12.0.2 HAS BEEN RELEASED!
Oracle7 12.0 is out (release 12.0.2).
From a head you ask to one you work with, in one chat:
- Looks up what it does not hold, and keeps it with its source.
- Explains a topic to the size you ask, every sentence cited.
- Builds programs from plain words: 10 of 10 on a sealed test.
- Repairs your code until your own tests pass, and audits servers.
- Carries projects through: 23 of 23 checked by what actually ran.
- New: the Oracle7 Hybrid Game Engine, Developer Preview.
- Nothing lost: 263 right, 0 wrong on the 600-question reference.
No language model. No GPU. Every answer proved, cited or declined.
https://t.co/Dg7LVbyJJb
Oracle7 12.0 is released, as 12.0.2.
Oracle7 is a new kind of geometric-symbolic compiler made of composable parts. There is no language model inside it. It runs on an ordinary CPU, with no GPU and no training, and every answer it gives is proved, cited or declined.
Summary: 11.0.1 was a head you asked. You got an answer from what it could check, a citation, or a decline. 12.0 keeps all of that, with the reference score unchanged, and becomes one you work with. Through one chat, in your terminal or as a clean page in your browser, you can talk to it, ask it, and give it work: it looks up what it does not hold, writes explanations from cited sources, builds programs from plain descriptions, repairs your code until your own tests pass, carries a project through to what actually ran, and looks at its own work. 12.0.2 is that release after one more round: we installed it clean, used it the way a first-time user would, and repaired every class of fault we found.
What 12.0 adds
- One chat door, in the terminal or a browser page, with conversations, replies bound to your own words, and /why on any reply to show how it was reached.
- Look-up: what it does not hold, it says so and offers to read Wikipedia; two sources agreeing is an answer, one source is offered with its citation, and what it learns is kept for next time.
- Writing: it explains a topic in cited paragraphs to the size you ask, and writes new sentences from claims it holds, each read back before it is said.
- Building: describe a program in plain words and it builds, runs and checks it; state a goal and it plans the steps backward.
- Your code and servers: it opens your folder or connects to your server, repairs code until your own tests pass, audits a server with the lines that show each finding, and changes nothing without your go.
- Projects: a project in plain words becomes requirements, each judged by what actually ran; several tables, pages and 3D views; it survives a restart, and "where are we?" is answered from the record.
- Itself: it reflects on its own work and can propose a change to itself, which an outside judge must pass, keeping the old version to roll back to.
- The harness: a Satellites tab, where each satellite runs its own exam, and a Machines tab, read-only over your own keys.
What 12.0.2 adds to that
- An explanation asked as a question is read from the page about its subject: 12 of 12, up from 3.
- Turns reach the part that does them, and a request it cannot write offers to read instead.
- A repair is judged by the tests that bear on it, and a timed check stops everything it started.
- A seventh satellite: the Oracle7 Hybrid Game Engine, Developer Preview.
Measured
- 600-question reference through the chat door: 263 right, 0 wrong, the same as 11.0.1. Nothing it could already do was lost.
- Work carried through, judged by what happened: 23 of 23 outcome checks through the browser page; 16 of 16 carried across a restart.
- Building programs from a description, sealed test run once: 10 of 10.
- Looking up 10 questions it had never seen, sealed: 0 answered wrongly.
- First use on a clean install, 50 turns frozen before the work: 49 of 49 (31 on 12.0.1).
- Both clean installs verify 402 of 402 parts against the roots anchored on Solana.
