Questions, answered

Fan apps move air. CyberKaze protects the Mac you work on while local AI runs. Overload is load on this machine. Under Overload with Protection On, listed AI is held in place so the machine is not fighting the AI for CPU, then the work continues in place. In short measured peak hybrid Overload windows (~90s) with Protection engaged (pause), the Mac you work on stays usable under that scoped test on one Mac. The Slow path and AI-only Protection are not covered by that evidence. Retaining memory: On. Protection Off. No easing, even under overload. The model stays loaded.

Protection applies to: Ollama, mlx_lm.server / mlx_lm serve (Apple), LM Studio headless, process named llama-server / llama-cpp-server. LM Studio GUI is never touched. Code editors, browsers, Mail, and generic Python are outside Protection. On Linux, the default target is Ollama only.

Only the allowlisted AI runtimes: Ollama, mlx_lm.server / mlx_lm serve (Apple), LM Studio headless, and processes named llama-server / llama-cpp-server. LM Studio GUI is never touched. Code editors, browsers, Mail, and generic Python are outside Protection. On Linux, the default target is Ollama only. Rename a custom llama binary to llama-server or llama-cpp-server if you want it eased.

No. The menu-bar app is the always-on surface: live status, Protection On/Off, and Live data. The header shows when checks are running. The CLI remains available for power users and scripts. Download the menu-bar build when you want protection without living in a terminal.

No. Fan apps move air. Macs Fan Control and TG Pro manage fans and sensors. CyberKaze protects the Mac you work on: listed AI is eased back under Overload, then the work continues. Protection Off. No easing, even under overload. Complementary tools. Different job, same Mac.

Fan apps move air and show degrees. CyberKaze watches Overload: load, and extra memory on disk while the CPU is still busy. Under Overload, listed AI is held in place so the machine is not fighting the AI for CPU, then continues. High extra memory on disk with a quiet CPU is Using extra memory on disk. Work is progressing. AI is left running. Degrees are a scoreboard. Holding the listed AI in place is the technique.

The check itself is cheap: load and extra memory on disk, once a minute. Monitor never eases listed AI. LIVE Protection is different. Under Overload, listed AI is held in place so the machine is not fighting the AI for CPU. That is the default. Retaining memory: On. Protection Off. No easing, even under overload. Using extra memory on disk (quiet CPU) leaves AI running.

Apple signed and notarized the macOS app with a Developer ID. Gatekeeper shows Paul Sully, Individual account, not Attest Dojo. It is offline-first: the license is a file on your machine, not a cloud login. High extra memory on disk with a quiet CPU is Using extra memory on disk. Work is progressing. AI keeps running. Listed AI is eased back under Overload. Retaining memory: On. Protection Off. No easing, even under overload. The model stays loaded. Work continues in place. You get a 10-day trial, then paid, with no money-back, because the trial is the evaluation. After pay we email a claim link; you download a signed license file. It does not auto-install. Seller is Attest Dojo, Osaka. Support is staff@attestdojo.com. Paul used to Force Quit and freeze; that relief was unexpected. It is not a freeze warranty. We will not claim watt savings, a degrees win, or room air conditioning.

As far as we can tell, CyberKaze is the first packaged Mac utility that tells the truth about Overload versus using extra memory on disk, then eases Ollama, MLX, and LM Studio headless under real stress and continues them when the machine is calm. The model stays loaded. That is a narrow claim. Other tools already freeze or slow apps. We did not invent freeze and thaw.

No. We do not sell a °C win, room AC, or an energy claim. Overload is load on this machine, or extra memory on disk while the CPU is still busy. We watch those signals. We do not race thermometer apps.

Using extra memory on disk is high extra memory on disk with a quiet CPU. That is not Overload. CyberKaze says so out loud and leaves AI running. The locked line is Using extra memory on disk. Work is progressing. Overload is the other reading: real load, or extra memory on disk while the CPU is still busy. Only Overload eases listed AI when Protection is On.

Overload is the customer name for real load on this machine, or extra memory on disk while the CPU is still busy. Protection On is the default: when Overload is real, listed AI is held in place so the machine is not fighting the AI for CPU, then the work continues in place. Retaining memory: On. The model stays loaded. Eased back means the listed AI is held in place until the machine is calm. Protection Off. No easing, even under overload.

Three honest rules. Extra memory on disk alone never stops anything, ever. Extra memory on disk only counts toward Overload when the CPU is already working hard. Overload means the CPU is genuinely working hard, or extra memory on disk plus a busy CPU. Every verdict prints its reason.

Yes, with an honest caveat. Apple Silicon exposes no readable thermal sensor, so on an M1 or M4 CyberKaze leans on the load reading and says so out loud. That is not a weakness, it is the point: load is the Overload signal, and it never lies the way a sensor number can. On Intel Macs and Linux it can read OS pressure too.

