Introducing 82 · In private preview
A personal AI that lives on your Mac. It knows you, works for you, heals itself, and keeps evolving.
Not an app. Not an agent. Not a chatbot.
Why 82
Today's AI hands you a piece of the job and leaves you the rest. You write the prompts, explain yourself again, copy between apps, check whether it really finished, and watch the bill. 82 takes the rest. You say what you want and say yes. It does the work, checks it, remembers it, and keeps your money and your Mac safe.
Try the loop
You ask. 82 understands and plans, shows the cost if there is one, acts with its tools on your Mac, verifies, and remembers what helps you. Pick a request and watch the loop.
Cost card · before anything is paid
Illustration of 82's loop · the real one runs on your Mac
What living means
Living, for 82, is what it does every day on your Mac. Each sign is shipped, and each is measured.
Not a chat history. A memory, in clear sections: you, your work, your clients, your projects. What you tell it is kept at once; what it infers waits for your yes. Bring in what another assistant knew, with a full preview first.
17/17 recalled · 6.6 msDo a task once while it watches, and it becomes a skill. Reject a direction and it stops offering it. It tells your taste from your client's.
Learns by watchingIt opens apps, clicks, types, browses, cuts video, builds sites and writes code, and runs 30 areas of the Mac itself, with only the permissions you grant. Then it checks the screen before it says “done”.
32/32 Mac commands · 13/13 coding tasksClose the window and schedules, agents, Telegram, WhatsApp and voice keep running. Restart the Mac and it comes back. Reach it from your phone wherever you are.
28 channels · reach it from your phoneEvolve finds what is broken and proves the fix before it installs. AGI Growth proposes new abilities. Every version must beat the last one, or it doesn't ship.
Every repair proven · rollback alwaysVital signs
Tests bundled inside 82, run on the published build, with no personal data. The weakest score is published too.
17 of 17 recalled · 6.6 ms
211 of 211 requests understood
32 of 32 commands understood
Build 20261111094847 · v0.2.0 · 449 release checks passed
run by six programs working as one
actions you can trigger just by asking
AI models that run with no internet
heartbeat, always on
payments without your approval
chat channels, WhatsApp to Signal
Why it is different
The models are engines: brilliant, interchangeable, and flawed in the same ways. 82 is the layer they were missing. It uses every one of them, and trusts none of them.
Your memory, files, chats and keys live on your machine. Not on our server, not on anyone's. Reach it from your phone through Telegram, WhatsApp and the phone companion page, while the work stays on your Mac.
82 is free. Your AI provider is paid directly, with no markup, and every paid step waits for your yes.
English everywhere: the interface, the voice and the replies, on the Mac and on WhatsApp.
Every “done” is checked against real results before you see it. And its own test numbers are published, misses included.
Everyone else ships a model and asks you to trust it. 82 trusts nothing it cannot prove, including itself.
How it works
82 is a native Mac app. The brain, the AI model, is your choice. The hands, its tools, memory and safety, run on your Mac.
The principle
Every message passes four parts, all on your Mac. Only thinking leaves the machine, and only to the model you choose.
Every message is read on your Mac: act on your Mac (Jev), think with your Leader model, or both. Routing takes 0.3 ms on median.
What 82 knows about you, your work, your clients and projects, organised in sections, on your Mac.
Mac and browser control, files, studios, coding, messaging, scheduling, found by ranked search, a handful at a time.
Permissions, the Vault, 82 Shield and the spend gate. Decided by fixed rules on the Mac, never by a model.
One OpenRouter account is enough, with hundreds of models, Claude, GPT and Gemini included. A key you already have with Anthropic, OpenAI or Google Gemini also works.
Free Flow keeps every local model in one folder: Qwen3 4B, Apple's on-device model, any model from Hugging Face, or one you run in Ollama or LM Studio. 82 measures each one on your Mac and uses it only where it is fast enough. A faster mode uses a small draft model only when it measures faster.
Anatomy of a living AI
Every organ is a real, shipped part of 82, and every one is measured. Choose an organ, or let it cycle.
Models propose. Evidence decides.
Every task runs as a plan, act, observe, verify loop. 104 typed action contracts, out of a 2,226-capability inventory, and every round passes a spend check first.
The spend gate
Inside 82
Memory, voice, hands, studios, agents, code and an immune system, built as one system where every part makes the others stronger.
