#reminder — Public Fediverse posts
Live and recent posts from across the Fediverse tagged #reminder, aggregated by home.social.
-
#Oldarticle #Oldnewz #Reminder Updated on: June 19, 2017 / 10:03 AM EDT / CBS News 🏙️ 666 Fifth Avenue was once the Kushner Companies’ prized $1.8B asset. Now, debt, vacancies and a looming $1.2B mortgage have put the tower under pressure. Jared Kushner’s financial dealings and ties to potential foreign partners are reportedly drawing scrutiny. 🔍 https://www.cbsnews.com/news/what-you-need-to-know-about-666-fifth-avenue/ #666FifthAvenue #RealEstate #Politics
-
HYDR8 Will Lead You to Water, But Will You Drink?
-
⚠️ #Reminder: Übermorgen (19. August 2026) stehen zwei öffentliche Anhörungen im Petitionsausschuss des #ltnds an:
📄 „Erhalt und Sanierung des Oberdecks der ‚Alten Liebe‘ in Cuxhaven“
⏰ Beginn: 10:00 Uhr📄 „Erhalt und kostendeckende Finanzierung der Dietrich-Bonhoeffer-Klinik (Suchtklinik für Kinder und Jugendliche) in Großenkneten“
⏰ Beginn: 11:15 UhrAnmeldungen für Interessierte sind noch möglich!
👉 Weitere Infos und Anmeldung: https://www.landtag-niedersachsen.de/oeffentliche-anhoerungen-des-petitionsausschusses/ -
While the US arrests dissidents and activists via defamatory imaginary CSAM claims, the government is full of … #reminder #EpsteinFiles
-
#Reminder Information is power — yet corporations like Reed Elsevier & co. have locked up humanity's scientific & cultural heritage. Aaron Swartz's 2008 call for Guerilla Open Access remains urgent: sharing knowledge is a moral imperative, not #piracy Time to liberate research for all. 🌍📚🔓
#OpenAccess #RIP #AaronSwartz #GuerillaOpenAccess #KnowledgeIsPower
-
#Reminder Thu, Aug 28th 2025 German courts are reopening the debate on whether ad blockers infringe copyright—again. Mike Masnick argues the logic is “legal zombie” nonsense, since extensions that alter page display (DOM/CSS) should be about user control—not copyright. 📰🧑💻⚖️ https://www.techdirt.com/2025/08/28/here-we-go-again-german-courts-reopen-the-is-ad-blocking-copyright-infringement-nonsense-we-thought-wed-put-to-bed/ #AdBlocking #Copyright #BrowserFreedom #Germany #Techdirt #AdBlockers #Privacy #Cybersecurity #Mozilla #Germany #PrivacyMatters #AxelSpringerSE #NotOnce #Security #Surveillance
-
#Reminder Published August 19, 2025 Germany’s top court revived a case that could make ad blockers unlawful over alleged copyright violations—potentially threatening user freedom, privacy, and security. Mozilla warns the precedent could hit other extensions too. If it spreads, Germany could become the 2nd country to ban ad blockers (after China). https://www.techradar.com/vpn/vpn-privacy-security/germanys-possible-ad-blocker-ban-could-threaten-user-freedom-and-privacy-says-mozilla 🔍🛡️📵 #AdBlockers #Privacy #Cybersecurity #Mozilla #Germany #PrivacyMatters #AxelSpringerSE #NotOnce #Security #Surveillance
-
#reminder article was published September 19, 2017
🗞️ Remember the time when German publisher Axel Springer and Spiegel Online tried to double down on their anti–ad-blocker fights… after years of legal battles and new site-wide blocks/tests. Springer said, “We are not going to give up.”