Overload is the name for real load on this machine, not a broken thermometer. When Protection is On and the machine is honestly overloaded, listed AI is held in place so the machine is not fighting the AI for CPU, then the work continues. You will never see Overload without what Protection is doing next to it. When Protection is Off, there is no easing, even under overload.

If overload is not eased back, listed AI can crowd the machine you work on. People Force Quit. The model leaves memory. Work has to start again. Protection eases that AI, retains memory, and your work stays in place. The result: no cold reload, and the work continues where it left off.

No. Mercy is a design rule. Default Protection eases listed AI: the listed AI is held in place until the machine is calm. Keeping things within safe limits. Retaining memory: On. The model stays loaded.

Unloading is a choice you make, not something Protection does on its own. When RAM is the problem, use Unload listed AI in the menu, or run cyberkaze unload-listed. It stops the running Ollama model, so the memory that model held is released, and the model reloads the next time you use it, so the step is reversible. This is separate from Protection, which eases listed AI under Overload and keeps the model loaded. Mercy is the design rule and it holds here too: CyberKaze holds listed AI with SIGSTOP and releases it with SIGCONT, and it never sends SIGTERM, so a running job is never killed out from under you. Unload is allowlisted AI only, and code editors, browsers, and Mail are never touched. Other allowlisted AI (MLX, LM Studio headless, llama-server) is not unloaded by this command; stop it from its own app if you need that memory back.

Only the allowlisted AI, and only under real Overload. Protection is On by default, so listed AI is eased back when Overload is real; code editors, browsers, Mail, and generic Python are never touched. When it acts, it logs the reason out loud, and there is a reversible escape hatch: turn Protection Off in the menu bar, or run cyberkaze disable. Either releases anything held and returns the machine to watching only.

Ten days of the paid tier on your own machine once Protection On is on: listed AI is held in place under Overload so the machine is not fighting the AI for CPU, then the work continues in place, plus quarantine and the other LIVE tools. Turn Protection On to feel the paid tier during the trial. Until then it watches and shows verdicts only. Protection Off. No easing, even under overload. The trial is the honest way to buy: you use the real product first, then decide. After it, the free monitor tier keeps watching forever.

The first paid-tier command, license status, or value starts a 10-day clock bound to local state. The moment it starts you get a welcome line explaining the deal: monitor tier free forever, paid tier runs through the trial, one-time per-machine license after it.

Trial ready means Protection is available now and the 10-day paid-tier clock has not started yet. Run a paid-tier command, license status, or value to start the trial. During the trial, every paid feature runs on your machine. After day 10, paid actions refuse until you activate a signed license file. Paying on Square does not install the license. The seller emails the file; you run cyberkaze license activate <file>. The free monitor tier keeps watching forever, licensed or not.

cyberkaze value prints Protection eased (all time) when CyberKaze has logged real easing events in heat-scheduler.log. It is all-time logged hours where Protection eased listed AI under Overload, with memory retained, not a live loss meter and not a room-heat or energy claim. Fresh installs may show nothing until real Overload days pass. The monitor tier records verdicts for free; the number grows when Protection is On and Overload is real.

Paid-tier commands refuse. The monitor tier keeps watching. Easing under Overload stays off until you activate a license. The menu bar shows Trial ended, Protection is report-only, and a Buy button that opens checkout (https://attestdojo.com/cyberkaze/checkout.html). Paying does not install a license. You receive a claim link by email, download a signed license file, then Activate in the menu or cyberkaze license activate <file>.

The whole truth-teller, free forever: verdict every minute, Overload history, extra-memory-on-disk flags, the report, and the value report. It observes and reports and never eases listed AI. It is the free funnel, and the full story of what the tool does.

Because the trial runs the paid tier and the monitor tier is free. The paid tier is what acts: listed AI is eased back under Overload, then continues. The monitor tier watches and never sheds. You pay once. Single is 20 USD introductory until 31 October 2026, then 27 USD regular; a Japan card is charged by Square in yen 税込, ¥2,980 until 31 October 2026, then ¥3,980. That is after 10 days of using the real thing, not before.

Pay on Square or crypto from the site checkout or the menu Buy button (checkout: https://attestdojo.com/cyberkaze/checkout.html). Neither rail installs a license. After payment we email a claim link. Open it, paste your machine code from cyberkaze license status, download the signed file, then run cyberkaze license activate <file> or Activate license in the menu. One purchase, one machine, no license server, nothing phones home. Staff@attestdojo.com is the fallback if the claim email is missing.

