Spotlight for Windows: semantic search across your PC, plus local LLA agents that fix files
Python
10
0 commits
updated Oct 14, 2026
A Spotlight-style launcher for Windows. Press Win+Y to search apps, files, document contents
and browser history by meaning, not just by name. It can also fix files for you with LLA, a
local mixture-of-experts model: type fix app.py it crashes on start and press Tab.
Everything runs on your PC.
Download LSearch-Setup.exe from Releases and run it. It installs per user,
with no admin prompt, and needs neither Python nor .NET. The setup wizard asks which folders to
index, whether to include browser history and, on PCs with 16 GB of RAM or more, whether to
download LLA (4.1 GB, from Hugging Face).
| Key | Action |
|---|---|
| ↑ / ↓ | Move selection |
| ↵ | Open (copies the result for calculator rows) |
| Ctrl+↵ | Show in Explorer |
| Tab | Run an LLA task, e.g. fix "C:\code\app.py" it crashes on start |
| Esc | Clear, then hide |
| Search only | With LLA | |
|---|---|---|
| OS | Windows 10 1809+ (64-bit) | same |
| RAM | 8 GB | 16 GB minimum, 32 GB recommended |
| CPU | 4 threads | 4 cores with AVX2; 8+ cores recommended |
| Disk | ~400 MB plus the index | +4.1 GB for the model |
LLA writes out its reasoning before each step, so it is slow on small CPUs (about 2 tokens/s on 4 threads).
BAAI/bge-small-en-v1.5 via fastembed/ONNX) and also indexed for keyword search..lsearch.bak backup
is kept, and shell commands are off unless agent.allow_commands is true.Config lives at %APPDATA%\LSearch\config.json (tray → Open config).
LLA runs through llama.cpp's llama-server on port 1114. To use another model, point llm at any
OpenAI-compatible server (llama.cpp, Ollama, LM Studio, vLLM):
"llm": { "enabled": true, "display_name": "MyModel",
"base_url": "http://127.0.0.1:11434/v1", "model": "my-model", "model_path": "" }
Or set model_path to a .gguf file and LSearch will start llama-server for it on demand.
Requires Python 3.10+, the .NET 10 SDK and, for the installer, Inno Setup 6.
pip install -r requirements.txt pyinstaller
powershell -File packaging\build.ps1 # -> dist\LSearch\ and dist\LSearch-Setup.exe
For development, cd winui && dotnet build -c Release and run LSearch.exe from bin\. It
starts the Python backend (pythonw -m lsearch.server, localhost:52123) from source. Debug flags:
--show, --pin, --query <text>, --setup.
Apache License 2.0 — see LICENSE.
Spotlight for Windows: semantic search across your PC, plus local LLA agents that fix files
Python
10
0 commits
updated Oct 14, 2026
A Spotlight-style launcher for Windows. Press Win+Y to search apps, files, document contents
and browser history by meaning, not just by name. It can also fix files for you with LLA, a
local mixture-of-experts model: type fix app.py it crashes on start and press Tab.
Everything runs on your PC.
Download LSearch-Setup.exe from Releases and run it. It installs per user,
with no admin prompt, and needs neither Python nor .NET. The setup wizard asks which folders to
index, whether to include browser history and, on PCs with 16 GB of RAM or more, whether to
download LLA (4.1 GB, from Hugging Face).
| Key | Action |
|---|---|
| ↑ / ↓ | Move selection |
| ↵ | Open (copies the result for calculator rows) |
| Ctrl+↵ | Show in Explorer |
| Tab | Run an LLA task, e.g. fix "C:\code\app.py" it crashes on start |
| Esc | Clear, then hide |
| Search only | With LLA | |
|---|---|---|
| OS | Windows 10 1809+ (64-bit) | same |
| RAM | 8 GB | 16 GB minimum, 32 GB recommended |
| CPU | 4 threads | 4 cores with AVX2; 8+ cores recommended |
| Disk | ~400 MB plus the index | +4.1 GB for the model |
LLA writes out its reasoning before each step, so it is slow on small CPUs (about 2 tokens/s on 4 threads).
BAAI/bge-small-en-v1.5 via fastembed/ONNX) and also indexed for keyword search..lsearch.bak backup
is kept, and shell commands are off unless agent.allow_commands is true.Config lives at %APPDATA%\LSearch\config.json (tray → Open config).
LLA runs through llama.cpp's llama-server on port 1114. To use another model, point llm at any
OpenAI-compatible server (llama.cpp, Ollama, LM Studio, vLLM):
"llm": { "enabled": true, "display_name": "MyModel",
"base_url": "http://127.0.0.1:11434/v1", "model": "my-model", "model_path": "" }
Or set model_path to a .gguf file and LSearch will start llama-server for it on demand.
Requires Python 3.10+, the .NET 10 SDK and, for the installer, Inno Setup 6.
pip install -r requirements.txt pyinstaller
powershell -File packaging\build.ps1 # -> dist\LSearch\ and dist\LSearch-Setup.exe
For development, cd winui && dotnet build -c Release and run LSearch.exe from bin\. It
starts the Python backend (pythonw -m lsearch.server, localhost:52123) from source. Debug flags:
--show, --pin, --query <text>, --setup.
Apache License 2.0 — see LICENSE.