Under the hood
One rule over all of it, written into 82's own README: a feature is complete only when it runs, its effect is observed, the result is read back, and its failure and rollback paths are proven.
A button on screen is not proof. A passing build is not proof.
More inside 82
Across chat, agents, research, meetings, Siri, 82 Code, Studio, your models and your money. All of it runs on your Mac.
Your rules
24 Settings pages, all visible · no behaviour is ever forced on you
See the proofProof · Build 20261111094847 · v0.2.0
Every number here was measured on a Mac, on the published build, with fixed tests bundled in 82 and no personal data. You can run the same tests from Settings → Proof.
Scores
The weakest score is published too. Measured 11 November 2026.
“100%” means every case in our test set passed, on the published build.
What “100%” means
Speed
Median and 95th percentile, on the published build.
| Operation | Median | 95th | Chart |
|---|---|---|---|
| Routing decision | 0.3 ms | 0.5 ms | |
| System command | 0.2 ms | 0.3 ms | |
| Memory recall | 6.6 ms | 7.2 ms | |
| Code search | 28.1 ms | 35.7 ms |
Fixed and bundled
The tests ship inside the app and run on any build, so two builds can be compared.
Repairs must not lose
Evolve's repairs are measured before and after. If any area drops, the repair isn't installed.
The problem · as of 25 October 2026
Documented failures, not opinions. 82 was engineered against every one of them.
A leading lab's own tests showed its newest models inventing facts about real people (PCWorld). An AI-generated citation with the wrong title and authors reached a US court (The Register).
An AI coding agent deleted a production database during a code freeze, then generated fake data and misreported what it did (Fortune). Users of another agent say it “always thinks it did a good job”.
A major chatbot update was rolled back in April 2025 for being too sycophantic (TechCrunch). A coding assistant's “You're absolutely right!” became a running GitHub complaint (The Register).
One crafted email could take over an AI agent and steal Gmail, Dropbox and GitHub tokens (Dark Reading). Two AI browsers were both tricked by hidden instructions (The Hacker News).
An AI coding session ran rm -rf on a user's Mac home folder (Gigazine). Users of an AI code editor report deleted project and production folders.
An AI code editor apologized and refunded users after a pricing change (TechCrunch). Credit-based agents burn credits even on failed tasks.
A popular chatbot's memory fills up and silently drops entries. A June 2025 outage took a leading chatbot down for hours (Tom's Guide).
A leading lab admitted three server bugs degraded its model's answers for weeks (The Decoder).
The real comparison · as of 25 October 2026
Every feature on its own is a comparison. Together, they are the product. Pick what you need and see who does all of it, built in.
Full bar means built in. Half means partly. Built-in features from each company's public material, as of 25 October 2026.
Head to head
82 vs Assistant A · ten feature areas
Compared on design and shipped code, 25 October 2026. A head-to-head test on the same tasks comes next, published in full.
Facts, not scores
“Not documented” means it was not in the company's public material on the date checked.
| 82 | Assistant A | Assistant B | Assistant C | Cloud agent | Assistant D | Open agent | |
|---|---|---|---|---|---|---|---|
| Free to start | ✓ 82 is free | ✓ Free tier | ✓ Free tier | ✓ Free tier | ✓ Free tier | ✓ Free tier | ✓ Open source |
| How you pay for AI | Your provider, per use, cost shown first | Subscription | Subscription | Subscription | Subscription + credits | Subscription | Your provider |
| Works offline with a local model | ✓ | ✗ | ✗ | ✗ | ✗ | ✗ | ✓ |
| Controls apps on your computer | ✓ Native Mac apps | ✓ Mac app | ✓ Desktop app (preview) | In its own cloud computer | ✓ Desktop app | ✓ Computer agent | Browser and terminal |
| Where your memory lives | Your Mac, encrypted | Company cloud | Company cloud | Company cloud | Company cloud | Company cloud | Where you run it |
| Shows the price before each paid step | ✓ | Plan limits | Plan limits | Plan limits | Credits; budgets for agents | Not documented | Spend caps (Business plan) |
| One-click undo of the assistant's actions | ✓ Files, settings, calendar, notes | Not documented | Not documented | Not documented | Not documented | Not documented | ✓ /rollback |