Slim (commercial use): https://t.co/Dg7LVbyJJb
Full (research): https://t.co/K2XGpxD4jU
Anchored on Solana: https://t.co/XGWtSJysn3
Yeah I think this will really help to properly showcase oracle7 and what the agents are actually capable of the institute is called the OUROBOROS research project and the world is a research campus with 60 different research teams to start off with nearly 500 researches on campus as well as then a whole economy with workers and homes for them as well
Basically building an immersive world based around the research institute with workers and activities and leisure time and individual personalities so that it can be properly shown off that it can be shown off that oracle7 can run entire cities, societies and worlds all on the oracle7 base plus the wander around game engine
Designing and building an entire autonomous world has never been cheaper to do or run
Planning on building a real 3d institute where the agents work and have offices using the wander around game engine that I built you’ll be able to explore around and walk up and ask the agents about the work that they are doing with even different wings basically gonna build a university/research campus out for the agents to explore and inhabit and live there while they research
The goal is to build a world that can generate itself indefinitely based upon the implied needs of the existing world and what is possible based on that as a fact of the reality of the world
I have never been interested in simulating things
If we aren’t building forms that emulate reality as a principle that generates itself indefinitely based upon its parts then we are just wasting time ducking around with questions that don’t help us at all at the end of the day
Planning on building a real 3d institute where the agents work and have offices using the wander around game engine that I built you’ll be able to explore around and walk up and ask the agents about the work that they are doing with even different wings basically gonna build a university/research campus out for the agents to explore and inhabit and live there while they research
Yeah this is gonna be so cool
I’ve had a vision for what I want the eventual irl research campus to look like gonna get it built and running so yall can explore
I’m really out here cooking hard behind the scenes
I’ve got list of nearly 100 research projects with teams of agents than I can start simultaneously once I build a few more parts
@prometheus7even is a research institute and this may finally allow yall to understand the vision and what the goal is here
Oracle7 can already do real research on its own as a specialist, going to have near a hundred teams of specialists all running together on cpu and with the ability to collaborate and know what the other teams are working on
🫡
The work continues on 12.5 is already building and then we just keep on going
Got some big things coming I’m just glad I was finally able to get this update out and with the game harness integrated still haven’t even gotten the 1.5 version of @WanderAroundio out which is powered $ORACLE7 and was built using the game engine that I released last night as a standalone repo and integrated into the harness
Lots to do!!
Oracle7: the road forwards
Oracle7 is a new kind of geometric-symbolic compiler made of composable parts. There is no language model inside it. It runs on an ordinary CPU, with no GPU and no training, and every answer it gives is proved, cited or declined. The trailer walks the whole road we are building it along: where it has been, where it is now, and where it goes next. Here is that road in words.
Where it has been
10.0, 26 October: a mind you can check. Every answer either proved, worked out in steps and checked; cited, read from a named record with its source; or declined, kept as a hole to fill rather than guessed.
11.0.1, 5 November: parts you can add. A part format in which every new piece of knowledge or skill arrives with its own exam, and is admitted only if it passes. On the 600-question reference: 263 right, 0 wrong.
12.0, 18 November: from answering to doing the work. One chat, in your terminal or browser, where Oracle7 looks up what it does not hold, explains topics from cited sources, builds programs from plain descriptions, repairs your code until your own tests pass, and carries projects through to what actually ran. The reference score stayed at 263 right and 0 wrong.
12.0.2, 19 November: the release after one more round of first-use testing, 49 of 49 on a sweep that 12.0.1 passed 31 of, and the seventh satellite: the Oracle7 Hybrid Game Engine, a developer preview, open source.
Where it is now
12.5 is the language core. A word, in Oracle7, is not a label for knowledge stored somewhere else: it is its definition, made of other words that are themselves definitions. Explaining something is unfolding those definitions until they meet. The reading of definitions and the meaning across words already work under independent judging: on the last fresh round, 26 of 30 opposites and 40 of 60 related pairs were found, with no false links between 60 unrelated pairs. What remains is the hardest part: answering why and how, by running what words do on a situation, and judging it all against an independent acceptance in conversation, writing and building.
Where it goes next
13.0, the laboratory, and receipts. Bring a dataset: it proposes hypotheses, registers them before reading the data, tests them, and reproduces each a second way. And every answer and check carries a receipt anyone can re-verify from the receipt alone, offline.
13.5, the first real plant. The Slow Fire kit on real sensors, holding its bars through a season.
14.0, engineering. A solar setup, a beam, a circuit or a pump designed to stated specifications, every design checked by independent open-source simulators.
15.0, self-sustaining growth. Most new abilities come from Oracle7's own growth loop, with the cost of each verified ability falling across releases.
