Lens for GNOME

Lens for GNOME is an AI powered local-first desktop search engine.

It serves as a fast launcher and smart search solution. It indexes local apps and files, extracts searchable text, watches folders for changes, runs local embeddings, supports OCR, syncs mail, and can use local GGUF language models through llama.cpp (such as those from HuggingFace).

The goal is simple: fast, private desktop search that helps you find information inside your files without sending your documents to a remote service. Lens for GNOME is 100% privacy-first and local-first.

The GNOME extension used to interface with this backend-service is published on the GNOME store here: https://extensions.gnome.org/extension/10337/lens/

What Lens Does

Lens turns content into a powerful searchable index on your computer.

It can:

• Index configured folders recursively.
• Watch folders for changes using inotify.
• Store searchable content in SQLite.
• Search exact text using SQLite FTS5.
• Search by meaning using local embeddings.
• Merge keyword and semantic results with Reciprocal Rank Fusion.
• Extract text from documents, images, PDFs, office documents, mail, and videos.
• Run local LLM-powered queries when enabled.
• Keep indexed data, embeddings, metadata, and models on your machine.
• Fast preview videos or images on hover, multi-monitor capable.

Screenshots

Filtering based on your data using AI

Questions about your data, powered by local AI - even if it's from i.e. screenshots/PDFs (OCR capable)

Filebrowsing and fast video preview.

Support for different AI models, which can be turned off

Different themes available:

Why Lens Exists

Most desktop search tools are good at finding file names.

Lens is built for finding information inside your files.

You can search for exact words, ask more natural questions, or find related content even when the document does not use the same wording as your query.

Example searches:

meeting notes about renewable energy pricing

documents mentioning my lease termination date

emails with an IBAN from last year

invoices less than 300 usd

source files related to the indexing pipeline

Features

Hybrid Search

Lens combines traditional full-text search with semantic search.

Keyword search is fast and precise. Semantic search helps when you know what you are looking for, but not the exact words used in the document.

Lens uses:

Capability	Technology
Exact search	SQLite FTS5
Semantic search	Local MiniLM embeddings through fastembed
Result merging	Reciprocal Rank Fusion
Local storage	SQLite

Local LLM Support

Lens can run local GGUF models through llama.cpp.

When enabled, local models can help with:

• Answering questions from indexed content.
• Summarizing matching documents.
• Routing natural-language queries.
• Handling more complex search requests.
• Turning vague searches into more useful retrieval steps.

The model manager can download supported GGUF models and stores them locally.

Hardware acceleration is detected when available, depending on the build and platform support. Supported backends may include:

• CUDA
• ROCm
• Apple Silicon
• Vulkan
• Other llama.cpp-supported backends

File Ingestion

Lens extracts text and metadata from common local content types, including:

• Plain text files
• Source code
• PDFs
• Images
• Office documents
• Spreadsheets
• Presentations
• Videos
• Local mail files

PDFs with embedded text are read directly. Scanned pages can fall back to OCR.

Videos are scanned for metadata and embedded subtitles through ffmpeg.

Images can be processed with OCR and QR-code extraction.

Mail Indexing

Lens can sync IMAP mail into local .eml files and index them like regular documents.

This allows local mail search without relying on a remote mail provider search API.

Mail indexing is useful when you want the same search experience across files, documents, and email.

Smart Extraction

During indexing, Lens can detect structured values and store them with the indexed content.

Supported extracted values include:

• URLs
• Email addresses
• IP addresses
• MAC addresses
• IBANs
• Dates

This makes it easier to search for specific technical, financial, or contact-related information.

Fast-Path Plugins

Some queries do not need the full search or LLM pipeline.

Lens includes fast paths for simple actions, including:

• Calculator queries
• Desktop application launching
• Filesystem browsing

These are handled directly to keep common actions instant.

Privacy Model

Lens is designed to be local-first.

By default:

• Document content is indexed locally.
• Embeddings are created locally.
• Search data is stored locally.
• Models are stored locally.
• No indexed document content is sent to an external AI service.

Network access is only needed for features that explicitly require it, such as:

• Downloading local models.
• Syncing configured IMAP mail.
• Accessing user-configured remote resources.

Installation and Setup

System Dependencies

Install the required system tools first:

sudo apt install tesseract-ocr ffmpeg poppler-utils curl

Dependency	Used For
tesseract-ocr	OCR for images and scanned PDFs
ffmpeg / ffprobe	Video metadata, subtitles, and thumbnails
poppler-utils	PDF rasterization for OCR through pdftoppm
curl	Model downloads

Increase the Inotify Watch Limit

Large folders may hit the default Linux inotify watch limit.

Increase it with:

echo 'fs.inotify.max_user_watches=524288' | sudo tee -a /etc/sysctl.conf
sudo sysctl -p

Building from Source

Install Rust and Cargo first.

Then build Lens:

git clone https://github.com/cwittenberg/gnome-lens.git
cd gnome-lens
cargo build --release

The release binary is created here:

target/release/gnome-lens

Running Lens

Start the daemon manually:

./target/release/gnome-lens

The GNOME extension can also manage the daemon automatically.

Command-Line Usage

Index a Folder

./target/release/gnome-lens index /path/to/folder

Force a Full Reindex

./target/release/gnome-lens reindex /path/to/folder

Query the Running Daemon

./target/release/gnome-lens "what are the terms of my apartment lease?"

Architecture

Lens runs as a local daemon and exposes search functionality to the GNOME desktop integration.

At a high level, the system contains:

Component	Purpose
File watcher	Tracks filesystem changes using inotify
Ingestion pipeline	Extracts text, metadata, and structured values
Search index	Stores searchable text in SQLite FTS5
Embedding pipeline	Creates local semantic vectors
Ranking layer	Merges exact and semantic results
Model manager	Handles local GGUF model downloads and paths
LLM runtime	Runs local llama.cpp queries when enabled
IPC layer	Serves requests from the desktop integration

Runtime Paths

Lens stores its local state under standard user directories.

Component	Default Path
IPC socket	~/.local/state/gnome-lens/gnome_lens.sock
Configuration	~/.config/gnome-lens/
Database, models, and local data	~/.local/share/gnome-lens/

When running inside Flatpak, Lens uses the matching Flatpak and XDG paths instead.

Storage

Lens stores the following locally:

• Search index
• Extracted text
• Extracted metadata
• Embeddings
• Model files
• Local mail cache, when IMAP sync is configured

No document content is sent to an external service by default.

Project Status

Lens for GNOME is under active development.

The project is focused on:

• Fast local indexing.
• Reliable folder watching.
• Strong GNOME desktop integration.
• Useful local AI-assisted search.
• Private-by-default behavior.
• Practical performance on normal desktop hardware.

License

This project is licensed under the GNU General Public License v3.0.

See the LICENSE file for details.