turbollm 1.6.1 → 1.6.2

package/LICENSE.md

@@ -1,105 +1,105 @@
-# Functional Source License, Version 1.1, ALv2 Future License
-
-## Abbreviation
-
-FSL-1.1-ALv2
-
-## Notice
-
-Copyright 2026 Mohit Soni
-
-## Terms and Conditions
-
-### Licensor ("We")
-
-The party offering the Software under these Terms and Conditions.
-
-### The Software
-
-The "Software" is each version of the software that we make available under
-these Terms and Conditions, as indicated by our inclusion of these Terms and
-Conditions with the Software.
-
-### License Grant
-
-Subject to your compliance with this License Grant and the Patents,
-Redistribution and Trademark clauses below, we hereby grant you the right to
-use, copy, modify, create derivative works, publicly perform, publicly display
-and redistribute the Software for any Permitted Purpose identified below.
-
-### Permitted Purpose
-
-A Permitted Purpose is any purpose other than a Competing Use. A Competing Use
-means making the Software available to others in a commercial product or
-service that:
-
-1. substitutes for the Software;
-
-2. substitutes for any other product or service we offer using the Software
-   that exists as of the date we make the Software available; or
-
-3. offers the same or substantially similar functionality as the Software.
-
-Permitted Purposes specifically include using the Software:
-
-1. for your internal use and access;
-
-2. for non-commercial education;
-
-3. for non-commercial research; and
-
-4. in connection with professional services that you provide to a licensee
-   using the Software in accordance with these Terms and Conditions.
-
-### Patents
-
-To the extent your use for a Permitted Purpose would necessarily infringe our
-patents, the license grant above includes a license under our patents. If you
-make a claim against any party that the Software infringes or contributes to
-the infringement of any patent, then your patent license to the Software ends
-immediately.
-
-### Redistribution
-
-The Terms and Conditions apply to all copies, modifications and derivatives of
-the Software.
-
-If you redistribute any copies, modifications or derivatives of the Software,
-you must include a copy of or a link to these Terms and Conditions and not
-remove any copyright notices provided in or with the Software.
-
-### Disclaimer
-
-THE SOFTWARE IS PROVIDED "AS IS" AND WITHOUT WARRANTIES OF ANY KIND, EXPRESS OR
-IMPLIED, INCLUDING WITHOUT LIMITATION WARRANTIES OF FITNESS FOR A PARTICULAR
-PURPOSE, MERCHANTABILITY, TITLE OR NON-INFRINGEMENT.
-
-IN NO EVENT WILL WE HAVE ANY LIABILITY TO YOU ARISING OUT OF OR RELATED TO THE
-SOFTWARE, INCLUDING INDIRECT, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES,
-EVEN IF WE HAVE BEEN INFORMED OF THEIR POSSIBILITY IN ADVANCE.
-
-### Trademarks
-
-Except for displaying the License Details and identifying us as the origin of
-the Software, you have no right under these Terms and Conditions to use our
-trademarks, trade names, service marks or product names.
-
-## Grant of Future License
-
-We hereby irrevocably grant you an additional license to use the Software under
-the Apache License, Version 2.0 that is effective on the second anniversary of
-the date we make the Software available. On or after that date, you may use the
-Software under the Apache License, Version 2.0, in which case the following
-will apply:
-
-Licensed under the Apache License, Version 2.0 (the "License"); you may not use
-this file except in compliance with the License.
-
-You may obtain a copy of the License at
-
-http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software distributed
-under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR
-CONDITIONS OF ANY KIND, either express or implied. See the License for the
-specific language governing permissions and limitations under the License.
+# Functional Source License, Version 1.1, ALv2 Future License
+
+## Abbreviation
+
+FSL-1.1-ALv2
+
+## Notice
+
+Copyright 2026 Mohit Soni
+
+## Terms and Conditions
+
+### Licensor ("We")
+
+The party offering the Software under these Terms and Conditions.