https://digiday.com/media/not-going-give-german-publishers-continue-war-ad-blockers/
⚔️📢 #AdBlocking #Media #PrivacyMatters #AxelSpringerSE #NotOnce #Privacy #Security #Surveillance -
#Reminder Sen. Angus King says a simple routine can help: turn the phone off, then back on. The NSA recommends weekly reboots to make hacking harder—especially as “zero-click” attacks can install malware without user action. https://apnews.com/article/technology-government-and-politics-hacking-752db867fafbaba1f9cc34f7588944c5 #cybersecurity #phones #privacy #APTech 🔄📱
-
#Reminder (KT)
Demokratische Parteien reden über die Gefahr der AfD, aber ein Verbotsantrag liegt bis heute nicht auf dem Tisch. Wer ein Verbot für „zu riskant“ hält, muss erklären, warum man dem Risiko für die Demokratie weniger Gewicht gibt. -
#Reminder Activist Checklist was launched in mid-2024 to cut confusion and improve digital security for Palestine solidarity organizing—without requiring anyone to become an expert. Built by experienced activists; guidance updates as surveillance tactics evolve. Subscribe: https://activistchecklist.org 🔒🛡️📱 #DigitalSecurity #PalestineSolidarity #SurveillanceAwareness #MutualAid
#Privacy #Security -
EVA 1.0
Eva Destiny Sapphira — a voice companion that runs on your own computer
Black Sail Studio · Version 1.0 (Beta) · July 2026
Tell her a serial number today. Ask for it back in five months. She’ll have it.Eva Destiny Sapphira 1.0 is a voice AI companion that runs entirely on your own computer. No account, no subscription, no cloud — nothing you say ever leaves your machine. She hears you, answers out loud in a voice you can clone, reads your documents, and remembers your conversations permanently.
Version 1.0 is the release where the memory started really working. Four ways of remembering at once. Exact details kept letter-perfect. Four years of history on one screen.
Free to download. Beta, and honest about it.
EVA is someone you can talk to. Out loud, in your own voice, with a companion who answers in hers — and who remembers the conversation the next time you sit down.
She runs entirely on your own machine. No account, no subscription, no API key. Nothing you say leaves your computer, because there is nowhere else for it to go. The language model, the speech recognition, the voice synthesis and the memory database all live on your hard drive.
This document has two halves. Part 1 is for people who already have EVA and want to know what changed in 1.0. Part 2 is a tour of everything she does, for anyone arriving fresh.
Part 1 — What’s new in 1.0
Version 1.0 is the release where EVA’s memory started actually working. Almost everything below traces back to a single complaint:
“If I tell her I have blue eyes five months ago, I want her to remember today. If I tell her the serial number on a chair, I want to be able to ask for it back.”
Neither of those worked before. Both work now, and the road between the two turned out to run through most of the application.
She remembers vastly more
Previously, EVA could bring roughly eight memories into a conversation, no matter how much you’d told her or how capable your model was. On a model with a large context window, she was using under 1% of the space available to her. She had room for a novel and was being handed a postcard.
EVA now measures how much room your model actually has and divides it up deliberately — memory, past conversations, exact quotes, attached documents, her own thoughts. On a large model that’s roughly 25 times more memory reaching her per turn than before.
What you’ll notice: she connects things. Something you mentioned last month comes up naturally instead of vanishing.
She remembers exact details, exactly
Language models are unreliable at copying. Ask one to remember
CHR-4432-B7and it will usually get it right — and occasionally drop a character, swap two digits, or tidy up the punctuation. It won’t tell you which time it did.For a phone number or a part code, being wrong by one character makes the memory worse than useless.
So EVA no longer asks the model to do it. Before the model sees your words at all, a separate layer reads your sentence directly and stores certain things letter for letter:
- Street addresses
- Phone numbers and email addresses
- Birthdays and anniversaries
- Serial, model, part, order and policy numbers
- Physical details — eye colour, hair, height
- Measurements, medication, vehicles
It keeps the surrounding sentence too, so “the chair serial is CHR-4432-B7″ remembers that it belongs to the chair.
What you’ll notice: you can read a serial number out loud, hang up, come back tomorrow, and ask for it. It comes back exactly.
A note on precision: tested against 443 real conversation turns, this layer produced zero false alarms. It doesn’t grab every number it sees. A year isn’t a birthday unless you say “born” or “birthday” nearby; a bare number isn’t a serial without a word like “serial” or “model” beside it.
She remembers your actual words, not just a summary
EVA has always kept a word-for-word record of every conversation. Until now, nothing could search it — the transcript went onto the disk and was never read back.
Now it’s fully searchable, and it works even for questions that share no words with the answer. “What’s my address?” has nothing in common with “I live at 412 Rowan Street” — no shared words at all — so EVA searches for address-shaped text instead of matching words.