The activation is a file, not a key. If you lose it, email staff@attestdojo.com with the machine's fingerprint from cyberkaze license status, and the license is re-issued for the same machine. Keep a backup copy of the file once you have it.

No money-back refunds on purchases, and here is why that is fair: you get the full 10-day trial before you pay anything, so no one pays for a tool they have not already used on their own machine. Digital downloads cannot be returned like a boxed product, and the trial is the evaluation window. If a license will not activate, or a purchase was a genuine mistake, email staff@attestdojo.com and we work it out case by case. One plain caveat: consumer rights vary by country and are yours to know; where the law grants a right we cannot waive, the law wins.

A license is bound to one machine, and moving to a new one is a simple, stated step. Retire the old machine, email the seller with the new machine's fingerprint, and the license is re-issued for that fingerprint. The old machine's copy is deactivated with cyberkaze license deactivate. No charge by default, no surprise.

Same answer, spelled out: the license is per machine, so a move needs a re-issue. The agreement permits a transfer when the original machine has been retired and the seller agrees, and in practice the seller re-issues a license for the new fingerprint, no charge by default. A pack works the same way, one signed file per machine, and a retired machine's slot is freed by the same agreement with the seller.

From the menu bar, turn Protection Off. Same hatch as the CLI: it releases anything held or throttled, marks management off in the shared preference, and the loop only watches. Turn Protection On (or cyberkaze enable) to manage again. To remove everything, cyberkaze uninstall archives your state and removes the app, never deleting your history without asking.

Updates are check-and-tell, never a silent broadcast. In the menu bar, Check for Updates checks attestdojo.com and tells you if a newer signed build exists. In the CLI, run cyberkaze check-update. CyberKaze never auto-installs. You download the new signed DMG and install over the old copy, which preserves your trial and license. There is no push notification and no remote patch channel.

No. State lives in your user data directory, not the app bundle. Installing over the top preserves the trial and the license.

You run cyberkaze check-update, or you visit the site. There is no silent push and no remote update channel, by design: updates are opt-in and visible. Everyone gets the same fix at the same time, and the fix ships as a new build you install.

That is usually a hosting glitch on our update server, not something you did wrong. Quit CyberKaze from the menu bar, open the Download page on this site, and install the latest signed DMG over your existing copy. Your trial and license stay in place. If Check for Updates keeps failing after a fresh install, email staff@attestdojo.com with your app version from About.

Email staff@attestdojo.com. That is the single channel, stated in the license, no ticket queue, no third-party helpdesk. If something is wrong with your machine's plumbing, the answer is often one command you run yourself: cyberkaze doctor --heal.

On older installs, a hung background check could leave Protection quiet until the next run. Current builds wrap each check in a 90-second deadline. Run cyberkaze doctor and look for agent-check-timeout. If it FAILs, run cyberkaze doctor --heal. Then cyberkaze status should show a fresh last check within a couple of minutes.

cyberkaze support report builds a sanitized bundle of everything a diagnosis needs: version, platform, license mode, doctor checks, recent verdicts. cyberkaze support send composes the email in your own mail app with the bundle already attached, so you just click send. Nothing is ever sent by the app itself; only your own mail app sends.

Open the support form at support.html#form on this site, or use Report a problem in Live data when the app opens that page with your version prefilled. Run cyberkaze support report, copy the bundle text to the clipboard, and paste it into the form message field. Email staff@attestdojo.com still works. Nothing is sent by CyberKaze itself until you submit the form or send the email yourself.

The person who sells it. Paul, at Attest Dojo. Replies are written by a person in plain language, in your language where possible, and the auto-reply is only a "got your mail" acknowledgment, never an answer.

Yes, on the site, when the site ships. Public threads for product discussion, feature requests, and answers to questions, on Attest Dojo's own domain. Login is by license number, so only paying customers post, and you set a display name on top. Support stays by email; the board is for public conversation.

No, not by default. Everything runs locally on your machine. CyberKaze serves your Mac, not a cloud. Telemetry exists but is opt-in, off by default, and you see exactly what a payload would contain before anything can be sent. Nothing is ever collected unless you explicitly turn it on.

If you opt in, the minimum needed to find and fix bugs: version, platform, load and swap readings, verdict counts, and error kinds. It never carries file paths, user names, hostnames, IPs, process command lines, model names, or any content. Every payload is scrub-verified before send, and cyberkaze telemetry preview shows you the exact contents first.

No. CyberKaze has no remote-execution channel into your machine, by design. A bug is fixed in the source and ships as a new build you install yourself. Support is a conversation, never remote control, and that is a deliberate trust and security decision.