+
+### The Software
+
+The "Software" is each version of the software that we make available under
+these Terms and Conditions, as indicated by our inclusion of these Terms and
+Conditions with the Software.
+
+### License Grant
+
+Subject to your compliance with this License Grant and the Patents,
+Redistribution and Trademark clauses below, we hereby grant you the right to
+use, copy, modify, create derivative works, publicly perform, publicly display
+and redistribute the Software for any Permitted Purpose identified below.
+
+### Permitted Purpose
+
+A Permitted Purpose is any purpose other than a Competing Use. A Competing Use
+means making the Software available to others in a commercial product or
+service that:
+
+1. substitutes for the Software;
+
+2. substitutes for any other product or service we offer using the Software
+   that exists as of the date we make the Software available; or
+
+3. offers the same or substantially similar functionality as the Software.
+
+Permitted Purposes specifically include using the Software:
+
+1. for your internal use and access;
+
+2. for non-commercial education;
+
+3. for non-commercial research; and
+
+4. in connection with professional services that you provide to a licensee
+   using the Software in accordance with these Terms and Conditions.
+
+### Patents
+
+To the extent your use for a Permitted Purpose would necessarily infringe our
+patents, the license grant above includes a license under our patents. If you
+make a claim against any party that the Software infringes or contributes to
+the infringement of any patent, then your patent license to the Software ends
+immediately.
+
+### Redistribution
+
+The Terms and Conditions apply to all copies, modifications and derivatives of
+the Software.
+
+If you redistribute any copies, modifications or derivatives of the Software,
+you must include a copy of or a link to these Terms and Conditions and not
+remove any copyright notices provided in or with the Software.
+
+### Disclaimer
+
+THE SOFTWARE IS PROVIDED "AS IS" AND WITHOUT WARRANTIES OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING WITHOUT LIMITATION WARRANTIES OF FITNESS FOR A PARTICULAR
+PURPOSE, MERCHANTABILITY, TITLE OR NON-INFRINGEMENT.
+
+IN NO EVENT WILL WE HAVE ANY LIABILITY TO YOU ARISING OUT OF OR RELATED TO THE
+SOFTWARE, INCLUDING INDIRECT, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES,
+EVEN IF WE HAVE BEEN INFORMED OF THEIR POSSIBILITY IN ADVANCE.
+
+### Trademarks
+
+Except for displaying the License Details and identifying us as the origin of
+the Software, you have no right under these Terms and Conditions to use our
+trademarks, trade names, service marks or product names.
+
+## Grant of Future License
+
+We hereby irrevocably grant you an additional license to use the Software under
+the Apache License, Version 2.0 that is effective on the second anniversary of
+the date we make the Software available. On or after that date, you may use the
+Software under the Apache License, Version 2.0, in which case the following
+will apply:
+
+Licensed under the Apache License, Version 2.0 (the "License"); you may not use
+this file except in compliance with the License.
+
+You may obtain a copy of the License at
+
+http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software distributed
+under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR
+CONDITIONS OF ANY KIND, either express or implied. See the License for the
+specific language governing permissions and limitations under the License.

package/README.md

@@ -136,9 +136,10 @@ the detail:
 - **Use the folders you already have.** Point TurboLLM at any directory of GGUFs — your
   existing LM Studio / Ollama / manual downloads — **no re-downloading.** It parses GGUF
   metadata (arch, params, quant, context, vision) for every file.
-- **Browse & download from Hugging Face**, in-app: search, see the file tree, pick a quant,
-  and download with **resume + SHA-256 verification**. Gated models (Llama, Gemma) work via
-  your own HF token, which **never leaves your machine**.
+- **Browse & download from Hugging Face**, in-app: a live, sortable list (trending / downloads
+  / likes / recently updated / newest) alongside a permanent detail pane — pick a quant, read
+  the **rendered model card**, and download with **resume + SHA-256 verification**. Gated
+  models (Llama, Gemma) work via your own HF token, which **never leaves your machine**.