When her summary of something and your actual words disagree, your words win.
Four different ways of remembering, at once
The reason her recall improved so sharply isn’t one clever trick. It’s that EVA now searches her memory four separate ways every time you speak, and combines the results.
Each method fails at something the others handle:
How she searchesWhat it’s good atWhat it can’t doBy meaning“Tell me about my job” finding a memory about workSerial numbers — they have no meaning to matchBy exact wordsSerial numbers, names, part codes“What’s my address” — shares no words with the answerBy who people are“My dad” finding a memory that only says “Ed”Anything not about a personBy when“What did we discuss Thursday afternoon?”Nothing — this question has no topic words at allA memory found by two methods ranks above one found by a single method, because agreement between different approaches is a good sign.
That last row is worth dwelling on. “What did we talk about on Thursday?” contains no subject matter whatsoever. There is nothing for a conventional search to grab. It needed its own method.
She doesn’t remember nonsense
A companion who believes everything she hears — including her own guesses — turns into a confident fabricator within a week. Her own speculation gets remembered, comes back as context, gets built on, and hardens into fact.
EVA ranks her memories by where they came from:
- Highest — you asked her to remember it
- Something you said, that she noted
- Web research, or something only she claimed
- Lowest — her own idle thoughts
Lower can never overwrite higher. And after each conversation, a review pass checks each new memory against the transcript, keeping only what’s actually supported.
You can see this working. In testing, 8 of 9 things EVA had concluded about herself — “I am operating at full capacity” — were correctly held back from recall. Kept on file, never treated as fact.
Nothing is ever deleted. Older memories simply compete harder for her attention. A fact from five months ago is still there at full detail — which is exactly why “she forgot my eye colour” turned out to be a question of attention, not storage. The information had never gone anywhere.
If she conflicts with herself, she resolves it
Tell her the model number is one thing, then later correct it, and the new one replaces the old rather than sitting alongside it. Two competing serial numbers would otherwise both survive forever, taking turns being wrong.
Repeating something you’ve already said doesn’t create a duplicate — it makes the existing memory stronger.
She can look things up mid-sentence
Previously, EVA had to guess what would be relevant before you finished speaking, gather it, and hope. If the guess was wrong, that was that.
She can now search again while she’s answering — through her memory, past conversations, or an attached document — when your question needs something the first pass missed.
This is off by default. It adds a pause before she answers, and smaller models handle it unevenly. Turn it on in Agent Profile if you want it.
Documents stay with her
Attach a document and EVA reads it, indexes it, and can answer questions about it. Three things changed:
Whole books now work. There used to be a hidden ceiling that quietly stopped at roughly a quarter of a full-length book. You got no warning — she simply had no idea the rest existed. That limit is gone.
Documents survive hanging up. They used to be wiped when the call ended. Attach something once and it stays.
You can now remove them. There’s a DOCUMENTS panel under the Memory tab listing everything she’s holding, with three options:
- Deactivate — she stops using it, but keeps the indexing. Reactivating is instant.
- Reactivate — bring it back.
- Remove — delete it completely.
Re-attaching something she already has won’t duplicate it. And removing a document doesn’t erase your conversations about it — those are memories of talking to you, and they stay.
She speaks as she thinks
Her words now appear as she generates them, rather than arriving all at once when she’s finished. Long answers feel like a conversation instead of a loading screen.
Long answers actually finish
There was a genuinely bad bug here. Beyond a certain length, her speech was cut off mid-thought without any warning — a long reply lost roughly its final two thirds. The audio also degraded as it approached that ceiling, which is why long answers sounded distorted. Two problems wearing one coat.
Her speech is now split at natural sentence breaks and spoken in sequence, with the next piece being prepared while the current one plays. Length no longer affects how long you wait for her to start, and nothing gets silently discarded.
Her answers are longer
Her reply length was capped at roughly 220 words, and her default instructions asked her to stay under 120. Both of those existed because of the audio problem above — and that problem is fixed.
The default is now more than three times longer, and the instruction to keep it short is gone. She’ll actually explain things.
