), mtpLabel],
     });
-    return { value: itemKey(item), label, hint: optionHint([basename(model.path), humanCapabilitySummary(caps), model.quant]) };
+    return { value: itemKey(item), label };
   }
   // managed
   const { model, backendId } = item;
   const backend = BACKENDS[backendId];
-  const details = [model.quant ? pc.dim(model.quant) : null, model.sizeBytes ? pc.dim(formatBytes(model.sizeBytes)) : null];
   const label = optionLabel({
     status: optionStatusTag("setup"),
     source: optionSourceTag(backendId, backend.label),
     name: model.label,
-    details,
   });
-  return { value: itemKey(item), label, hint: optionHint([model.id, "available through local backend"]) };
+  return { value: itemKey(item), label };
 }
 
 function actionsForItem(item) {
@@ -435,7 +416,7 @@ function printModelCards(items, { runningProfilesNow, drafters }) {
     for (const item of newModels) {
       const { model, drafter } = item;
       const caps = detectCapabilities(model.path, model.mmprojPath);
-      const mtpAvailable = caps.mtp || Boolean(drafter) || Boolean(model.hfRepo && /gemma-?4/i.test(`${model.hfRepo} ${model.label}`));
+      const mtpAvailable = caps.mtp || Boolean(drafter);
       const mtpLabel = mtpAvailable
         ? pc.green("MTP ✓")
         : (caps.architecture === "gemma4")
@@ -543,7 +524,6 @@ async function printProfileDetails(profile) {
   ];
   if (!isManaged) {
     detailRows.push(
-      ...(profile.hfRepo ? [["Hugging Face", `${profile.hfRepo}${profile.hfVariant ? `:${profile.hfVariant}` : ""}`]] : []),
       ["Local file", fileMissing ? pc.red(`${profile.modelPath} (not found)`) : profile.modelPath ?? "unknown"],
       ["Vision file", profile.mmprojPath ? (existsSync(profile.mmprojPath) ? profile.mmprojPath : pc.red(`${profile.mmprojPath} (not found)`)) : "none"],
       ["Model size", profile.modelPath && existsSync(profile.modelPath) ? formatBytes(statSync(profile.modelPath).size) : "unknown"],
@@ -566,7 +546,7 @@ async function printProfileDetails(profile) {
 
 function printGgufModelDetails(model, drafter) {
   const caps = detectCapabilities(model.path, model.mmprojPath);
-  const mtpAvailable = caps.mtp || Boolean(drafter) || Boolean(model.hfRepo && /gemma-?4/i.test(`${model.hfRepo} ${model.label}`));
+  const mtpAvailable = caps.mtp || Boolean(drafter);
   const mtpLabel = mtpAvailable
     ? pc.green("MTP ✓")
     : (caps.architecture === "gemma4")
@@ -584,7 +564,6 @@ function printGgufModelDetails(model, drafter) {
   ];
   console.log("\n" + renderSection("Downloaded model", renderRows(overviewRows)));
   const detailRows = [
-    ...(model.hfRepo ? [["Hugging Face", `${model.hfRepo}${model.hfVariant ? `:${model.hfVariant}` : ""}`]] : []),
     ["Local file", model.path],
     ["Vision file", model.mmprojPath ?? "none"],
     ["Detected", capabilitySummary(caps)],

package/src/profile-setup.mjs

@@ -34,16 +34,14 @@ export async function configureLocalProfile(prompt, profile) {
   let configured = profile;
   // Re-detect capabilities from the model file and check for drafters
   // so that re-setup can pick up MTP availability, vision changes, etc.
-  const hfSource = profile.hfRepo ? { hfRepo: profile.hfRepo, hfVariant: profile.hfVariant } : {};
-  const freshCaps = { ...detectCapabilities(profile.modelPath, profile.mmprojPath), ...hfSource };
+  const freshCaps = detectCapabilities(profile.modelPath, profile.mmprojPath);
   let drafterPath = profile.drafterPath ?? null;
-  if (!drafterPath && !profile.hfRepo) {
+  if (!drafterPath) {
     const { drafters } = await scanGgufModels();
     const drafter = matchDrafter(profile.modelPath, drafters);
     if (drafter) drafterPath = drafter.path;
   }
-  const hfGemma4Mtp = Boolean(profile.hfRepo && /gemma-?4/i.test(`${profile.hfRepo} ${profile.label}`));
-  const hasMtp = freshCaps.mtp || Boolean(drafterPath) || hfGemma4Mtp;
+  const hasMtp = freshCaps.mtp || Boolean(drafterPath);
   const caps = { ...freshCaps, mtp: hasMtp };
   // If MTP is newly available, switch backend and add drafter path
   if (hasMtp && configured.backend !== "llama-cpp-mtp") {
@@ -73,7 +71,7 @@ export async function configureLocalProfile(prompt, profile) {
     console.log(renderSection("MTP detected", renderRows([
       ["Backend", "llama.cpp MTP"],
       ["Port", String(LLAMA_CPP_MTP_PORT)],
-      ["Flags", `--spec-type draft-mtp --spec-draft-n-max 4${configured.drafterPath && !configured.hfRepo ? " --spec-draft-model <drafter>" : ""}`],
+      ["Flags", `--spec-type draft-mtp --spec-draft-n-max 4${configured.drafterPath ? " --spec-draft-model <drafter>" : ""}`],
     ])));
     if (drafterInfo) console.log(pc.dim(drafterInfo));
     const useMtp = await prompt.yesNo("Use MTP speculative decoding?", true);
@@ -144,7 +142,7 @@ function applyMtpDefaults(profile) {
   const edits = {
     values: { "--spec-type": "draft-mtp", "--spec-draft-n-max": 4 },
   };
-  if (profile.drafterPath && !profile.hfRepo) {
+  if (profile.drafterPath) {
     edits.values["--spec-draft-model"] = profile.drafterPath;
   }
   return applyProfileFlags({ ...profile, backend: "llama-cpp-mtp", providerId: "llama-cpp-mtp" }, flags, edits);