 - **Import from any URL** — not just Hugging Face. Paste a direct `.gguf` link (model-author
   sites, mirrors, private servers); it disk-space-checks and downloads through the same manager.
 - **Quant recommendation per GPU** and a **VRAM-fit verdict** so you pick a quant that
@@ -319,7 +320,9 @@ the engine as a swappable component.
    support are hidden; ones it adds (e.g. low-bit KV cache, NextN) light up.
 
 No prebuilt for your OS? The **build-from-source guide** checks your toolchain (git / CMake /
-CUDA / MSVC), hands you the exact build commands, then drops you into the folder scan above.
+CUDA / a compiler — MSVC on Windows, gcc/clang on Linux), hands you the exact build commands
+(or a 1-click **"Build it for me"** on Windows and Linux), then drops you into the folder scan
+above.
 
 **Auto-provisioned default.** Don't want to fetch anything? On first run TurboLLM downloads
 the right upstream prebuilt for your GPU automatically — and a **backend picker** lets you

package/bin/turbollm.mjs

@@ -1,25 +1,25 @@
-#!/usr/bin/env node
-// Hand-written launcher shim — NOT bundled by tsup. It runs before the bundled
-// daemon (dist/cli.js) loads, so it can do two things that must happen before
-// node:sqlite is imported (the bundle hoists that import, emitting an experimental
-// warning that PowerShell renders as a scary red error block on every run):
-//   1. Guard the Node version with a friendly message (node:sqlite needs Node 22+).
-//   2. Register a 'warning' filter that swallows the node:sqlite experimental notice
-//      while still printing every other warning.
-// Then it hands off to the real CLI via dynamic import (same process; argv intact).
-const major = Number(process.versions.node.split('.')[0])
-if (major < 22) {
-  process.stderr.write(
-    `TurboLLM requires Node.js 22 or newer.\n` +
-      `You are running Node.js ${process.versions.node}.\n` +
-      `Please upgrade: https://nodejs.org\n`,
-  )
-  process.exit(1)
-}
-
-process.on('warning', (w) => {
-  if (w.name === 'ExperimentalWarning' && /SQLite/i.test(w.message)) return
-  process.stderr.write(`${w.stack ?? `${w.name}: ${w.message}`}\n`)
-})
-
-await import('../dist/cli.js')
+#!/usr/bin/env node
+// Hand-written launcher shim — NOT bundled by tsup. It runs before the bundled
+// daemon (dist/cli.js) loads, so it can do two things that must happen before
+// node:sqlite is imported (the bundle hoists that import, emitting an experimental
+// warning that PowerShell renders as a scary red error block on every run):
+//   1. Guard the Node version with a friendly message (node:sqlite needs Node 22+).
+//   2. Register a 'warning' filter that swallows the node:sqlite experimental notice
+//      while still printing every other warning.
+// Then it hands off to the real CLI via dynamic import (same process; argv intact).
