Dictation, autocomplete, chat apps — they quietly download models and keep them in memory long after you stop using them. LLM Radar surfaces the ones it can see — what they cost, and what you can unload or delete.
Download coming soonIllustrative mock of the Activity screen — not a live screenshot.
Install, scan, decide. About a minute.
Drag it into Applications. Grant Full Disk Access so it can see model files inside other apps' folders. No account, no sign-in.
It finds apps that can run a model, model files on disk, and which ones are loaded right now. Ten runtimes out of the box, plus a catch-all scan.
Unload what's idle, delete what wasn't observed in use. Nothing is removed until you confirm, and deletions go to the Trash.
Dictation, chat, autocomplete — each keeps its own models in the background. LLM Radar is the one place that shows all of them together.
One central place for local-model apps LLM Radar knows — plus a catch-all scan — so a model is understood as a model.
Models not observed in use during monitoring, grouped with disk cost. Deletions go to the Trash and name every file first.
No account, no sync. Anonymous usage counters are on by default and switch off in Settings — no paths, model names, or prompts, ever. Optional crash reports only when you enable them and the build includes a Sentry DSN. The other optional network call downloads the list of known apps.
Ollama, LM Studio, MLX, llama.cpp, Core ML, whisper.cpp, Docker and more — plus a catch-all for anything we haven't met yet.
Model weights, KV cache, app overhead — so you know what unloading actually buys you before you do it.
A running total and one-click unload without opening anything. You pick what shows: icon, memory, busiest app, or a graph.
A model pinned to the GPU costs far more than the same work on the Neural Engine. LLM Radar is the thing that tells you which is which.
The list updates without an app update. Anything not on it is still caught by the runtime-agnostic scan.
No card and no subscription. Monitoring stays free; Pro is a one-time lifetime licence.
A file of AI weights that runs on your Mac instead of on a server. It makes an app work offline and keeps your text private — but it has to sit in memory while it works, and it is often several gigabytes.
Yes. The app keeps running and reloads its model the next time it needs one. You lose a second or two on the next request; you get the memory back immediately. Where a runtime has no unload API, LLM Radar says so rather than pretending.
Usage varies with the number of apps and model folders on your Mac. LLM Radar samples less often on battery, and you can pause monitoring from the menu bar. You can verify its current CPU and memory use at any time in Activity Monitor.
To find model files that live inside other apps' folders. It reads names, sizes and dates only — never file contents, never anything outside model directories. Skip the permission and LLM Radar still shows running processes, just not what is sitting on disk.
Only if it was that app's only model, and LLM Radar tells you before you confirm. It also tells you whether the owning app can simply download it again. Deletions go to the Trash by default, so nothing is irreversible unless you choose that in Settings.
Almost nothing. There is no LLM Radar account and no sync. Anonymous product-usage counters are sent by default — which screens and features were used, the app and macOS version, Apple silicon yes/no, and memory to the nearest GB — and you can switch them off in Settings → Privacy & Data. They never contain file paths, model names, model contents, prompts, process lists, your email, or your licence key. Crash reports are off by default; if you enable them and the build includes Sentry, anonymous stack traces may be sent — never model files, paths, or process lists. Licence activation sends the key and a random installation label to Lemon Squeezy; paid licences are revalidated periodically. See Privacy.
Unknown apps may still appear as running processes, but model attribution is best for the runtimes and apps listed above. Tell us about a missing app and we can add a verified adapter.
It runs, but Neural Engine metrics and most MLX detection are Apple silicon only. LLM Radar is built for M-series Macs first, on macOS 14 or later.
Every Pro feature is available for 7 days from first launch, with no card required. After that, monitoring remains free. Pro is a one-time $9.99 licence for one Mac or $14.99 for up to three Macs.
Yes. Production builds check a signed Sparkle feed and only install Developer ID-signed, notarized updates.
LLM Radar reads process and file metadata locally (names, sizes, dates, memory use). It does not upload your models, prompts, chat content, or anything a model produces. There is no account and no sync.
LLM Radar sends anonymous product-usage counters so we can see which features are worth keeping and which screens people never reach. This is on by default, and you can switch it off at any time in Settings → Privacy & Data. Switching it off takes effect immediately; no relaunch is needed.
What is sent: the event name, a random installation id that is generated on this Mac and is not linked to you or to your licence, the app version and build, the macOS version, whether the Mac is Apple silicon, memory rounded to the nearest whole GB, the licence state (trial, expired, or licensed), days since first launch, and coarse counts, size buckets, and durations — which screens were opened, which features were used, and how long a screen was open.