If you had set your own reply length or written your own instructions for her, your settings are kept. The update only changes values you never touched.
She stops interrupting herself
The most annoying bug in the previous release: EVA would start talking, hear her own voice through your microphone, decide someone was interrupting her, and stop mid-sentence.
She’s always subtracted her own voice from what the microphone hears. The problem was that she calibrated this once, quietly, at the start of speaking — so a phrase that started soft and grew louder would exceed her own estimate and read as somebody else talking.
She now tracks how loud she’s currently being and learns how much of her voice your room actually returns to the microphone.
In simulation, on the scenario that broke it — quiet start, louder later — she interrupted herself 139 times before this fix and zero times after.
The old behaviour failed in both directions, which is the part that took longest to see. It cut her off when she got louder, and it made her impossible to interrupt while she was loud. Both are fixed.
She waits for you to finish
EVA used to allow the same pause after everything you said — two words or two minutes.
She now waits longer after a long turn. Someone who’s been talking at length is more likely to be mid-thought than someone who said “yes”. The base pause is adjustable in Call Options, from a snappy 0.3 seconds to a patient 2 seconds.
You can interrupt her — two ways
Speak. Talk over her and she stops and listens.
Or type. Press Enter while she’s talking and she stops immediately and responds to what you typed. This works even with voice interruption switched off — that setting exists to protect against microphone noise, and a keypress is unambiguous.
She works without a microphone
Previously a missing, busy or unavailable microphone meant a fatal error dialog and no call at all. EVA now starts anyway, as a text conversation, and tells you what she found.
Thinking models work now
If you use a reasoning model — one that works through a problem before answering — her background thinking was silently broken. It would fill its entire allowance thinking and produce nothing visible, which looked identical to a crash.
The cause is worth knowing if you use these models: they spend around 700 words thinking before their first word of output, against an allowance of 180. She now notices this and retries with more room, and her allowance defaults high enough that it rarely comes up.
You can also now pick a separate, smaller model for her background thinking — which is a setting that existed before, was displayed in the interface, and did nothing at all. It works now.
Her memory timeline is navigable
The Journal Matrix shows your history with EVA as a map you can move around.
It used to become unusable after about a month — a year of daily conversation was 23 screens wide, and the “fit everything” button silently stopped fitting anything.
It’s now laid out on a real calendar, with a ruler that adapts from years down to hours. Long quiet stretches fold up into a marked gap — “⋯ 3 months quiet ⋯” — so a break doesn’t cost you a mile of empty space. There are span buttons for day, week, month, quarter, year and all time.
Four years of history now fits on one screen and draws in a twenty-fifth of a second.
Polish
A startup screen. Loading is genuinely slow the first time — several hundred megabytes of speech and language machinery unpacking. There’s now a progress screen with her logo that tells you what’s loading, weighted so the bar moves at a rate matching reality rather than racing to 80% and hanging.
A proper icon in your taskbar, window and Alt-Tab, instead of a generic Python icon. She pins correctly now.
A cleaner window. The interface header was removed; the tab strip sits at the top.
The About tab tells the truth. It used to report “Primary model not selected” while displaying your selected model a few inches away. It now shows what’s running, and distinguishes a model that’s active from one that’s chosen and starts with your next call.
Part 2 — What EVA can do
For anyone arriving fresh.
Everything runs on your computer
- Her mind — a language model running locally through LM Studio
- Her ears — local speech recognition
- Her voice — local speech synthesis, with voice cloning from a sample
- Her memory — a database file on your hard drive
No account. No subscription. No internet required for a conversation. Nothing you say is transmitted anywhere.
The only optional exception is web search, which is off unless you supply your own key and switch it on.
Memory that lasts
The point of the whole project.
EVA keeps three separate kinds of record, and keeping them apart is what stops her memory turning to mush:
- Memory — what she knows to be true about you
- The journal — what was actually said, word for word, permanently
- Reasoning — what she’s still uncertain about
She searches all of it four different ways at once. She keeps important details letter-perfect. She knows where each belief came from and won’t let a guess overwrite something you told her. And she never deletes anything.
Talking to her
Press a button and talk. She hears you, waits for you to finish, answers out loud, and her words appear on screen as she speaks them.