+const major = Number(process.versions.node.split('.')[0])
+if (major < 22) {
+  process.stderr.write(
+    `TurboLLM requires Node.js 22 or newer.\n` +
+      `You are running Node.js ${process.versions.node}.\n` +
+      `Please upgrade: https://nodejs.org\n`,
+  )
+  process.exit(1)
+}
+
+process.on('warning', (w) => {
+  if (w.name === 'ExperimentalWarning' && /SQLite/i.test(w.message)) return
+  process.stderr.write(`${w.stack ?? `${w.name}: ${w.message}`}\n`)
+})
+
+await import('../dist/cli.js')

package/dist/cli.js

@@ -1427,7 +1427,12 @@ function readinessTimeoutMs(kind) {
   return kind === "vllm" || kind === "sglang" ? 6e5 : 12e4;
 }
 function pyEngineEnv(kind, dataDir, binPath) {
-  if (kind !== "mlx" && kind !== "vllm" && kind !== "sglang") return void 0;
+  if (kind !== "mlx" && kind !== "vllm" && kind !== "sglang") {
+    if (process.platform === "win32") return void 0;
+    const dir = dirname2(binPath);
+    const existing = process.env.LD_LIBRARY_PATH;
+    return { ...process.env, LD_LIBRARY_PATH: existing ? `${dir}:${existing}` : dir };
+  }
   const hfHome = join9(dataDir, "hf-cache");
   const hubCache = join9(hfHome, "hub");
   mkdirSync6(hubCache, { recursive: true });
@@ -4148,6 +4153,28 @@ var HfError = class extends Error {
   }
   code;
 };
+var SORT_PARAM = {
+  trending: "trendingScore",
+  downloads: "downloads",
+  likes: "likes",
+  modified: "lastModified",
+  created: "createdAt"
+};
+function libraryFilterFor(engineKind) {
+  if (engineKind === "mlx") return "filter=mlx&";
+  if (engineKind === "vllm") return "";
+  return "filter=gguf&";
+}
+function toSearchItem(m) {
+  return {
+    repo: m.id ?? m.modelId ?? "",
+    downloads: m.downloads ?? 0,
+    likes: m.likes ?? 0,
+    updatedAt: m.lastModified ?? m.createdAt ?? "",
+    gated: m.gated === true || m.gated === "auto" || m.gated === "manual",
+    tags: Array.isArray(m.tags) ? m.tags : []
+  };
+}
 var SPLIT_RE2 = /^(.*)-(\d{5})-of-(\d{5})\.gguf$/i;
 var HfClient = class {
   constructor(tokenFn, version2) {
@@ -4157,24 +4184,26 @@ var HfClient = class {
   tokenFn;
   version;
   cache = /* @__PURE__ */ new Map();
-  /** Search repos (spec 10 §2). Returns up to 30 rows sorted by downloads.
-   *  Format filter adapts to the active engine kind:
-   *  - llama-server / TurboQuant → filter=gguf
-   *  - mlx                       → filter=mlx (HF library tag)
-   *  - vllm                      → no format filter (searches all HF repos) */
-  async searchModels(query, engineKind) {
+  /** Search repos by text (spec 10 §2). Returns up to 30 rows. `sort` defaults to HF's own
+   *  relevance ranking ('best-match' — omits `sort=` entirely, a genuinely different order
+   *  from `sort=downloads`, not just an alias for it). Format filter adapts to the active
+   *  engine kind (never hardcoded to GGUF) — see {@link libraryFilterFor}. */
+  async searchModels(query, engineKind, sort = "best-match") {
     const q = query.trim();
-    const formatFilter = engineKind === "mlx" ? "&filter=mlx" : engineKind === "vllm" ? "" : "&filter=gguf";
-    const url = `${BASE}/api/models?search=${encodeURIComponent(q)}${formatFilter}&sort=downloads&direction=-1&limit=30&full=false`;
+    const sortParam = sort === "best-match" ? "" : `sort=${SORT_PARAM[sort]}&direction=-1&`;
+    const url = `${BASE}/api/models?search=${encodeURIComponent(q)}&${libraryFilterFor(engineKind)}${sortParam}limit=30&full=false`;
     const raw = await this.getJson(url);
-    return raw.map((m) => ({
-      repo: m.id ?? m.modelId ?? "",
-      downloads: m.downloads ?? 0,
-      likes: m.likes ?? 0,
-      updatedAt: m.lastModified ?? m.createdAt ?? "",
-      gated: m.gated === true || m.gated === "auto" || m.gated === "manual",
-      tags: Array.isArray(m.tags) ? m.tags : []
-    }));
+    return raw.map(toSearchItem);
+  }
+  /** Browse repos with no search term (spec 10 §7 rewrite) — the live equivalent of
+   *  https://huggingface.co/models?library=gguf&sort=trending, replacing what used to be a
+   *  hardcoded "Featured" list. 'best-match' has no meaning without a query, so it falls
+   *  back to 'trending'. Format filter adapts to the active engine kind, same as search. */
+  async browseModels(sort, engineKind, limit = 30) {
+    const effective = sort === "best-match" ? "trending" : sort;
+    const url = `${BASE}/api/models?${libraryFilterFor(engineKind)}sort=${SORT_PARAM[effective]}&direction=-1&limit=${limit}&full=false`;
+    const raw = await this.getJson(url);
+    return raw.map(toSearchItem);
   }
   /** Repo detail (spec 10 §3): card data + the GGUF file tree, with split parts
    *  grouped and quant/mmproj detected per file. */
@@ -9123,6 +9152,36 @@ var ALL = [
     // install endpoint of its own.