Works on: Mac, Apple silicon first. Intel intended, not yet proven on our hardware. macOS 14+ for the extra. CLI on the same Macs (Node 20+). Linux: x86_64 CLI. Default target is Ollama. No extra. Protection applies to: Ollama, mlx_lm.server / mlx_lm serve (Apple), LM Studio headless, process named llama-server / llama-cpp-server. Reads this machine's swap and load. Same rules on every Mac. Does not run: Windows (including Ryzen AI Halo stock Windows, Copilot+ PCs); iPad, ChromeOS; AMD NPU stacks (Lemonade, DirectML); MLX on non-Apple; Code editors, browsers, Mail, LM Studio GUI, generic Python (Protection does not apply). Linux is a split, not the main product. The Mac extra will not run on Halo. Linux CLI may ease Ollama on a Halo box if Ollama is the process. That is not a tested first-class SKU. Exact list: Download, Works on / Does not run.

Yes, as an x86_64 CLI split, not the main product. Default target is Ollama. No extra. The Mac extra will not run on Linux. Linux CLI may ease Ollama on a Halo box if Ollama is the process. That is not a tested first-class SKU. Exact list on Download.

No. Windows, including Ryzen AI Halo stock Windows and Copilot+ PCs, does not run. Also does not run: iPad, ChromeOS; AMD NPU stacks (Lemonade, DirectML); MLX on non-Apple. Code editors, browsers, Mail, LM Studio GUI, and generic Python are outside Protection. Exact list on Download.

On macOS, no. v1 reads load, swap, and thermal sysctls with read-only calls and never needs sudo. A future root helper for suspending other users' processes is the reason this ships as a signed DMG, not the App Store, but nothing in v1 requires elevation. On Linux, sensing needs no elevation; installing the systemd service uses root.

Prices are listed in USD. The MONITOR tier is free forever. LIVE Single is 20 USD introductory until 31 October 2026, then 27 USD regular. Packs: Small up to 3 is 74 USD, Team up to 5 is 107 USD, Fleet up to 10 is 188 USD, Site up to 25 is 403 USD. Enterprise, 50 machines and up, is from 671 USD, about ¥99,800 税込, a custom quote by email. A Japan card is charged by Square in yen 税込: Single ¥2,980 until 31 October 2026, then ¥3,980, and the packs at ¥10,980 / ¥15,980 / ¥27,980 / ¥59,980. No subscription. Checkout: https://attestdojo.com/cyberkaze/checkout.html. Paying does not install a license. You get a claim link by email, download a signed file, then cyberkaze license activate <file>.

One machine, one license: Single is 20 USD introductory until 31 October 2026, then 27 USD. More machines, buy the pack that fits your ceiling: Small up to 3 is 74 USD, Team up to 5 is 107 USD, Fleet up to 10 is 188 USD, Site up to 25 is 403 USD. Enterprise, 50 machines and up, is from 671 USD, about ¥99,800 税込, a custom quote by email. A Japan card pays the same ladder in yen 税込 through Square: ¥2,980 until 31 October 2026 then ¥3,980, and ¥10,980 / ¥15,980 / ¥27,980 / ¥59,980 for the packs. Packs are up-to-N, so buy the one that fits your fleet, not the biggest one, and a pack counts Macs and servers together.

Because this is a one-time tool, not a service. A small, fixed price matches a small, fixed job, and it does not make you pay forever for something that runs on your own machine. One-time payment, no subscription fatigue.

Yes. One engine, one price per machine, Mac or server, and a pack counts a mixed fleet against the same machine count. The cost to serve is identical, so charging differently would just be a made-up split.

Japan Square checkout is 税込. The listed yen is the card charge, and no extra 10% is stacked on those yen prices. The USD figures are the amounts the crypto checkout charges. For sales outside Japan, the seller handles tax as it arises.

Yes. Pay with crypto (BTC, ETH, SOL, USDC on Polygon, and other enabled coins) via NOWPayments hosted checkout from the checkout page. Prices are fixed in USD (20 USD intro Single until 31 October 2026, then 27 USD regular; Small 74, Team 107, Fleet 188, Site 403 USD; Enterprise, 50 machines and up, is from 671 USD, a custom quote by email). We settle in USDC on Polygon. Card buyers in Japan should use Square and pay yen 税込. NOWPayments does not mint or install a license. After payment we email a claim link; paste your machine code from cyberkaze license status, download the signed file, then Activate. Staff@attestdojo.com is the fallback. Network and NOWPayments service fees are paid by the buyer; cross-coin payments may include a conversion fee.

They live in the terms document, linked in the site footer and at checkout. In plain language: one-time per-machine license, the 10-day trial runs every paid feature before you pay, no money-back refunds because the trial is the evaluation window, telemetry is off by default, support is by email, and there is no remote patch channel into your machine. The terms document is a v1 draft and says plainly it is not legal advice and has not been lawyer-reviewed yet.