Repository navigation
Loach v1.4.0 knows your hardware, searches inside messages, renders LaTeX and locks itself #45
ztcs-software
announced in
Announcements
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
Loach v1.4.0 starts at the beginning. Onboarding was rebuilt and the setup now recommends a model your hardware can actually run, search finally looks inside transcripts, math renders properly, and the app can lock itself when you walk away.
👉 Download Loach v1.4.0
or update directly from the app in Settings → Updates.
A setup that knows your machine
First launch got a rework. When Ollama has no models yet, setup reads your hardware - GPU VRAM if you have a discrete card, system RAM otherwise - and recommends the largest model that will run comfortably. Every entry in the catalog gets a plain verdict: Runs well, Runs OK, Runs slowly, Too big for this machine or Not enough disk space. Only the last one blocks anything; the rest are advice.
The download runs in the background while you finish setup and follows you into the chat as a banner above the composer. If you try to send before it's done, Loach tells you the model is still downloading instead of letting Ollama fail with "model not found".
There's also a new Tools step. The built-in utilities open on a Recommended selection, with All OFF / Recommended / All ON presets and a switch per tool. Web fetch is the exception: it's the only tool that reaches the network, so it stays off until you turn it on yourself.
Search inside messages
Cmd/Ctrl + K now searches transcripts, not just titles. Message hits get a MESSAGE badge and a snippet around the match; pick one and Loach opens the chat and jumps to that turn. Title matches still sort first, but message hits keep a reserved block at the bottom so a common word can't crowd them out.
To narrow the search, use the scope dropdown - everywhere, chats, messages, spaces, snippets - or type the same thing as an in: filter, like in:messages borrow. Archived chats, Private Chat and text inside attachments are not searched.
LaTeX math
Formulas now render.$$…$$ , $…$ , (…), […] and ```math fences are typeset with KaTeX, with no setting to turn on. Since $ is also a currency sign, $…$ only becomes math when the span actually reads like math - "it costs $5 and $10" stays prose, $x^2$ renders. TeX that doesn't parse shows its own source instead of an error. KaTeX ships inside the app, so nothing is fetched from the network.
Auto-lock
Until now the app lock only gated launch. Once a lock is configured, Settings → Security offers two opt-in triggers: Lock after inactivity (1, 5, 15 or 30 minutes) and Lock when minimized. A reply already streaming keeps going behind the lock screen, and native save or open dialogs don't trigger the minimize lock. Cmd/Ctrl + Shift + L locks immediately from anywhere in the app.
Storage breakdown
Settings → Data now shows what Loach keeps on disk: the database's size and path, split into chats, attachments, Spaces and everything else, plus how many local models Ollama has and what they weigh. The models are listed for context only - Ollama stores them outside Loach's data folder.
In-app updates for .deb and .rpm
Linux package installs now update in place, like the AppImage. Loach downloads the signed package and installs it through pkexec, which is why the update asks for your password. There's still no apt or yum repository, so updates are in-app only. If you installed v1.2.3 or earlier from a .deb or .rpm, download a newer package once from the releases page and in-app updates take over from there.
Other changes
Fixes
👉 View the full changelog in GitHub Releases
All reactions