     installEndpoint: ""
   },
+  {
+    // Upstream publishes NO Linux CUDA prebuilt (download.ts availableBackends() — Linux
+    // NVIDIA users get Vulkan from the main llama.cpp entry above). Now that the guided
+    // build (ADR-100/build-runner.ts) runs on Linux too, offer official llama.cpp + CUDA
+    // there as its own "build from source" card — same guided-build flow as ik_llama.cpp/
+    // TurboQuant below, just pointed at the official repo. Windows/macOS never see this
+    // (platforms:['linux']): Windows already gets a CUDA prebuilt, and macOS has no CUDA.
+    id: "llama.cpp-cuda-linux",
+    name: "llama.cpp (CUDA)",
+    kind: "llama-server",
+    description: "Official llama.cpp compiled with CUDA for NVIDIA GPUs on Linux \u2014 no prebuilt is published upstream, so this uses the guided in-app build.",
+    provision: "github-release",
+    homepage: "https://github.com/ggml-org/llama.cpp",
+    repo: "ggml-org/llama.cpp",
+    platforms: ["linux"],
+    support: "experimental",
+    installEndpoint: "",
+    note: "Builds the official repo with -DGGML_CUDA=ON and bundles its CUDA runtime libraries next to the binary, so it runs standalone. Manages like any other engine afterward (rebuild on a newer commit, disable, delete).",
+    variants: [
+      {
+        id: "llama.cpp-cuda-source",
+        label: "CUDA (NVIDIA) \u2014 build from source",
+        repo: "ggml-org/llama.cpp",
+        requires: { platform: ["linux"], gpuVendor: ["nvidia"] },
+        stability: "experimental",
+        speed: "fast",
+        hasPrebuilt: false
+      }
+    ]
+  },
   {
     id: "vllm",
     name: "vLLM",
@@ -9377,7 +9436,8 @@ var INSTALL_URLS = {
   cmake: "https://cmake.org/download/",
   cuda: "https://developer.nvidia.com/cuda-downloads",
   // The "Desktop development with C++" workload from the VS Build Tools installer.
-  msvc: "https://visualstudio.microsoft.com/downloads/"
+  msvc: "https://visualstudio.microsoft.com/downloads/",
+  gcc: "https://gcc.gnu.org/install/"
 };
 async function runVersion(cmd, args, env) {
   const { stdout, stderr } = await execFileP5(cmd, args, {
@@ -9400,6 +9460,10 @@ function parseNvccVersion(out) {
   const m = out.match(/release\s+([\d.]+)/i);
   return m ? m[1] : "";
 }
+function parseCompilerVersion(out) {
+  const m = out.match(/(?:g\+\+|gcc|clang(?:\+\+)?)[^\d]*(\d+(?:\.\d+)*)/i);
+  return m ? m[1] : "";
+}
 async function checkGit(env) {
   let found = false;
   let version2;
@@ -9461,13 +9525,30 @@ async function checkMsvc(env) {
     installUrl: INSTALL_URLS.msvc
   };
 }
+async function checkGcc(env) {
+  let found = false;
+  let version2;
+  try {
+    version2 = parseCompilerVersion(await runVersion("g++", ["--version"], env)) || void 0;
+    found = true;
+  } catch {
+    try {
+      version2 = parseCompilerVersion(await runVersion("clang++", ["--version"], env)) || void 0;
+      found = true;
+    } catch {
+      found = false;
+    }
+  }
+  return { id: "gcc", name: "C++ compiler (g++/clang++)", found, version: version2, installUrl: INSTALL_URLS.gcc };
+}
 async function checkBuildPrereqs(toolchainDirs = []) {
-  if (process.platform !== "win32") {
-    return { supported: false, tools: [] };