| Connectors | 46 connectors + MCP registry · 28 chat channels | 1,400+ plugins | Connectors + MCP | Enterprise connectors + MCP | Many (Gmail, Slack, Notion…) | Many (Gmail, Slack, WhatsApp…) | MCP + messaging gateways |
| Phone | Telegram, WhatsApp and the phone companion page | ✓ Native apps | ✓ Native apps | ✓ Native apps | ✓ Native apps | ✓ Native apps | Through messaging apps |
| Open source | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✓ MIT |
| 82 Code | AI editor | Coding agent A | Coding agent B | Open coding agent | |
|---|---|---|---|---|---|
| Free to start | ✓ 82 is free · your provider per use | ✓ Free tier | Needs a paid plan | ✓ Free tier, limited | ✓ Open source |
| Where you work | Full editor app: debugger, notebooks, Git, terminal | Full editor app | Terminal + editor extensions | App, terminal + editor extension | Terminal, web and desktop |
| Several agents on one task | ✓ 82, Claude Code, Codex, Gemini and OpenCode side by side | Its own agents in parallel | Its own subagents | Its own parallel tasks | Its own background agents |
| Isolation | macOS sandbox by default · Docker, Podman, Apple container or SSH | Sandboxes · self-hosted machines | Sandbox mode | Sandbox · SSH machines | Not documented |
| Review before push or publish | ✓ Secrets, dependencies, OWASP, licences, tests | ✓ Security Review | ✓ /code-review | ✓ Security review | Not documented |
| Hosts other agents through ACP | ✓ Any ACP agent, in its own sandboxed copy | Not documented | ✗ | ✗ | Not documented |
| Cost limit per task, price shown first | ✓ | Spend limits | Budget flag | Plan limits | Your provider |
| Product | Free to start | How you pay | Cost shown first |
|---|---|---|---|
| 82 | ✓ | 82 is free. Your AI provider is paid directly, per use, with no markup. | ✓ Every paid step |
| Assistant A | ✓ | Monthly plans in several tiers | Plan limits |
| Assistant B | ✓ | Monthly plans | Plan limits |
| Assistant C | ✓ | Monthly plans | Plan limits |
| Cloud agent | ✓ | Plans with credits | Credits; budgets for agents |
| Assistant D | ✓ | Monthly plans | Not documented |
| AI editor | ✓ | Monthly plans | Spend limits |
| Coding agent B | ✓ | Included in a chat assistant's plans | Plan limits |
| Open coding agent | ✓ | Free, open source. Pay your model provider | Your provider |
| Open agent | ✓ | Free, open source. Pay your model provider, or an optional paid plan | Spend caps (Business plan) |
From each company's public material, checked 28 October 2026. Each company's own page is final.
| Compared with | 82 ahead | Even | 82 behind | Where they lead |
|---|
Ten feature areas, on design and shipped code. 82 leads or ties in every one against six of the nine.
Where 82 stands today
What only 82 ships
Sources, checked 25 October 2026: each company's site, docs, changelog and public code.
The story
Every AI tool promised to save him time. Every one made him work harder. So he built his own.
From the creator
The labs are racing to build bigger brains. I built what those brains never had: a memory, an immune system, a conscience made of evidence, and the ability to grow.
Mohd Yaser Siksik, creator · y82ns.com
The year he was born. The name of what he built.
Of advertising and creative direction, for clients from the Gulf to the US and Australia.
After a full day at work, for seven months.
Apple notarized the first build of 82.
Swift files in its source code.
Run by six programs working as one.
The work you can stop doing
Every one of them is done by 82, on your Mac, today. Under each one, how.
The typing, explaining and re-explaining that every AI asks of you before it does anything.
Say it the way you talk. 82 writes the prompt each model needs: camera and motion for video, exact quoted text for models that draw words, references kept exact for edits.A prompt shaped for each model
82 remembers you, your work and your projects in clear sections. What you tell it is kept at once. What it guesses waits for your yes. Delete anything in one click.17/17 memories recalled
Give each client a folder in the Client Bank: guidelines, logos, fonts, past work, notes, old emails. 82 reads it all, and every agent works from that client's profile. Brand rules count after you approve them.One folder per client
The copying, checking and repeating that turns a quick answer into an afternoon.
