Starting a blackout shuts down running agentic IDEs and kills any active or future chat sessions in seconds.
Scheduled blackouts
Weekends. After midnight. Whenever you keep promising yourself you'll stop. Recurring windows that turn the chatbots off without asking how you feel about it.
Real enforcement
A root daemon sinks DNS for the blocked services. New tab, new browser, app reinstall, reboot — still blocked. Quitting the app doesn't lift it either.
You hold the keys. Sort of.
You choose how hard each block bites — from a nudge you can wave off to a self-ransom that costs real money to break.
How it works
Pick which AI services go dark — the chatbots, the desktop apps, the CLIs — and when.
When a blackout starts, running AI apps are quit after a short countdown — not so you can rush to finish, but so the stop lands as a deliberate breath instead of a jolt. There is never a good time to stop; that's the point. At worst you lose a minute of chat scrollback. For the rest of the window, DNS lookups for the blocked providers are refused at the system level. The daemon remembers its state: restarting the app, or the whole machine, lands you right back in the blackout you scheduled. The only doors out are the ones you chose ahead of time, and they're labeled by severity.
What gets blocked
An ever-growing list of AI providers — Claude, ChatGPT, Cursor, Gemini, and more — reached in the three places they actually live.
Web pages
A root daemon refuses each provider's domains at the DNS level, system-wide. A new tab, another browser, even a reinstall stays dark.
Desktop apps
Apps holding a live AI connection — Cursor, VS Code, the Claude app — are quit when a blackout starts, because an already-open one can keep streaming until it's closed. Reopen it right away if you like: the editor works fine, the AI just stays dark until the blackout ends.
Command-line tools
CLI agents like Claude Code, Codex, and cursor-agent are ended too — the same cut, applied to the terminal.
One mechanism drives all three: a single source-of-truth list, shared by the DNS sink and the app-quit logic, so coverage can't drift between them. When the blackout lifts, your DNS is restored and the apps it closed reopen on their own — you pick up where you left off.
Want the exact list? Every domain, app, and CLI — generated straight from the source so it can't go stale — is on the full blocked-services page. Missing something? The app's “Report a Coverage Gap” button tells us which domain slipped through.
The severity ladder
- Gentle Nudge A speed bump. The block is advisory and you can end it with one click. For people whose problem is momentum, not willpower.
- Hard Stop & Think Blocks for real, but you can write your way out: a short reflection on what you're reaching for and why right now. The questions are chosen deliberately — answering them tends to build the mental pathways that lead away from the chatbox, or at minimum toward a clearer understanding of why you keep ending up there.
- Serious AF Self-Ransom No free exit. The only way out before the timer ends is paying the penalty you set when you armed the block — real, non-refundable money. Negative reinforcement added to the decision matrix at the exact moment you're weighing whether to break your own blackout.
Self-ransom (optional)
The app is free, and most blocks let you stop for free. But some people want an enforcement option with real teeth. Self-ransom is the break-glass version: you pre-agree on an amount you'd pay to lift a block early, for the rare moment you genuinely need to. You never have to use it.
If you do break the glass, the payment goes to the app's maintainers — toward servers and upkeep. That neutral destination is the point: it's what makes pre-committing to it actually hold. How self-ransom works →
The honest version of the pitch
Sure — if you're determined enough, you'll figure out how to get around the blocks in a pinch. But you'll be doing it without LLM tools, so good luck with that. And every minute you spend trying to subvert the rule is a minute you didn't spend with an LLM — which was the goal. The block wins either way.
Known limitations
What LessLLM does and doesn't cover — so you hear it from us, not after.
- macOS only. macOS 14+. No Windows, no iOS, no mobile.
- A determined person can bypass it. It raises friction; it isn't a cage. Encrypted DNS (DoH / “Secure DNS” in Chrome-based browsers), VPNs, and custom proxies can route around the DNS block.
- An already-open connection can finish. A browser tab or app holding a live connection when a blackout starts may keep streaming until it's closed. To cut these, LessLLM also quits known AI apps at blackout start.
- Your own API keys go direct. Tools that stream a model straight from the
provider (e.g. Cursor →
api.anthropic.com) are only fully covered when those providers are blocked too. - New AI domains can lag. Coverage is defined in the app's source; when a service adds a domain, it may need adding. There's a one-click “Report a Coverage Gap” for exactly that.
- Local / on-device models aren't blocked. LessLLM blocks network access to hosted services; a model running on your machine needs no network.
- Managed / corporate Macs with their own DNS or MDM profiles may behave differently.
FAQ
Which services get blocked, how the blocking actually holds, whether you can turn it off, and what it costs — on the FAQ page.
LessLLM is a self-control tool — not medical treatment, therapy, addiction treatment, or crisis support. If you're struggling in a way that affects your wellbeing, please reach out to a qualified professional.