feat: integrate Google semantic search and branded desktop launcher

tools/embeddinggemma2/build_runtime.py

@@ -0,0 +1,58 @@
+"""Build a pinned isolated native worker; development VM builds use --backend cpu."""
+import argparse
+import os
+from pathlib import Path
+import shutil
+import subprocess
+import sys
+
+ROOT = Path(__file__).resolve().parents[2]
+TAG = "b11461"
+COMMIT = "4d756bc72bf00a4aacf410ae15a2d315f3db400d"
+
+
+def run(arguments, cwd=None):
+    subprocess.run([str(value) for value in arguments], cwd=cwd, check=True)
+
+
+def main():
+    parser = argparse.ArgumentParser(description="Build Searchi's pinned Google embedding worker")
+    parser.add_argument("--backend", choices=["cpu", "auto", "metal", "vulkan", "cuda"], default="cpu")
+    parser.add_argument("--jobs", type=int, default=2)
+    args = parser.parse_args()
+    if not 1 <= args.jobs <= 16:
+        parser.error("jobs must be 1–16")
+    source = ROOT / ".searchi/references/llama.cpp"
+    build = ROOT / ".searchi/runtime" / args.backend
+    if not source.exists():
+        source.parent.mkdir(parents=True, exist_ok=True)
+        run(["git", "clone", "--depth", "1", "--branch", TAG,
+             "https://github.com/ggml-org/llama.cpp.git", source])
+    actual = subprocess.check_output(["git", "rev-parse", "HEAD"], cwd=source, text=True).strip()
+    if actual != COMMIT:
+        raise RuntimeError("Runtime reference checkout differs from pinned commit; refusing to reset it")
+    patch = Path(__file__).with_name("no-logits.patch")
+    check = subprocess.run(["git", "apply", "--check", str(patch)], cwd=source, capture_output=True)
+    if check.returncode == 0:
+        run(["git", "apply", patch], cwd=source)
+    else:
+        reverse = subprocess.run(["git", "apply", "--reverse", "--check", str(patch)], cwd=source, capture_output=True)
+        if reverse.returncode:
+            raise RuntimeError("Runtime memory patch cannot be applied or verified")
+    cmake_exe = shutil.which("cmake")
+    if not cmake_exe:
+        try:
+            import cmake
+            cmake_exe = str(Path(cmake.CMAKE_BIN_DIR) / ("cmake.exe" if os.name == "nt" else "cmake"))
+        except ImportError as error:
+            raise RuntimeError("Install CMake first (python -m pip install cmake)") from error
+    worker_source = ROOT / "crates/searchi-semantic/native"
+    run([cmake_exe, "-S", worker_source, "-B", build,
+         "-DCMAKE_BUILD_TYPE=Release", f"-DLLAMA_CPP_DIR={source}",
+         f"-DSEARCHI_ACCELERATION={args.backend}"])
+    run([cmake_exe, "--build", build, "--config", "Release", "--parallel", args.jobs])
+    print(f"Runtime build complete: {build}")
+
+
+if __name__ == "__main__":
+    main()

tools/embeddinggemma2/download.py

@@ -0,0 +1,54 @@
+"""Explicitly provision the pinned Google model; search never calls this script."""
+import argparse
+import hashlib
+import json
+import os
+from pathlib import Path
+import urllib.request
+
+
+def verified(path, spec):
+    if not path.is_file() or path.stat().st_size != spec["bytes"]:
+        return False
+    with path.open("rb") as source:
+        return hashlib.file_digest(source, "sha256").hexdigest() == spec["sha256"]
+
+
+def main():
+    parser = argparse.ArgumentParser(description="Download the verified Google EmbeddingGemma 2 Q8 model (865 MB)")
+    parser.add_argument("--model-dir", type=Path, default=Path(".searchi/models/embeddinggemma2"))
+    parser.add_argument("--text-only", action="store_true", help="Download only the 310 MB text backbone")
+    args = parser.parse_args()
+    manifest = json.loads(Path(__file__).with_name("model-manifest.json").read_text())
+    for spec in manifest["files"][:1] if args.text_only else manifest["files"]:
+        destination = args.model_dir / spec["path"]
+        if verified(destination, spec):
+            print(f"Verified {spec['path']}", flush=True)
+            continue
+        destination.parent.mkdir(parents=True, exist_ok=True)
+        temporary = destination.with_suffix(".part")
+        url = f"https://huggingface.co/{manifest['export_repository']}/resolve/{manifest['revision']}/{spec['path']}"
+        try:
+            request = urllib.request.Request(url, headers={"User-Agent": "Searchi/0.1"})
+            print(f"Downloading {spec['path']} ({spec['bytes']} bytes)", flush=True)
+            with urllib.request.urlopen(request, timeout=60) as response, temporary.open("wb") as output:
+                total = 0
+                last_report = 0
+                while chunk := response.read(1024 * 1024):
+                    total += len(chunk)
+                    if total > spec["bytes"]:
+                        raise ValueError("Model download exceeds pinned size")
+                    output.write(chunk)
+                    if total - last_report >= 50 * 1024 * 1024:
+                        print(f"Downloaded {total}/{spec['bytes']} bytes", flush=True)
+                        last_report = total
+            if not verified(temporary, spec):
+                raise ValueError("Model size or SHA256 mismatch")
+            os.replace(temporary, destination)
+            print(f"Verified SHA256: {spec['sha256']}", flush=True)
+        finally:
+            temporary.unlink(missing_ok=True)
+
+
+if __name__ == "__main__":
+    main()