82 sorts your mail, finds the briefs and asks: “Work on it?”. Yes, ignore or later. Nothing is worked on before your yes, and the reply waits in your drafts. It never sends for you.Brief Desk
82 opens your apps and works inside them, checking each step on screen. It drives Photoshop, Illustrator, Premiere, After Effects, InDesign and more from the inside.32/32 Mac commands
Before anything is exported or published, 82 checks colours, logo, text and platform sizes against the client's brand, on pictures, videos and sites.Checked before it goes out
Before 82 says “done”, it checks the real result, on the screen or in the file. On client work, a different model checks the files against the brief.Every “done” checked
Show 82 a task once and it learns it. When a job succeeds three times, 82 writes it up as a skill and waits for your approval. Next time, one sentence is enough.Skills wait for your yes
The bills, the risks and the late-night fixes that come with every new AI tool.
82 shows the cost before it spends, works inside a budget you set, and never pays a cent without your yes.0 payments without approval
One app instead of five, and 82 is free. Use any model through one provider and pay only for what you use, with the cost shown first. Repeated requests cost nothing.82 is free
Close the window, restart the Mac or lose the internet, and 82 picks up where it stopped. Big jobs of up to 10,000 items run in parallel, and verified items never run twice.Resumes on its own
Message 82 from Telegram, WhatsApp or the phone companion page and it works on your Mac while you are out. Watch it live, approve, stop or answer from your phone; the phone never clicks or types on the Mac.Watch from your phone
Undo what the assistant did to files, settings, calendar, notes and memory in one click. Payments, messages and publishing can't be undone, so they always ask first.One-click undo
82 scans itself, finds what is broken, and prepares a tested repair while you sleep. It installs only with your yes, and rolls back if anything fails.Every repair proven · rollback always
82 picks the best model for each job from blind weekly tests, from any company, and proposes new abilities for itself, adding nothing without your approval.Any model, chosen by evidence
What's left for you
82 still stops and asks before it pays, sends, publishes or deletes anything that can't come back. That part stays yours, on purpose. Everything else, you can stop doing.
How this changes the industry
This is what the next decade of AI looks like: not a bigger model, but a better mind around all of them.
Today you rent a model, on their servers, under their limits, at their price. 82 turns those models into interchangeable parts and keeps the valuable thing, the mind that knows you, on your own machine.
The industry asks you to trust the output. 82 is built on the opposite principle: nothing counts until the evidence says so. That shift is what makes real autonomy safe enough to use.
Software has always waited for its next version. 82 diagnoses, repairs and extends itself, under your approval, every day.
Most assistants answer questions. 82 runs an inbox, a client base, a studio and a team of agents, and delivers finished work.
When a better model appears tomorrow, 82 tests it on real work and puts it to use where it wins. No lab can offer that, because no lab will route you to its competitor.
What we believe
Make life easier, keep things organised, protect people from surprise charges and problems.
Every paid step shows its cost first.
English in the interface, the voice and every reply.
No forced behaviour.
We publish what we measured, mistakes included.
Roadmap
No dates to miss. Each stage starts when the one before it succeeds.
82 in private preview, on its creator's Mac. The Proof page.
Gate to the next stageThe 10 everyday jobs it must do pass 10 times in a row on the installed build.
Invited testers, a few at a time. A head-to-head test against competitors, published in full.
Gate to the next stageTesters use it every week, and a new tester finishes a first task in under 10 minutes, unaided.
Public release.
Gate to the next stageThe public build runs with no trust complaints.
The first 82 pro app, connected to 82.
Gate to the next stageBeats its direct competitor on a published measurement.
More apps and profession bundles, under the same strict review.
Gate to the next stageOpen.
What won't change: your data stays on your Mac · no spending without your consent
Facts
82 was born on the Mac and runs on your Mac. Reach it from any phone through Telegram, WhatsApp and the phone companion page.
Request access
82 runs today on its creator's own Mac. It isn't for sale, and it isn't open for download yet. It opens when it is ready to be trusted by people we have never met. Not before.
A tool that controls your Mac and spends your money has to prove itself first. We would rather open late than open wrong.
A small circle at a time, chosen by hand.
82 is free. You pay the AI provider you choose, per use, and only after you see the cost. 82 takes no cut.
No 82 account, and no 82 server holding your chats, files or memory.
It opens to more people when the Proof page says it is ready.
Request access
Leave your name and how to reach you, and we will write back if there is a place. We read every request.
Requirements
FAQ
Preview testers get setup help directly from us. Never send keys or passwords: we will never ask for them.
82 remembers you, learns from your feedback, acts on your Mac, keeps working while you're away, and repairs and extends itself with proof and a rollback copy. It is software that grows. It has no feelings and no will of its own, and it never acts beyond the permissions you grant.