15.5, off-grid. The whole checked system on a phone-class board, offline.
16.5, every subject. Whole courses built from verified answers and lessons.
17.5, your library. Your own documents and measurements become private parts on your own machine, checked the same way, with nothing sent anywhere.
18.0, the long ledger. Any answer replayed against what was known when it was given, across every release.
18.5, check anything. Paste an article or an invoice and every fact and number comes back checked, wrong or unknown, each with its source.
19.5, everywhere. The same checks in a browser extension, a spreadsheet function and a phone keyboard, all calling your local install.
How each step is judged
Every stage on this road ships only when it passes a test written down before the work began, with its failures published. That is why the dates are not on the planned stages: each one is done when its proof passes, not when a calendar says so.
Oracle7 12.0.2: https://t.co/Dg7LVbyJJb
Game engine: https://t.co/BudWrks5ko
https://t.co/zDXSqSiEN7
12.0.2 HAS BEEN RELEASED!
Oracle7 12.0 is out (release 12.0.2).
From a head you ask to one you work with, in one chat:
- Looks up what it does not hold, and keeps it with its source.
- Explains a topic to the size you ask, every sentence cited.
- Builds programs from plain words: 10 of 10 on a sealed test.
- Repairs your code until your own tests pass, and audits servers.
- Carries projects through: 23 of 23 checked by what actually ran.
- New: the Oracle7 Hybrid Game Engine, Developer Preview.
- Nothing lost: 263 right, 0 wrong on the 600-question reference.
No language model. No GPU. Every answer proved, cited or declined.
https://t.co/Dg7LVbyJJb
Oracle7 12.0 is released, as 12.0.2.
Oracle7 is a new kind of geometric-symbolic compiler made of composable parts. There is no language model inside it. It runs on an ordinary CPU, with no GPU and no training, and every answer it gives is proved, cited or declined.
Summary: 11.0.1 was a head you asked. You got an answer from what it could check, a citation, or a decline. 12.0 keeps all of that, with the reference score unchanged, and becomes one you work with. Through one chat, in your terminal or as a clean page in your browser, you can talk to it, ask it, and give it work: it looks up what it does not hold, writes explanations from cited sources, builds programs from plain descriptions, repairs your code until your own tests pass, carries a project through to what actually ran, and looks at its own work. 12.0.2 is that release after one more round: we installed it clean, used it the way a first-time user would, and repaired every class of fault we found.
What 12.0 adds
- One chat door, in the terminal or a browser page, with conversations, replies bound to your own words, and /why on any reply to show how it was reached.
- Look-up: what it does not hold, it says so and offers to read Wikipedia; two sources agreeing is an answer, one source is offered with its citation, and what it learns is kept for next time.
- Writing: it explains a topic in cited paragraphs to the size you ask, and writes new sentences from claims it holds, each read back before it is said.
- Building: describe a program in plain words and it builds, runs and checks it; state a goal and it plans the steps backward.
- Your code and servers: it opens your folder or connects to your server, repairs code until your own tests pass, audits a server with the lines that show each finding, and changes nothing without your go.
- Projects: a project in plain words becomes requirements, each judged by what actually ran; several tables, pages and 3D views; it survives a restart, and "where are we?" is answered from the record.
- Itself: it reflects on its own work and can propose a change to itself, which an outside judge must pass, keeping the old version to roll back to.
- The harness: a Satellites tab, where each satellite runs its own exam, and a Machines tab, read-only over your own keys.
What 12.0.2 adds to that
- An explanation asked as a question is read from the page about its subject: 12 of 12, up from 3.
- Turns reach the part that does them, and a request it cannot write offers to read instead.
- A repair is judged by the tests that bear on it, and a timed check stops everything it started.
- A seventh satellite: the Oracle7 Hybrid Game Engine, Developer Preview.
Measured
- 600-question reference through the chat door: 263 right, 0 wrong, the same as 11.0.1. Nothing it could already do was lost.
- Work carried through, judged by what happened: 23 of 23 outcome checks through the browser page; 16 of 16 carried across a restart.