+  const platform = process.platform;
+  if (platform !== "win32" && platform !== "linux") {
+    return { supported: false, os: "other", tools: [] };
   }
   const env = buildEnv(toolchainDirs);
-  const tools = await Promise.all([checkGit(env), checkCmake(env), checkCuda(env), checkMsvc(env)]);
-  return { supported: true, tools };
+  const tools = platform === "win32" ? await Promise.all([checkGit(env), checkCmake(env), checkCuda(env), checkMsvc(env)]) : await Promise.all([checkGit(env), checkCmake(env), checkCuda(env), checkGcc(env)]);
+  return { supported: true, os: platform === "win32" ? "windows" : "linux", tools };
 }
 
 // src/engines/build-runner.ts
@@ -9497,8 +9578,9 @@ function sourceBuildDirOf(binPath, enginesRoot) {
   if (!m) return null;
   return join19(enginesRoot, "build", m[1]);
 }
-function pickGenerator(hasNinja) {
-  return hasNinja ? "Ninja" : "NMake Makefiles";
+function pickGenerator(hasNinja, isWindows) {
+  if (hasNinja) return "Ninja";
+  return isWindows ? "NMake Makefiles" : "Unix Makefiles";
 }
 var CMAKE_CONFIGURE_ARGS = ["-DGGML_CUDA=ON", "-DCMAKE_BUILD_TYPE=Release"];
 function vcvarsBatch(vcvars, cmakeArgs) {
@@ -9573,6 +9655,56 @@ function copyCudaRuntimeDlls(env, destDir, log) {
   if (copied > 0) log(`Bundled ${copied} CUDA runtime DLL(s) next to the binary so the engine is self-contained.`);
   return copied;
 }
+var CUDA_RUNTIME_SO_PREFIXES = ["libcudart.so", "libcublas.so", "libcublasLt.so", "libnvrtc.so", "libnvrtc-builtins.so", "libnvJitLink.so"];
+function cudaSoSourceDirs(env) {
+  const key = Object.keys(env).find((k) => k.toLowerCase() === "path") ?? "PATH";
+  const pathDirs = (env[key] ?? "").split(delimiter2).filter(Boolean);
+  const nvccDir = pathDirs.find((dir) => {
+    try {
+      return existsSync17(join19(dir, "nvcc"));
+    } catch {
+      return false;
+    }
+  });
+  if (!nvccDir) return [];
+  const toolkitRoot = dirname6(nvccDir);
+  return [join19(toolkitRoot, "lib64"), join19(toolkitRoot, "lib"), join19(toolkitRoot, "targets", "x86_64-linux", "lib")].filter((d) => {
+    try {
+      return existsSync17(d);
+    } catch {
+      return false;
+    }
+  });
+}
+function copyCudaRuntimeLibs(env, destDir, log) {
+  const sources = cudaSoSourceDirs(env);
+  if (sources.length === 0) {
+    log("Note: could not find the CUDA toolkit shared libraries to bundle \u2014 the engine may need LD_LIBRARY_PATH set to run.");
+    return 0;
+  }
+  const seen = /* @__PURE__ */ new Set();
+  let copied = 0;
+  for (const dir of sources) {
+    let names;
+    try {
+      names = readdirSync5(dir);
+    } catch {
+      continue;
+    }
+    for (const name of names) {
+      if (seen.has(name)) continue;
+      if (!CUDA_RUNTIME_SO_PREFIXES.some((p) => name.startsWith(p))) continue;
+      try {
+        copyFileSync(join19(dir, name), join19(destDir, name));
+        seen.add(name);
+        copied++;
+      } catch {
+      }
+    }
+  }
+  if (copied > 0) log(`Bundled ${copied} CUDA runtime librar${copied === 1 ? "y" : "ies"} next to the binary so the engine is self-contained.`);
+  return copied;
+}
 async function findVcvarsall() {
   try {
