JW7 retweeted
Big progress on Oracle7 12.0! 🔮 Two weeks of working through a difficult part, and it’s finally coming together. The dedication behind the scenes is what makes this journey worth following. Can’t wait to see what’s next!
@Oracle7vn @Psiloceyeben $ORACLE7 #ORACLE7
JW7 retweeted
Oracle7 is building beyond the usual crypto narrative. The focus is on real utility, practical tools, and technology people can explore and understand.
There’s more to come. 🔮
@Oracle7vn @Psiloceyeben $ORACLE7 #ORACLE7
JW7 retweeted
$ORACLE7 is the kind of technology I could genuinely imagine seeing in the traditional tech market, with the ambition to build something worth hundreds of millions or even billions.
I’ve seen what the team is building, listened to the dev and that’s why I took my position early.
Personally, I see $1M MC as a milestone, not the ceiling.
The market hasn’t fully caught up with the technology yet. Let the builders build. 👀
NFA / DYOR.
JW7 retweeted
I keep adding to my $OPACUS, $ORACLE7 and $MONARK positions at these levels.
The market can shake, sentiment can turn and charts can bleed, but I’m focused on what these teams are building.
I’m not here for every pump. I’m here for the long game.
The market can’t scare conviction out of me. 🦾
NFA / DYOR.
JW7 retweeted
Oracle7 DeepSpace 2.4 🛰️
A spacecraft 23 light-hours from Earth can't wait two days for help. DeepSpace reads its own telemetry, finds the fault, fixes it, and writes down why.
Every test was written down before it ran, then run once (simulation):
✅ 1,683/1,683 fresh faults handled, 0 harmful actions
✅ Stuck sensors and lying power readings: 400/400 each
🛰️ A fleet of 8 spacecraft, Voyager 1 to Europa Clipper: what one learns protects the others (21 times in the sealed week)
🛑 35/35 deliberately wrong rules refused, no spacecraft worse
🪐 Built from NASA's public record of Psyche (no Psyche telemetry): 599/600 propulsion faults handled with no harm
Code + full sealed record, on request:
huggingface.co/psiloceyeben/…
JW7 retweeted
Also taking the time to completely redesign the home page so it’s more primitive and easier to understand the project from the outset
JW7 retweeted
A project shouldn’t lose its direction just because a session ends. Oracle7’s upcoming browser harness is being designed to preserve project progress, track what’s missing, and continue work after reconnecting.
That’s the kind of practical detail that matters. 🔮
@Oracle7vn
JW7 retweeted
The bigger vision for Oracle7 is not just completing tasks, but building reusable abilities from experience and testing them before putting them to work.
Less guesswork, more structured problem-solving. The journey continues. 🔮
@Oracle7vn $ORACLE7 #ORACLE7
Everybody who used claude as LLM knows “You’re absolutely right”.
Now imagine this happens on a space/satellite/defense/aero application.
Data sovereignty, operational independence and no GPU servers or AI-infrastructure matter.
@Oracle7vn
JW7 retweeted
What makes Oracle7 interesting is the focus on verification. A task isn’t finished just because code runs. The actual result must be checked against the requirements.
Building software that can prove its work is a challenge worth watching. 🔮
@Oracle7vn @Psiloceyeben $ORACLE7
JW7 retweeted
Imagine giving a computer a task in plain English and having it turn that request into a working project, test the result, and catch errors before calling it complete.
That’s the direction Oracle7 is building toward with its next stage. 🔮
@Oracle7vn $ORACLE7 #ORACLE7
JW7 retweeted
Just a reminder
All this is on CPU
No language model. No subscriptions.
You own everything.
$ORALCE7
What 12.0 will be able to do in the broswer harness (why it has taken so long for me to get this update out basically)
Starting. On your PC you run python o7_browser.py. It connects to box five, starts Oracle7 there, and opens a chat page in your browser. Before anything else it shows which build is running, so you know exactly what you’re talking to.
Describing a project. You write the request the way you’d say it, for example: “Could you build a report that joins the orders with the customers by customer id, keeps the orders over 100, works out each one’s total as quantity times price, and saves it to a file.” Oracle7 reads it into a numbered list of requirements:
what it should produce: a file, a database, a web page, or printed lines;
which tables it reads, and how they connect: joined on a shared column, stacked like monthly files, or a value looked up in a reference table;
what to do with the rows: filter, total, average, count, group, sort, keep the top few;
new values worked out from columns: times, plus, minus, divided by, or from examples you give;
anything you rule out, such as “no database” or “don’t write a file”.
If a word means nothing to it, it doesn’t guess. It names that word as a gap and tells you what would fill it, for example “say what ‘stale’ means”. Nothing gets written while a gap is open.
Building and checking. Once every requirement is clear, it writes a small standalone Python program, runs it, and checks every requirement against what actually came out: the file’s rows and values, the database contents, the page itself. The check works out the right answer on its own, without trusting the program it just wrote. A program that joins the wrong records is caught even if it has the right number of rows. Only when everything passes does it say done.
Changing it. You correct it the way you’d talk:
“No, I meant over 85” replaces the old value.
“Drop 3” removes a requirement, and it stays gone through later changes.
“Actually, also save the totals by month” adds to the project.
If a correction doesn’t name anything it can change (“make it smarter”), it says so plainly instead of pretending it updated something.
Missing pieces. If a table isn’t there yet, it says which one. Add the file, say “it’s there now”, and it carries on. It doesn’t blindly retry when nothing has changed; it takes a different step instead, like pointing to the exact gap or suggesting the column you probably meant.