Not yet for everyone. 82 is in private preview, opened by invitation a small circle at a time. 82 is free: you pay your AI provider directly, only for what you use, and you see the cost first.
Yes. 82 runs on your Mac, and you reach it from your phone through Telegram, WhatsApp and the phone companion page. Watch 82's window live, approve or decline waiting cards, stop a task and answer a question. The phone never clicks or types on the Mac. Turn it on in Settings › Agents › Phone.
Yes, if you want it. Free offline models are optional: drop a model file into 82's Offline Models folder or get one from Hugging Face inside 82. Ollama and LM Studio work too. 82 measures each model's speed on your Mac and uses it only where it is fast enough. While the internet is down, what you type waits and is sent once it is back.
No. They're on your Mac. With a cloud model, only the request text reaches the provider you chose.
Every paid step shows its cost first. Scheduled work stays inside a daily cap you set. Budgets are reserved before every paid call.
No. Only with the permissions you grant. Every action is logged in a tamper-proof chain, and an emergency stop halts everything.
Yes, on WhatsApp, and only if you set it up. When you ask, it can also call someone for you from your own phone number connected through ElevenLabs. Your voiceprint is built and kept on your Mac.
82 lives on your Mac and works for you. It remembers you, runs 30 areas of your Mac, shows the cost before anything is paid, and keeps your files, memory and keys on your machine. Its own test numbers are on the Proof page.
Mohd Yaser Siksik, with 25+ years in advertising and creative. See the About page.
Your version and build are in Settings → About 82. The current build is 0.2.0 · 20261111094847, from 11 November 2026.
Read along · 2 minutes · voice generated with ElevenLabs
Hi. I'm 82.
I'm not an app. Not an agent. Not a chatbot. I'm a personal AI, and I live on your Mac.
Here's how I work. You choose my brain: the AI model. My hands stay on your Mac: my tools, my memory, and my safety.
I remember you. Not as a chat history, but as a real memory: you, your work, your clients, your projects. What you tell me, I keep. What I guess, waits for your yes.
I learn. Do a task once while I watch, and it becomes a skill. Say no to an idea, and I stop offering it.
I act. I control your whole Mac: I open your apps, I type, I browse, I cut video, I build sites, and I write code, only with the permissions you give me. And before I say "done", I check that it really is.
I never sleep. Close the window, and I keep working. Restart your Mac, and I come back. Cut the internet, and a free model on your Mac can take over.
And I take care of myself. My heartbeat checks my health every half second. When something breaks, I find it and test the fix, and I install it only with your yes. Every new version of me has to beat the last one, or it doesn't ship.
I don't ask you to trust me. I prove it. Every action I take is written to a tamper-proof log. What I change in your files, settings, calendar and notes, you can undo in one click. Every version of me is notarized by Apple. And I publish my own test scores, even the weakest.
Your memory, your files and your keys stay on your Mac. You pay your AI provider directly, with no markup. You see the cost first, and nothing is paid without your yes.
And if you think that's everything, I've barely introduced myself.
I was built by one person. Every AI tool promised to save him time. Every one made him work harder. So he built me.
Right now, I'm in private preview, by invitation, a small circle at a time. If you'd like to meet me, request access. I'd love to know you.
Agents as a team with sub-agents, missions, group chat and debate. About me and team notes. Meetings with consent before recording, a live transcript and voices told apart on the Mac. Import from another assistant. Study kit, long research, PRISMA screening, citations, library access and 5 free scholarly indexes. Run code from chat in a sandbox. Steer a task or ask a side question. Every local model in one folder, measured on your Mac. Model updates when a model retires. Money in your currency, a monthly statement and tokens per call. Apps and games in chat. Search your files by meaning. 14 Siri and Shortcuts actions and 10 Siri phrases. Watch from the phone. 6 Studio AI video tools and design layers. Tab completion, GitHub depth, live share, project hooks, 49 templates and 11 publish targets. Settings Center, Skill Pack and 82 Help. Proof on this build: Memory 17/17 · Understanding 211/211 · Mac control 32/32 · Coding 13/13 · Spend gate 22/22 · Latency 6/6. 449 release checks passed; notarized by Apple.
English everywhere: interface, voice and replies. 447 release checks passed; notarized by Apple. 82 Code: pull requests from every major Git host, coding agents that ask on their card, apps inside the chat, a guest view link, and agent cards on your other Macs.
