Changelog

Your folders often hold more than what you want AI to read: thousands of screenshots, gigabytes of screen recordings, dependency packages, and private material you would rather keep away from agents. You can now filter by file type, and use exclusion rules to skip specific files and subfolders. On top of that, the official models get a full upgrade, and remote MCP now hides more than 200 kinds of sensitive information automatically.

The index, local models and other app data take up a lot of disk space, and many of you asked to change the default location. It's here now, on Windows, macOS and Linux. Linkly AI is also now listed in Claude's connector directory.

Linkly AI reaches its first stable release, v1.0. This version redesigns the whole app: all your material is laid out in view. The library, clips, index status and agent usage each have a screen of their own, so you can manage your knowledge files with ease and see how agents are using these knowledge assets.

Linkly AI v0.14.1 adds Linux arm64 installers, so arm64 machines such as NVIDIA DGX can now run Linkly AI and let an agent search the material on the machine directly. Chinese search also gets better word segmentation and ranking, so whole sentences typed without spaces hit more precisely. On top of that come 20-odd improvements and fixes covering spreadsheet parsing, scanned-document OCR, disconnected external drives and cloud library uploads.

Linkly AI now supports Excel and CSV. To let an agent search the spreadsheet data on your machine, you no longer upload files one at a time — thousands of Excel files can be searched and compared where they are. We also reworked the built-in OCR: it now schedules recognition models to balance speed against quality, for an average speed-up of more than 3×.

Semantic indexing on Windows and Linux now runs on your graphics card, which makes it considerably faster. Indexing work is also distributed by CPU core count, so the slow crawl large libraries used to settle into is back to normal speed.

System requirements moved back down in this release: macOS is supported from 13.4 again, and the Linux glibc baseline drops to 2.35. Machines that could not install the previous version because they were too old can install this one.

Audio and video transcription can now run in the cloud. Drop a meeting recording, a podcast or a video lecture into a library and it comes back faster and more accurate than the local engine manages, without spending your machine's resources. The same release brings six note insight skills — pick an angle and have an assistant read your own writing back to you.

This release requires macOS 15.0 or newer. On anything older the app no longer quits without explanation; it tells you why.

Pro users can switch semantic indexing to the cloud. It is considerably faster, search results are sharper, and — the part you feel most — your own CPU and memory stop being tied up for long stretches. A large library can index in the background while you carry on working.

Local indexing is not going anywhere. It stays the default and stays fully offline; the cloud is a switch you can flip back at any time. This release also gave notes a timeline and a tag filter.