- Building programs from a description, sealed test run once: 10 of 10.
- Looking up 10 questions it had never seen, sealed: 0 answered wrongly.
- First use on a clean install, 50 turns frozen before the work: 49 of 49 (31 on 12.0.1).
- Both clean installs verify 402 of 402 parts against the roots anchored on Solana.
Slim (commercial use): https://t.co/Dg7LVbyJJb
Full (research): https://t.co/K2XGpxD4jU
Anchored on Solana: https://t.co/XGWtSJysn3
Oracle7: the road forwards
Oracle7 is a new kind of geometric-symbolic compiler made of composable parts. There is no language model inside it. It runs on an ordinary CPU, with no GPU and no training, and every answer it gives is proved, cited or declined. The trailer walks the whole road we are building it along: where it has been, where it is now, and where it goes next. Here is that road in words.
Where it has been
10.0, 26 October: a mind you can check. Every answer either proved, worked out in steps and checked; cited, read from a named record with its source; or declined, kept as a hole to fill rather than guessed.
11.0.1, 5 November: parts you can add. A part format in which every new piece of knowledge or skill arrives with its own exam, and is admitted only if it passes. On the 600-question reference: 263 right, 0 wrong.
12.0, 18 November: from answering to doing the work. One chat, in your terminal or browser, where Oracle7 looks up what it does not hold, explains topics from cited sources, builds programs from plain descriptions, repairs your code until your own tests pass, and carries projects through to what actually ran. The reference score stayed at 263 right and 0 wrong.
12.0.2, 19 November: the release after one more round of first-use testing, 49 of 49 on a sweep that 12.0.1 passed 31 of, and the seventh satellite: the Oracle7 Hybrid Game Engine, a developer preview, open source.
Where it is now
12.5 is the language core. A word, in Oracle7, is not a label for knowledge stored somewhere else: it is its definition, made of other words that are themselves definitions. Explaining something is unfolding those definitions until they meet. The reading of definitions and the meaning across words already work under independent judging: on the last fresh round, 26 of 30 opposites and 40 of 60 related pairs were found, with no false links between 60 unrelated pairs. What remains is the hardest part: answering why and how, by running what words do on a situation, and judging it all against an independent acceptance in conversation, writing and building.
Where it goes next
13.0, the laboratory, and receipts. Bring a dataset: it proposes hypotheses, registers them before reading the data, tests them, and reproduces each a second way. And every answer and check carries a receipt anyone can re-verify from the receipt alone, offline.
13.5, the first real plant. The Slow Fire kit on real sensors, holding its bars through a season.
14.0, engineering. A solar setup, a beam, a circuit or a pump designed to stated specifications, every design checked by independent open-source simulators.
15.0, self-sustaining growth. Most new abilities come from Oracle7's own growth loop, with the cost of each verified ability falling across releases.
15.5, off-grid. The whole checked system on a phone-class board, offline.
16.5, every subject. Whole courses built from verified answers and lessons.
17.5, your library. Your own documents and measurements become private parts on your own machine, checked the same way, with nothing sent anywhere.
18.0, the long ledger. Any answer replayed against what was known when it was given, across every release.
18.5, check anything. Paste an article or an invoice and every fact and number comes back checked, wrong or unknown, each with its source.
19.5, everywhere. The same checks in a browser extension, a spreadsheet function and a phone keyboard, all calling your local install.
How each step is judged
Every stage on this road ships only when it passes a test written down before the work began, with its failures published. That is why the dates are not on the planned stages: each one is done when its proof passes, not when a calendar says so.
Oracle7 12.0.2: https://t.co/Dg7LVbyJJb
Game engine: https://t.co/BudWrks5ko
https://t.co/zDXSqSiEN7
Oracle7: the road forwards
Oracle7 is a new kind of geometric-symbolic compiler made of composable parts. There is no language model inside it. It runs on an ordinary CPU, with no GPU and no training, and every answer it gives is proved, cited or declined. The trailer walks the whole road we are building it along: where it has been, where it is now, and where it goes next. Here is that road in words.