Long work and restarts. Long jobs keep running in the background with their progress shown, and “stop” halts them straight away. If the connection drops or the server restarts, nothing is lost: “continue” picks up from your original words. Each project stays tied to its folder, so switching between projects never mixes them up.
Asking about it. You can ask “how does it combine the tables?”, “what’s still missing?”, “is it done?”, “what does it make?” or “what have we tried?”. Oracle7 answers from the project’s saved record without rebuilding anything, and its replies keep done, failed and still-missing clearly apart.
Pages and 3D. You can ask for a web page, including a 3D view (“in 3D as cubes”). Each object stands for one record, sized by its value, and clicking it or its table row selects the same record. The check confirms the view shows the right records.
Growing itself. “Build into yourself: …” makes a new ability from Oracle7’s own records. It’s tested before it goes in, can be rolled back, and afterwards shows up when you ask what it has added. It runs when you ask for it in its own words, and later projects can use it. Functions it works out from your examples are kept and reused too.
Underneath. The original Oracle7 is unchanged: its question-answering head scores 263 right and 0 wrong on the reference set. It has no language model and runs on an ordinary CPU, and every claim it makes is backed by a record.
JW7 retweeted
What 12.0 will be able to do in the broswer harness (why it has taken so long for me to get this update out basically)
Starting. On your PC you run python o7_browser.py. It connects to box five, starts Oracle7 there, and opens a chat page in your browser. Before anything else it shows which build is running, so you know exactly what you’re talking to.
Describing a project. You write the request the way you’d say it, for example: “Could you build a report that joins the orders with the customers by customer id, keeps the orders over 100, works out each one’s total as quantity times price, and saves it to a file.” Oracle7 reads it into a numbered list of requirements:
what it should produce: a file, a database, a web page, or printed lines;
which tables it reads, and how they connect: joined on a shared column, stacked like monthly files, or a value looked up in a reference table;
what to do with the rows: filter, total, average, count, group, sort, keep the top few;
new values worked out from columns: times, plus, minus, divided by, or from examples you give;
anything you rule out, such as “no database” or “don’t write a file”.
If a word means nothing to it, it doesn’t guess. It names that word as a gap and tells you what would fill it, for example “say what ‘stale’ means”. Nothing gets written while a gap is open.
Building and checking. Once every requirement is clear, it writes a small standalone Python program, runs it, and checks every requirement against what actually came out: the file’s rows and values, the database contents, the page itself. The check works out the right answer on its own, without trusting the program it just wrote. A program that joins the wrong records is caught even if it has the right number of rows. Only when everything passes does it say done.
Changing it. You correct it the way you’d talk:
“No, I meant over 85” replaces the old value.
“Drop 3” removes a requirement, and it stays gone through later changes.
“Actually, also save the totals by month” adds to the project.
If a correction doesn’t name anything it can change (“make it smarter”), it says so plainly instead of pretending it updated something.
Missing pieces. If a table isn’t there yet, it says which one. Add the file, say “it’s there now”, and it carries on. It doesn’t blindly retry when nothing has changed; it takes a different step instead, like pointing to the exact gap or suggesting the column you probably meant.
Long work and restarts. Long jobs keep running in the background with their progress shown, and “stop” halts them straight away. If the connection drops or the server restarts, nothing is lost: “continue” picks up from your original words. Each project stays tied to its folder, so switching between projects never mixes them up.
Asking about it. You can ask “how does it combine the tables?”, “what’s still missing?”, “is it done?”, “what does it make?” or “what have we tried?”. Oracle7 answers from the project’s saved record without rebuilding anything, and its replies keep done, failed and still-missing clearly apart.
Pages and 3D. You can ask for a web page, including a 3D view (“in 3D as cubes”). Each object stands for one record, sized by its value, and clicking it or its table row selects the same record. The check confirms the view shows the right records.
Growing itself. “Build into yourself: …” makes a new ability from Oracle7’s own records. It’s tested before it goes in, can be rolled back, and afterwards shows up when you ask what it has added. It runs when you ask for it in its own words, and later projects can use it. Functions it works out from your examples are kept and reused too.
Underneath. The original Oracle7 is unchanged: its question-answering head scores 263 right and 0 wrong on the reference set. It has no language model and runs on an ordinary CPU, and every claim it makes is backed by a record.
JW7 retweeted
The most interesting projects are those building something people can actually explore and use. Oracle7 is experimenting with AI reasoning, autonomous development, and real-world sensor technology. I’m here to watch the progress. 🔮 @Oracle7vn $ORACLE7 #ORACLE7
I hope soon @Oracle7vn can talk with mushrooms too 🍄🟫
Save this post 🫵🏻
From AI reasoning to talking plants, Oracle7 is exploring ideas beyond the usual boundaries of technology. 🌱 The upcoming experiments show just how ambitious this vision is. Keep building, @Oracle7vn! 🔮 $ORACLE7 #ORACLE7
JW7 retweeted
Imagine being able to talk directly to a plant and see how it responds through Oracle7’s sensor technology. 🌱 This is a fascinating experiment, and a live stream where people can interact with the plant would make it even more exciting. Big things ahead for @Oracle7vn🔮
Just an update on the sensor project and the talking plants
I’m going to be moving in the next two weeks and plan on setting up a dedicated place for the experiment in my new place and should be able to get a live stream up as well where people can talk to the plant whenever they would like
It should be interesting very excited to get this part of the project up and running!