package/src/profiles.mjs

@@ -134,15 +134,11 @@ export function normalizeProfile(profile) {
 export async function createProfileFromModel(model, backendId, drafterPath) {
   const { detectCapabilities } = await import("./autodetect.mjs");
   const caps = detectCapabilities(model.path, model.mmprojPath);
-  // If a drafter is provided, this model supports MTP regardless of filename.
-  // For Hugging Face Gemma 4 GGUF repos, prefer llama.cpp's -hf companion-file
-  // discovery so root mtp-*.gguf drafters do not need offgrid-ai file matching.
-  const hfGemma4Mtp = Boolean(model.hfRepo && /gemma-?4/i.test(`${model.hfRepo} ${model.label}`));
-  const hasMtp = caps.mtp || Boolean(drafterPath) || hfGemma4Mtp;
+  // If a drafter is provided, this model supports MTP regardless of filename
+  const hasMtp = caps.mtp || Boolean(drafterPath);
   const backend = backendId ?? (hasMtp ? "llama-cpp-mtp" : "llama-cpp");
-  const hfSource = model.hfRepo ? { hfRepo: model.hfRepo, hfVariant: model.hfVariant } : {};
   const { flags, argv } = computeFlags(
-    { ...caps, mtp: hasMtp, ...hfSource },
+    { ...caps, mtp: hasMtp },
     model.path,
     model.mmprojPath,
     drafterPath ?? null,
@@ -154,12 +150,11 @@ export async function createProfileFromModel(model, backendId, drafterPath) {
     backend,
     providerId: backend,
     modelAlias: model.aliasSuggestion,
-    source: model.hfRepo ? "huggingface" : model.source,
+    source: model.source,
     modelPath: model.path,
-    mmprojPath: model.hfRepo ? null : model.mmprojPath,
-    ...hfSource,
+    mmprojPath: model.mmprojPath,
     drafterPath: drafterPath ?? null,
-    capabilities: summarizeCapabilities({ ...caps, mtp: hasMtp, ...hfSource }),
+    capabilities: summarizeCapabilities({ ...caps, mtp: hasMtp }),
     preset: null, // no presets — auto-detected
     flags,
     commandArgv: argv,
@@ -177,8 +172,6 @@ function summarizeCapabilities(caps) {
     quant: caps.quant,
     metaCtx: caps.metaCtx,
     mmprojProjectorType: caps.mmprojProjectorType,
-    hfRepo: caps.hfRepo,
-    hfVariant: caps.hfVariant,
     ctxSize: caps.ctxSize,
   };
 }

package/src/scan.mjs

@@ -1,6 +1,6 @@
 import { statSync } from "node:fs";
 import { readdir } from "node:fs/promises";
-import { basename, dirname, join, sep } from "node:path";
+import { basename, dirname, join } from "node:path";
 import { getModelScanDirs } from "./config.mjs";
 import { readGgufMetadata } from "./gguf.mjs";
 
@@ -67,7 +67,6 @@ async function scanOneDir(root) {
         source: "local-gguf",
       });
     } else {
-      const hf = inferHfRef(path, name);
       models.push({
         path,
         mmprojPath,
@@ -76,8 +75,7 @@ async function scanOneDir(root) {
         quant: quantFromName(name),
         sizeBytes,
         backend: "llama-cpp",
-        source: hf ? "huggingface" : "local-gguf",
-        ...(hf ? { hfRepo: hf.repo, hfVariant: hf.variant } : {}),
+        source: "local-gguf",
       });
     }
   }
@@ -160,23 +158,6 @@ function quantFromName(name) {
   return name.match(/(Q\d_K_[A-Z]+|Q\d_[01]|UD-[A-Z0-9_]+)/)?.[1];
 }
 
-export function inferHfRef(path, name = basename(path).replace(/\.gguf$/i, "")) {
-  const parts = path.split(sep);
-  const hubIndex = parts.lastIndexOf("hub");
-  const hfCacheRepo = hubIndex >= 0 ? parts.find((part, index) => index > hubIndex && part.startsWith("models--")) : null;
-  if (hfCacheRepo) {
-    return { repo: hfCacheRepo.replace(/^models--/u, "").replace(/--/gu, "/"), variant: quantFromName(name) ?? name };
-  }
-
-  const lmStudioIndex = parts.lastIndexOf(".lmstudio");
-  if (lmStudioIndex >= 0 && parts[lmStudioIndex + 1] === "models") {
-    const org = parts[lmStudioIndex + 2];
-    const repo = parts[lmStudioIndex + 3];
-    if (org && repo) return { repo: `${org}/${repo}`, variant: quantFromName(name) ?? name };
-  }
-
-  return null;
-}
 
 function safeReadGgufMetadata(path) {
   try {