Where it has been
10.0, 26 October: a mind you can check. Every answer either proved, worked out in steps and checked; cited, read from a named record with its source; or declined, kept as a hole to fill rather than guessed.
11.0.1, 5 November: parts you can add. A part format in which every new piece of knowledge or skill arrives with its own exam, and is admitted only if it passes. On the 600-question reference: 263 right, 0 wrong.
12.0, 18 November: from answering to doing the work. One chat, in your terminal or browser, where Oracle7 looks up what it does not hold, explains topics from cited sources, builds programs from plain descriptions, repairs your code until your own tests pass, and carries projects through to what actually ran. The reference score stayed at 263 right and 0 wrong.
12.0.2, 19 November: the release after one more round of first-use testing, 49 of 49 on a sweep that 12.0.1 passed 31 of, and the seventh satellite: the Oracle7 Hybrid Game Engine, a developer preview, open source.
Where it is now
12.5 is the language core. A word, in Oracle7, is not a label for knowledge stored somewhere else: it is its definition, made of other words that are themselves definitions. Explaining something is unfolding those definitions until they meet. The reading of definitions and the meaning across words already work under independent judging: on the last fresh round, 26 of 30 opposites and 40 of 60 related pairs were found, with no false links between 60 unrelated pairs. What remains is the hardest part: answering why and how, by running what words do on a situation, and judging it all against an independent acceptance in conversation, writing and building.
Where it goes next
13.0, the laboratory, and receipts. Bring a dataset: it proposes hypotheses, registers them before reading the data, tests them, and reproduces each a second way. And every answer and check carries a receipt anyone can re-verify from the receipt alone, offline.
13.5, the first real plant. The Slow Fire kit on real sensors, holding its bars through a season.
14.0, engineering. A solar setup, a beam, a circuit or a pump designed to stated specifications, every design checked by independent open-source simulators.
15.0, self-sustaining growth. Most new abilities come from Oracle7's own growth loop, with the cost of each verified ability falling across releases.
15.5, off-grid. The whole checked system on a phone-class board, offline.
16.5, every subject. Whole courses built from verified answers and lessons.
17.5, your library. Your own documents and measurements become private parts on your own machine, checked the same way, with nothing sent anywhere.
18.0, the long ledger. Any answer replayed against what was known when it was given, across every release.
18.5, check anything. Paste an article or an invoice and every fact and number comes back checked, wrong or unknown, each with its source.
19.5, everywhere. The same checks in a browser extension, a spreadsheet function and a phone keyboard, all calling your local install.
How each step is judged
Every stage on this road ships only when it passes a test written down before the work began, with its failures published. That is why the dates are not on the planned stages: each one is done when its proof passes, not when a calendar says so.
Oracle7 12.0.2: https://t.co/Dg7LVbyJJb
Game engine: https://t.co/BudWrks5ko
https://t.co/zDXSqSiEN7
GTAOralce7- A GTA inspired game powered by Oralce7
MAYHEM MODE
The first choice in Mayhem happens before the clock starts. Choose three from the practice baton, exhibition staff, Motor Works tool kit, neon ray gun and arcade sidearm. Your first pick starts in your hand. Use 1, 2 or 3 to switch, or open Pocket. If you want to watch a fight before committing, close the picker and choose your loadout when you’re ready.
Once your run starts, time counts on foot. At twelve seconds the first reinforcements enter near you. They come from the city’s existing population, and they can pick fights with one another. Later waves can bring as many as six fighters and arrive as often as every six seconds. Getting into a car pauses your clock, so a long drive won’t carry you up the board.
Press J for Score. A knockdown banks your time and brings up WASTED. You can also finish the run yourself and post it. The board keeps the ten longest completed times in that world alongside your personal best.
There’s a dedicated Mayhem arena you can enter straight away. In the shared main city, ten distinct online players can vote within a minute to start a five-minute Mayhem round. When it ends, everyone goes back to their regular Free Roam progress.
https://t.co/Dyho73PgpV
Meet GTAOracle7. We built a playable neighborhood inspired by Washington, D.C.’s U Street with the Wander Around 3D engine. Oracle7 powers the decisions of 3,000 simulated residents. One of my favorite things you can do there is leave a job for somebody else.