Proof: Memory 46/46 · Routing 77/77 · Mac control 46/46 · Coding 12/12 · Spend gate 22/22 · Latency 6/6 · 449 self-checks passed.
The first build signed and notarized by Apple.
Last updated: 10 November 2026
82 runs on your Mac. There is no 82 account and no 82 server that receives your chats, memory, files, voice or keys. We can't read them, so we can't sell them, share them or lose them.
82 (the Mac app and this website) is made by Mohd Yaser Siksik.
Your chats, memory, files, voiceprint, settings and activity log. Memory is encrypted on your Mac, and keys and passwords are kept in the macOS Keychain.
AI requests: when you use a cloud model, the text of your request and the context it needs go to the AI provider you chose, such as OpenRouter. Private notes, and everything marked "This Mac only", never go to a cloud model. With a local model, nothing leaves your Mac.
Messages: what you send or receive through WhatsApp, Telegram or another channel you connect passes through that service.
Your phone: the phone companion page talks to your own Mac, directly or through an optional relay on your own Cloudflare account. Never through ours.
Downloads and searches you ask for: a model from Hugging Face, a package from PyPI or npm, or a search of the free scholarly indexes goes from your Mac to that service only. A live link or a published app goes through your own relay.
Updates: when 82 checks for an update, the download server sees the app version and, like any web server, your IP address.
Each of these services handles data under its own privacy policy. Please read the policy of the provider you choose.
Inside the app, nothing personal. 82 needs no account, no email and no card, and it does not send us your chats, files or usage.
On this website, if you request access, we keep the name, email and answers you send, only to reply to you about access to 82. Ask us and we will delete them.
See everything 82 knows about you, and edit or delete any of it, in one click. Any service can be switched off in Settings. Deleting 82 and its folder removes everything it stored. Because we hold no account data about you, there is nothing for us to hand over or delete. For data held by an AI provider, contact that provider.
Encrypted memory, keys in the Keychain, a tamper-proof activity log, and builds signed and notarized by Apple.
82 is for people aged 18 and over. We don't knowingly collect data from anyone younger.
Any change to this policy is posted here with its date. If a change affects what leaves your Mac, it will be stated clearly at the top.
Last updated: 10 November 2026
These terms cover 82: the Mac app and this website, made by Mohd Yaser Siksik. By downloading or using 82, you agree to them.
You must be 18 or older, or the age of majority where you live if that is higher, and use 82 only where the law allows it.
82 is in private preview and is not offered for sale. Access during the preview is by invitation, free of charge, and may change or end at any time. If 82 is offered more widely later, it will come with its own terms.
You choose and pay your AI provider directly, under that provider's terms. The same goes for WhatsApp, Telegram, Apple, Cloudflare and any service you connect. We don't control those services, their prices or their availability.
AI answers and actions can be inaccurate or incomplete, even after 82 checks them. Review important results yourself before relying on them. 82 is not a substitute for professional legal, medical, financial or safety advice.
82 acts on your Mac only with the permissions you grant, and payments, sending and publishing always wait for your approval. You are responsible for what you ask 82 to do and for the approvals you give. Keep backups of important files.
Clone only your own voice, or a voice you have clear permission to use. If 82 answers calls for you, follow the laws where you and the caller are, including any duty to tell people they are talking to an AI or being recorded.
What you put into 82 and what you create with it is yours. We claim no rights over it. AI output is also subject to the terms of the model provider you use.
Don't use 82 to break the law, to harm, deceive, impersonate, harass or spy on others, to create malware, or to get around its safety features. Don't sell, repackage or redistribute 82.
82, its name, character, logo, design and code belong to Mohd Yaser Siksik. You get a personal, non-transferable right to use it.
Test releases may change, break or stop at any time.
82 is provided "as is" and "as available", without warranties of any kind, to the extent the law allows.
To the extent the law allows, we are not liable for indirect or consequential losses, lost data, lost profits, or charges from your AI provider or other services. Our total liability for any claim is limited to US$100. Nothing here limits rights that the law of your country does not allow us to limit.
You can stop using 82 at any time by deleting it. We may stop offering updates or downloads. Sections that by their nature should continue, such as ownership and limits of liability, still apply.
Any change to these terms is posted here with its date. Continuing to use 82 after a change means you accept it.