What is never sent: file paths, model file names, model contents, prompts or chat content, process lists or command lines, your macOS username, your email address, your licence key, and any hardware serial number. The event format accepts only fixed names, numbers, and yes/no values — there is no free-form text field for user content to travel in. These counters go to PostHog, and nowhere else.
Crash reports are off by default and always user-toggleable. When enabled and configured, anonymous stack traces and app version may be sent to Sentry — never model files, paths, or process lists. Licence activation and periodic validation contact Lemon Squeezy with the licence key, a random installation label, and Lemon Squeezy order metadata. Separately, an optional toggle can download an updated list of known apps.
This website uses Google Analytics 4 to measure aggregate page visits, referrals, download clicks, checkout visits, and purchases. Browser privacy settings, consent choices, or ad blockers may prevent that collection. Download buttons also pass through a first-party endpoint on this domain, and a checkout click is posted to a matching first-party endpoint, so those two actions can still be counted when GA4 is blocked. Each of those events is limited to the time, release or plan, button location, campaign labels, referrer hostname, broad browser/bot category, and a request ID; LLM Radar does not add an IP address or persistent user identifier to it. These website events are stored in PostHog, the same place the app’s anonymous usage counters go, so a download and a first app launch can be counted together. Website measurement and app telemetry stay separate switches: blocking this site’s scripts does not change the in-app toggle, and turning the in-app toggle off does not change this site. Licence activation and validation continue to be handled by Lemon Squeezy as described above. Read Google's privacy information.
Used only to find model files inside other apps’ folders. Skip it and LLM Radar still shows running processes, just not on-disk inventory.
Email faisal@dataweal.org for questions about licences, purchases, refunds, or the app. This is the fastest way to reach a human; we aim to reply within two business days.
Open Send feedback in the app, under Settings → About. It drafts a message in your own mail app — nothing is sent until you press send. When reporting a crash or bug, attach Export diagnostics… from Settings → Privacy & Data, after checking that it contains no information you do not want to share.
Describe what you expected and what happened, and attach the exported diagnostics. Do not paste secrets, Sentry DSNs, or Sparkle private keys.
These Terms of Service govern your use of the LLM Radar macOS app and this website, provided by dataweal (“we”, “us”). By downloading, installing, or using LLM Radar you agree to these terms; if you do not agree, do not use the app. Effective date: August 19, 2026.
We grant you a personal, non-exclusive, non-transferable licence to install and use LLM Radar on the number of Macs included in your plan — one Mac for the Single licence, up to three for the Personal licence. Monitoring features stay free after the trial. You may not resell, redistribute, sublicense, or reverse-engineer the paid components, or remove its licensing and update mechanisms.
Every Pro feature is free for 7 days from first launch, with no card required. After the trial, monitoring stays free and Pro is a one-time purchase: $9.99 for one Mac or $14.99 for up to three Macs, with no subscription. Purchases and licence keys are processed by Lemon Squeezy, our merchant of record, who also handles applicable taxes. Prices are charged in the currency shown at checkout and may change for future purchases.
If Pro does not work as described, contact us within 14 days of purchase and we will review a refund through Lemon Squeezy. Because monitoring is free and Pro includes a full no-card trial, please evaluate the app during the trial before buying.
LLM Radar can unload models and move model files to the Trash. You are responsible for confirming these actions. Deletions go to the Trash by default so they can be recovered, but you should keep your own backups of anything important, and only remove files you are authorised to manage.
LLM Radar is provided “as is” and “as available”, without warranties of any kind. Detection and attribution across runtimes are best effort; a model shown as “reclaimable” means only that it was not observed in use during the monitoring window, not that it is safe to delete. We do not warrant that the app is error-free or that it identifies every app, model, or file.
To the maximum extent permitted by law, we are not liable for indirect, incidental, or consequential damages, or for lost data, lost profits, or the cost of recovering files, arising from your use of LLM Radar. Nothing in these terms limits liability that cannot be limited under applicable law.
Payments and licensing are handled by Lemon Squeezy. Optional crash reporting uses Sentry and is off by default. App updates use Sparkle. This website uses Google Analytics 4. Anonymous product-usage counters from the app and this website use PostHog, and the app’s counters can be switched off in Settings → Privacy & Data. How each provider handles data is described in our Privacy section and each provider’s own policy.
We may update these terms as the app changes. Material changes will be reflected here with a new effective date. Continuing to use LLM Radar after an update means you accept the revised terms.
These terms are governed by the laws of Ontario, Canada. Questions about these terms can be sent to faisal@dataweal.org or raised through the Support section below.
Find out what your Mac has been running all this time.
Download coming soon