     const programFilesX86 = process.env["ProgramFiles(x86)"] || "C:\\Program Files (x86)";
@@ -9691,10 +9823,11 @@ function neutralizeGratuitousAsm(srcDir, log) {
   }
 }
 async function runBuild(req, hooks, signal) {
+  const isWindows = process.platform === "win32";
   const env = { ...buildEnv(req.toolchainDirs), GIT_TERMINAL_PROMPT: "0", GCM_INTERACTIVE: "never" };
   hooks.phase("preparing");
   const prereqs = await checkBuildPrereqs(req.toolchainDirs);
-  if (!prereqs.supported) throw new Error("In-app build is currently Windows + CUDA only.");
+  if (!prereqs.supported) throw new Error("In-app build is currently Windows or Linux (with CUDA) only.");
   const missing = prereqs.tools.filter((t) => (t.id === "git" || t.id === "cmake" || t.id === "cuda") && !t.found);
   if (missing.length > 0) {
     const names = missing.map((t) => t.name).join(", ");
@@ -9702,10 +9835,17 @@ async function runBuild(req, hooks, signal) {
       `Missing build prerequisite(s): ${names}. Install them, or \u2014 if they live in a conda env / custom path \u2014 add that folder under Build environment so TurboLLM can find them.`
     );
   }
-  const vcvars = await findVcvarsall();
-  if (!vcvars) {
+  let vcvars = null;
+  if (isWindows) {
+    vcvars = await findVcvarsall();
+    if (!vcvars) {
+      throw new Error(
+        'Could not locate the Visual Studio C++ build environment (vcvarsall.bat). Install the "Desktop development with C++" workload from the Visual Studio Build Tools.'
+      );
+    }
+  } else if (!prereqs.tools.some((t) => t.id === "gcc" && t.found)) {
     throw new Error(
-      'Could not locate the Visual Studio C++ build environment (vcvarsall.bat). Install the "Desktop development with C++" workload from the Visual Studio Build Tools.'
+      "Could not find a C++ compiler. Install one (e.g. `sudo apt install build-essential` on Debian/Ubuntu) and retry."
     );
   }
   const buildRoot = join19(req.enginesRoot, "build", buildDirName(req.repoUrl, req.branch));
@@ -9721,24 +9861,35 @@ async function runBuild(req, hooks, signal) {
   const commit = (await runStep("git", ["-C", srcDir, "rev-parse", "HEAD"], { env, signal, onLine: () => {
   } })).trim();
   neutralizeGratuitousAsm(srcDir, hooks.log);
-  const generator = pickGenerator(onPath(env, "ninja.exe"));
+  const generator = pickGenerator(onPath(env, isWindows ? "ninja.exe" : "ninja"), isWindows);
   hooks.log(
-    generator === "Ninja" ? "Using the Ninja generator (drives nvcc directly \u2014 no Visual Studio CUDA integration needed)." : "Ninja not found on PATH \u2014 using the NMake generator (works, but single-threaded and slower; add a folder containing ninja.exe under Build environment for much faster builds)."
+    generator === "Ninja" ? "Using the Ninja generator (drives nvcc directly \u2014 no Visual Studio CUDA integration needed)." : isWindows ? "Ninja not found on PATH \u2014 using the NMake generator (works, but single-threaded and slower; add a folder containing ninja.exe under Build environment for much faster builds)." : "Ninja not found on PATH \u2014 using Unix Makefiles (works, but single-threaded and slower; add a folder containing ninja under Build environment for much faster builds)."
