Offline Installation Guide

Linkly AI’s search, text recognition (OCR), speech transcription (ASR), and semantic search all run locally on your machine — everyday use needs no internet connection. But two steps of the normal installation flow do depend on the network: downloading the installer, and the automatic download of the AI model files (about 870 MB in total) on first run. If the target machine has no internet access at all (offline environment, corporate intranet, air-gapped network), follow this page: download everything on a connected machine, carry it over, and install.

For the model files, use the official offline model pack from this page. Do not download model files from other sources or assemble the directory layout by hand — model files are easily corrupted during manual transfer and re-encoding, which leads to AI features failing with confusing errors.

Step 1: What to download first

On a machine with internet access, download these two things:

The app installer: pick the target machine’s platform (Windows / macOS / Linux) on the official download page. Version v0.9.0 or later is required (earlier versions use a different model directory layout, so the offline pack doesn’t apply to them).
The official offline model pack: a single zip file, about 850 MB, containing every model needed for OCR, speech transcription, and semantic search, plus a readme and a SHA-256 checksum list.
(Only if the target machine runs Windows) two system runtime installers — offline, the app installer cannot fetch them from the internet automatically:
- WebView2 Runtime (Evergreen Standalone Installer, x64): Windows 11 and most Windows 10 machines with Edge already have it; stripped-down enterprise / LTSC systems may not — bring it along so installation can’t get stuck
- Microsoft VC++ 2015–2022 Redistributable (x64): the app’s local AI components depend on it; when missing, the installer tries to download it online and fails

Download the Offline Model Pack

linkly-ai-offline-models-20260825.zip · about 850 MB. Keep the zip as-is after downloading — don’t open or edit anything inside it.

Step 2: What to copy over

Using a USB drive, external disk, or internal file share, copy both files as-is to the offline machine (the desktop or any temporary location is fine):

The app installer (.exe / .dmg / .AppImage, etc.)
The offline model pack zip (copy the whole zip — don’t extract first and copy the pieces)
(Windows) the WebView2 and VC++ runtime installers

Step 3: How to install

Install the app

Windows: first install the WebView2 Runtime and the VC++ Redistributable in turn (machines that already have them skip harmlessly), then double-click the app installer and follow the wizard
macOS: open the .dmg and drag Linkly AI into the Applications folder
Linux: install according to your package format (for AppImage, make it executable and run it)

Launch once, then quit completely

Start Linkly AI normally once — the app creates its data directory automatically (offline, its model download attempt will fail; ignore that). Then exit the app fully: right-click the L icon in the system tray and choose Quit.

Open the data folder inside the app data directory

Find it for your operating system (create it if it doesn’t exist yet):

Windows
macOS
Linux

Press Win + R to open Run, paste, and hit Enter:

%APPDATA%\ai.linkly.desktop\data

In Finder choose Go → Go to Folder… (⌘ + ⇧ + G), paste, and hit Enter:

~/Library/Application Support/ai.linkly.desktop/data

mkdir -p ~/.local/share/ai.linkly.desktop/data
xdg-open ~/.local/share/ai.linkly.desktop/data

You can also click Open next to Data Directory under Settings → About, then enter data/.

Extract the model pack and drop the models folder into data

Extract the model pack zip anywhere — you get a models folder (alongside two info files, README.txt and SHA256SUMS.txt). Copy the whole models folder into the data directory you opened in the previous step — if data already contains a models folder, merge/overwrite.When done, the final layout must match the directory layout below exactly — for example, the file data/models/Qwen3-Embedding-0.6B-Q8_0.gguf should exist.

Start the app and verify

Launch Linkly AI and open Settings → AI: the model section should show as ready rather than waiting for a download. OCR (text recognition in scans/images) and semantic search work right away.Audio/video transcription is off by default: when you need it, turn on Audio transcription and Video transcription under Settings → Index — the app then works through your existing media files with the bundled speech model, no rescan needed.

Directory layout after installation

Once in place, the data/models folder should contain the following (file names must match exactly):

models/
├── Qwen3-Embedding-0.6B-Q8_0.gguf      ← semantic search model
├── ocr/
│   ├── .ready                          ← readiness marker (ships in the pack; leave it alone)
│   ├── v6/
│   │   ├── pp-ocrv6_small_det.onnx
│   │   ├── pp-ocrv6_small_rec.onnx
│   │   ├── ppocrv6_dict.txt
│   │   ├── korean_pp-ocrv5_mobile_rec.onnx
│   │   ├── ppocrv5_korean_dict.txt
│   │   ├── latin_pp-ocrv5_mobile_rec.onnx
│   │   └── ppocrv5_latin_dict.txt
│   └── structure/
│       ├── pp-doclayout-m.onnx
│       ├── slanet_plus.onnx
│       └── table_structure_dict_ch.txt
└── asr/
    └── sense-voice/
        ├── model.int8.onnx
        └── tokens.txt

ocr/.ready is a hidden file (its name starts with a dot) that marks the OCR models as ready. Do not open, edit, or recreate it — encoding differences introduced by text editors will make the app reject it. If you suspect it’s broken, just re-extract the offline pack.

Troubleshooting

The app still asks to download models after installation

Almost always a layout mismatch. Check against the tree above item by item, paying attention to:

Whether the zip was extracted directly into data/models, producing a doubly nested data/models/models/... — move the inner models contents up one level
Whether the hidden ocr/.ready file made it through (some extraction tools skip dot-files by default; in Finder press ⌘ + ⇧ + . to show hidden files)
Whether it ended up on a different drive or in another user’s home directory

The zip won't open or fails to extract

The file was corrupted during download or transfer. Download the pack again and try a different USB drive or transfer method. The pack ships with a SHA256SUMS.txt — the commands in its README.txt let you verify each file and pinpoint where things broke. Don’t attempt repairs or partial extraction — corrupted model files cause runtime errors that are much harder to diagnose.

AI features ask to download again after an app upgrade

A new app version may have switched to newer models. Download the latest offline pack and place it again following Step 3 on this page.

App version is older than v0.9.0

Older versions use a different model directory layout, so this pack doesn’t apply to them. Get the latest installer following Steps 1 and 2, upgrade the app, then continue.

What works offline

This pack covers the local AI models — not every feature. Once installed:

Feature	Offline availability
File indexing, full-text search, Launcher, notes	✅ Always available (no models needed)
Semantic search, OCR, audio/video transcription	✅ Covered by this pack
AI chat (Chatbot)	❌ Relies on cloud models; can be pointed at an OpenAI-compatible endpoint inside your network (e.g. Ollama / LM Studio on the LAN) — see the custom provider section of Model Service
Agent Skills installation	❌ Requires an internet download; not part of this pack
Account sign-in, cloud libraries, app auto-update	❌ Cloud features

Getting Started

Core Features

Connect Your AI Tools

For AI Agents

Offline Installation Guide

Step 1: What to download first

Download the Offline Model Pack

Step 2: What to copy over

Step 3: How to install

Directory layout after installation

Troubleshooting

What works offline

​Step 1: What to download first

Download the Offline Model Pack

​Step 2: What to copy over

​Step 3: How to install

​Directory layout after installation

​Troubleshooting

​What works offline

Step 1: What to download first

Step 2: What to copy over

Step 3: How to install

Directory layout after installation

Troubleshooting

What works offline