tools/embeddinggemma2/model-manifest.json

@@ -0,0 +1,15 @@
+{
+  "model_id": "google/embeddinggemma-2",
+  "export_repository": "ggml-org/embeddinggemma-2-GGUF",
+  "revision": "bfcd298762cc34d0357ece5ebdd31791a3a374d8",
+  "dimensions": 768,
+  "quantization": "Q8_0",
+  "license": "Apache-2.0",
+  "runtime_tag": "b11461",
+  "query_prefix": "task: search result | query: ",
+  "document_template": "title: {title} | text: {text}",
+  "files": [
+    {"path":"embeddinggemma-2-Q8_0.gguf","bytes":309855456,"sha256":"2188ac1deca4b77dffefd603c2776a9d76d9d74ec01841392982ebb840b09135"},
+    {"path":"mmproj-embeddinggemma-2-Q8_0.gguf","bytes":554821024,"sha256":"c4a8a52691ecef40618438928bdf9e68379b854e24166f292592353db0aab64f"}
+  ]
+}

tools/embeddinggemma2/no-logits.patch

@@ -0,0 +1,12 @@
+diff --git a/src/llama-context.cpp b/src/llama-context.cpp
+--- a/src/llama-context.cpp
++++ b/src/llama-context.cpp
+@@ -2133,7 +2133,7 @@ uint32_t llama_context::output_reserve(int32_t n_outputs) {
+     const auto n_embd     = hparams.n_embd;
+     const auto n_embd_out = hparams.n_embd_out();
+ 
+-    bool has_logits     = true;
++    bool has_logits     = !(cparams.embeddings && !cparams.causal_attn);
+     bool has_embd       = cparams.embeddings;
+     bool has_embd_nextn = cparams.embeddings_nextn;
+ 

tools/embeddinggemma2/stage_runtime.py

@@ -0,0 +1,45 @@
+"""Stage the native worker for installers; optional offline model bundle."""
+import argparse
+import hashlib
+import json
+import os
+from pathlib import Path
+import shutil
+
+ROOT = Path(__file__).resolve().parents[2]
+
+
+def main():
+    parser = argparse.ArgumentParser(description="Stage Searchi runtime resources")
+    parser.add_argument("--backend", choices=["cpu", "auto", "metal", "vulkan", "cuda"], default="cpu")
+    parser.add_argument("--bundle-model", action="store_true")
+    args = parser.parse_args()
+    name = "searchi-gemma-worker.exe" if os.name == "nt" else "searchi-gemma-worker"
+    source = ROOT / ".searchi/runtime" / args.backend
+    if os.name == "nt":
+        source /= "Release"
+    source /= name
+    if not source.is_file():
+        raise RuntimeError("Build the pinned native worker before staging")
+    destination = ROOT / "apps/desktop/src-tauri/resources"
+    destination.mkdir(parents=True, exist_ok=True)
+    shutil.copy2(source, destination / name)
+    # Include the upstream runtime license with the statically linked helper.
+    shutil.copy2(ROOT / ".searchi/references/llama.cpp/LICENSE", destination / "LLAMA-LICENSE.txt")
+    if args.bundle_model:
+        manifest = json.loads(Path(__file__).with_name("model-manifest.json").read_text())
+        model_destination = destination / "embeddinggemma2"
+        model_destination.mkdir(exist_ok=True)
+        for spec in manifest["files"]:
+            artifact = ROOT / ".searchi/models/embeddinggemma2" / spec["path"]
+            if artifact.stat().st_size != spec["bytes"]:
+                raise RuntimeError("Model size mismatch")
+            with artifact.open("rb") as stream:
+                if hashlib.file_digest(stream, "sha256").hexdigest() != spec["sha256"]:
+                    raise RuntimeError("Model checksum mismatch")
+            shutil.copy2(artifact, model_destination / spec["path"])
+    print(f"Staged native worker: {destination / name}")
+
+
+if __name__ == "__main__":
+    main()