   );
   hooks.phase("configuring");
-  const configureBat = join19(buildRoot, "_tllm_configure.bat");
-  writeFileSync8(configureBat, vcvarsBatch(vcvars, ["-G", generator, "-B", buildSubdir, "-S", srcDir, ...CMAKE_CONFIGURE_ARGS]));
-  await runStep("cmd.exe", ["/c", configureBat], { cwd: buildRoot, env, signal, onLine: hooks.log });
+  const configureArgs = ["-G", generator, "-B", buildSubdir, "-S", srcDir, ...CMAKE_CONFIGURE_ARGS];
+  if (isWindows) {
+    const configureBat = join19(buildRoot, "_tllm_configure.bat");
+    writeFileSync8(configureBat, vcvarsBatch(vcvars, configureArgs));
+    await runStep("cmd.exe", ["/c", configureBat], { cwd: buildRoot, env, signal, onLine: hooks.log });
+  } else {
+    await runStep("cmake", configureArgs, { cwd: buildRoot, env, signal, onLine: hooks.log });
+  }
   hooks.phase("compiling");
-  const compileBat = join19(buildRoot, "_tllm_build.bat");
-  writeFileSync8(compileBat, vcvarsBatch(vcvars, ["--build", buildSubdir, "-j", "--target", "llama-server"]));
-  await runStep("cmd.exe", ["/c", compileBat], { cwd: buildRoot, env, signal, onLine: hooks.log });
+  const compileArgs = ["--build", buildSubdir, "-j", "--target", "llama-server"];
+  if (isWindows) {
+    const compileBat = join19(buildRoot, "_tllm_build.bat");
+    writeFileSync8(compileBat, vcvarsBatch(vcvars, compileArgs));
+    await runStep("cmd.exe", ["/c", compileBat], { cwd: buildRoot, env, signal, onLine: hooks.log });
+  } else {
+    await runStep("cmake", compileArgs, { cwd: buildRoot, env, signal, onLine: hooks.log });
+  }
   hooks.phase("registering");
   const binPath = resolveServerBinary(buildSubdir) ?? resolveServerBinary(buildRoot);
   if (!binPath) {
     throw new Error("Build finished but no llama-server binary was found in the output \u2014 see the log above.");
   }
-  copyCudaRuntimeDlls(env, dirname6(binPath), hooks.log);
+  if (isWindows) copyCudaRuntimeDlls(env, dirname6(binPath), hooks.log);
+  else copyCudaRuntimeLibs(env, dirname6(binPath), hooks.log);
   return { binPath, commit, buildRoot };
 }
 
@@ -10353,8 +10504,8 @@ function registerApi(app2, d) {
   app2.post("/api/v1/build/run", async (c) => {
     if (!isLocalRequest(c, d))
       return err2(c, 403, "forbidden", "Building an engine is only available on the machine running TurboLLM.");
-    if (process.platform !== "win32")
-      return err2(c, 409, "unsupported_platform", "In-app build is currently Windows + CUDA only.");
+    if (process.platform !== "win32" && process.platform !== "linux")
+      return err2(c, 409, "unsupported_platform", "In-app build is currently Windows or Linux (with CUDA) only.");
     const b = await body2(c);
     const repoUrl = (b.repoUrl ?? "").trim();
     if (!/^https?:\/\//i.test(repoUrl))
@@ -11270,11 +11421,13 @@ function registerApi(app2, d) {
     });
     return c.json(modelDirsPayload(d));
   });
+  const SORT_OPTIONS = ["best-match", "trending", "downloads", "likes", "modified", "created"];
   app2.get("/api/v1/hf/search", async (c) => {
     const q = (c.req.query("q") ?? "").trim();
-    if (!q) return c.json({ results: [] });
+    const rawSort = c.req.query("sort") ?? "";
+    const sort = SORT_OPTIONS.includes(rawSort) ? rawSort : "best-match";
     try {
-      const results = await d.hf.searchModels(q, d.registry.active()?.kind);
+      const results = q ? await d.hf.searchModels(q, d.registry.active()?.kind, sort) : await d.hf.browseModels(sort, d.registry.active()?.kind);
       const withLocal = results.map((r) => ({ ...r, localCount: localCountFor(d, r.repo) }));
       return c.json({ results: withLocal });
     } catch (e) {