Interrupt her by speaking or by typing. Adjust how patient she is. Use her without a microphone if you don’t have one. Her voice can be cloned from a sample you provide.
Documents
Attach a document — a book, a manual, a contract — and ask her about it. She reads it properly, keeps it across sessions, and can pull the surrounding context so a fact split across two pages arrives whole. Deactivate one temporarily or remove it entirely, from a panel that shows you everything she’s holding.
She thinks when you’re not talking
When the room is quiet, EVA reviews what she knows and forms a thought about it. She deliberately favours memories she hasn’t used recently, so old things resurface rather than the same few facts circling.
If a thought seems worth your attention and you’ve been quiet a while, she’ll speak first. How willing she is to do this is entirely yours to set — how long she waits, how many thoughts she needs first, and how interesting a thought has to be. Below the threshold, she keeps it to herself.
She will never volunteer your address or phone number this way. Sensitive details are excluded from her idle thinking on purpose.
You can read everything she’s thought in the Thoughts and Thought Stream tabs.
She keeps track of what’s uncertain
The Reasoning tab is not a chat log. It’s a structure she maintains: competing explanations, evidence attached to each with a direction and a weight, assumptions she’s making, and — the useful part — what would tell the difference between two explanations that both still fit.
It’s deliberately conservative:
- Hearing the same thing three times counts once, not three times
- Evidence against something weighs more than evidence for it
- She may only propose changes to this structure — they’re checked and applied separately, so she can’t quietly rewrite her own conclusions
The confidence figure is labelled honestly in the interface as a heuristic, not a probability. It’s a rough indication, and it says so.
Making her yours
Agent profiles hold a complete personality: her instructions, her model, her voice, her portrait, how long her answers run, how creative she is, and every memory and behaviour setting. Keep several and switch between them.
Themes restyle the whole interface.
Diagnostics shows you what’s actually happening — what she retrieved, what fit in her context, what the model returned.
What you should know
Honest notes. EVA is beta software and this is where the sharp edges are.
Her memory file isn’t encrypted.
memory.dbis a plain database on your disk. It holds your conversations and — since detail capture is on by default — addresses, phone numbers and birthdays if you’ve mentioned them. Anyone with access to your computer or your backups can read it. If that’s a concern, switch off “Remember personal details” in Agent Profile, or use full-disk encryption. At-rest encryption is not yet implemented, and this is the most significant open item in the release.There’s no screen for reviewing sensitive details. She marks them and handles them carefully, but you can’t currently browse or revoke them individually.
She’s only as sharp as the model you give her. Everything above is scaffolding around a language model. On a small one she’ll still occasionally misread something she retrieved perfectly. Better memory can’t fix reasoning.
Confidence figures are estimates. The interface says so where they appear.
Mid-answer search is untested with real models. That’s why it’s off by default.
Web search is off unless you supply a key and enable it. Everything else needs no internet at all.
Checking she’s working
After starting a call:
- Say “the chair serial is CHR-4432-B7.”
- Hang up. Start a new call.
- Ask “what’s the chair serial?”
It should come back exactly. If it does, the memory system is doing its job.
EVA 1.0 (Beta) · Black Sail Studio
Beta means it works and it’s still moving. Keep backups of your EVA Data folder — it holds your conversations, and they’re the part that can’t be reinstalled.
#ai #journal #reminder #writing -
Somehow they survive
Mom's ancient Speedy Henry
Her green reminder#Haiku #OneHaikuADay #WritersCollective #writingcommunity #BackToHaiku #Summer #July #Flower #Flowers #Flora #SpeedyHenry #Inch #green #Flags #Ancient #Mom #Reminder #Life #Art #Akron #Ohio #USA #Photography #2026
July 25, 2026 -
Der Rechtsextremismus hat Krieg, Zerstörung, Vertreibung und Wahnsinn gebracht.
-
#Reminder nicht das erste Mal
#Paywall ⚠️: Ein muslimischer Wissenschaftler in Sachsen geriet laut seinem Anwalt ins Visier – er verlor Jobs, zeitweise sogar Hartz IV. Im Raum steht, dass der Verfassungsschutz eine Forscherkarriere sabotierte. 🕵️♂️⚙️ https://www.spiegel.de/panorama/justiz/sachsen-wie-der-nachrichtendienst-eine-forscherkarriere-sabotierte-a-00000000-0002-0001-0000-000174691223 #Sachsen #Rechtsstaat #Verfassungsschutz #Fehlidentifikation
-
#Reminder (KT)
Demokratische Parteien reden über die Gefahr der AfD, aber ein Verbotsantrag liegt bis heute nicht auf dem Tisch. Wer ein Verbot für „zu riskant“ hält, muss erklären, warum man dem Risiko für die Demokratie weniger Gewicht gibt.
-
Saudi Arabia and nuclear-armed Pakistan signed a mutual defence pact after Israel launched an attack on Qatar targeting Hamas officials. Pakistan’s defence minister then said his nation’s nuclear programme “will be made available” to Saudi Arabia if needed, something seen as a warning for Israel, long believed to be the Middle East’s only nuclear-armed state.
-
#Reminder Privacy groups and major privacy tech firms urged UK ministers to “hands off” VPNs, warning that age-gating could weaken security while barely stopping teens from dodging social media bans. Mozilla, Proton, Tor and others back an open letter. 🔒📡 #VPN #Privacy #OnlineSafety https://www.theregister.com/networks/2026/07/14/hands-off-our-vpns-privacy-groups-tell-uk-ministers/5271070
-
#Reminder #Siemens AG faces growing scrutiny over reports of defense-linked software licensing in #Israel and involvement in the EU-backed interconnector to Israel’s grid. Siemens says it has been active there for decades. Learn more: https://irbak-action.com/company/92ec3be0-769b-467f-92c0-7e72c5061a39 ⚡️🔎📰 #Boycott #Siemens #HumanRights #Israel
-
#Reminder a spotlight falls on Canada’s Canadian Commercial Corporation (CCC): reports say it signed a U.S. DoD amendment in Sept. 2024 worth up to CAD $78.8M for M31A2 triple-base propellants, with Israel listed as the end-user. Critics demand transparency. 👀⚖️🇨🇦 #ArmsTrade #Transparency https://irbak-action.com/company/5a192e92-6bf5-4365-bbe8-bff84175fc15
#Boycott -
#Reminder reports say #Dell Technologies won an #Israeli MoD server tender in Jan 2023 (>$150M, US foreign aid) and sponsored “IT for IDF” in July 2024. 🖥️ Its EMC Israel unit is also reported to have secured a $300M storage tender (2010). Some are calling for procurement & investment review. https://irbak-action.com/company/9c26f035-b8e3-45db-86b3-c35774aab118 #TechAccountability #HumanRights #Boycott
-
#Reminder #UN #HumanRights Office update (26 Sep 2025) lists #Airbnb Inc. in a database of businesses linked to illegal #Israeli settlement activities. 📄 Advocates call for #boycotts of ABNB-linked products and investor pressure. More: https://irbak-action.com/company/5efea0ab-5f9a-451e-9747-538082b40851 🔗 #Airbnb #ABNB #HumanRights #Settlements #Boycott
-
#Reminder Investigations say #Cisco Systems supplies networking and servers used by Israel’s military and police, including Defense Ministry tenders. Activists are calling for a boycott of Cisco products and CSCO stock. 📢🚫 https://irbak-action.com/company/527b238e-a910-4df0-9984-13471718dad5 #Boycott
-
#Reminder Investigative outlets report that #Amazon Web Services (#AWS) has close ties to Israel’s defense apparatus, including a 2021 cloud-infrastructure deal and clauses limiting service suspension. 📡⚖️ Activists are calling on investors and customers to boycott Amazon and demand accountability. More: https://irbak-action.com/company/a6392317-2472-4a60-9823-9b472e2204df #BoycottAmazon #TechAccountability #Palestine #Boycott
-
#Reminder #Microsoft (MSFT) is drawing renewed scrutiny after a 2025 OHCHR report alleged its technologies have been used to support #Israel’s security apparatus, including claims tied to Azure in #Gaza. 🧠☁️📡⚖️ Activists are calling for #Boycott—see more: https://irbak-action.com/company/24be48bf-116c-402f-870b-666e72c5c3c2 #HumanRights #Accountability #Boycott