Say the radio wagon breaks down. You can take the repair call, pick up Reese’s tool kit at Motor Works and follow the marker to the orange wagon. Leave the call on Dispatch long enough and a resident can take it. Oracle7 weighs who’s nearby and whether they work at the garage. The person who gets the call goes to collect the tools and does the repair. You might spend that whole time in the record shop.
That’s what “powered by Oracle7” means here. Wander Around gives us the streets you can drive, the doors you can open and the people you see on the sidewalk. Oracle7 weighs what those people can do next. A resident finishing a shift can head home. Someone with cash can stop at a shop that still has what they want. When there’s a show at The Low Note, people can turn up for it. Open Residents, pick someone and you can read the reason behind their latest decision. Walk up and press G to talk. Come back later and someone you’ve met can recognize you.
The band van in the trailer is my pick for a first mission. Call Imani to book Last Set on U. Bring a friend, load the full rig at Motor Works and decide who’s driving and who’s calling ahead for soundcheck. Playing solo? Give Imani the keys and make the call from the passenger seat. Get to Shaw Print with the rig and soundcheck confirmed inside 65 game seconds and the crew earns $90. Run late and you can still play the $54 acoustic set. There’s a separate $12 tip, and Imani remembers how you split it.
You can put your share toward a place of your own. Call June and $30 in game buys the key to an apartment where you can stash purchases and sleep to recover. Take the jade sedan outside Eastside Records for a pursuit drill and watch what the patrol does when it loses sight of you. Or head to Eastbank Municipal Airport, fly the Lark trainer around the river circuit and land it to earn your flight card.
At 0:25 the trailer cuts to Mayhem. The player in orange has a neon ray gun, three reinforcements have just joined the fight, and the survival clock is ticking through wave two. You choose three of five weapons before each run. Residents fight each other and the players. Reinforcements arrive after twelve seconds on foot, then the waves get bigger and closer together. When WASTED hits, your time goes on the board if it makes the top ten. You can enter the dedicated arena on your own or with friends. In the shared city, ten distinct online players can vote within a minute to switch the whole district into Mayhem for five minutes. When the round ends, everyone returns to Free Roam.
If you watch the trailer and remember one name, make it Imani. Call her, let her drive, and see what she says the next time you ring.
https://t.co/Dyho73PgpV
GTAOralce7- A GTA inspired game powered by Oralce7
MAYHEM MODE
The first choice in Mayhem happens before the clock starts. Choose three from the practice baton, exhibition staff, Motor Works tool kit, neon ray gun and arcade sidearm. Your first pick starts in your hand. Use 1, 2 or 3 to switch, or open Pocket. If you want to watch a fight before committing, close the picker and choose your loadout when you’re ready.
Once your run starts, time counts on foot. At twelve seconds the first reinforcements enter near you. They come from the city’s existing population, and they can pick fights with one another. Later waves can bring as many as six fighters and arrive as often as every six seconds. Getting into a car pauses your clock, so a long drive won’t carry you up the board.
Press J for Score. A knockdown banks your time and brings up WASTED. You can also finish the run yourself and post it. The board keeps the ten longest completed times in that world alongside your personal best.
There’s a dedicated Mayhem arena you can enter straight away. In the shared main city, ten distinct online players can vote within a minute to start a five-minute Mayhem round. When it ends, everyone goes back to their regular Free Roam progress.
https://t.co/Dyho73PgpV
A watchtower, a marble archway, a grove of trees, a fountain, glowing lamps and a cottage. Everything in this video was built by the Oracle7 Hybrid Game Engine, Developer Preview v0.1.0, and nothing in it is staged: each object is what the engine's own edit layer accepted, drawn by its own renderer.
Every object is a recipe: a short, declarative list of parts. Boxes, cylinders, spheres, cones and planes, each with a size, a position and one of twenty materials, from stone and marble to water, copper and glowing rune light. The watchtower is a tapered stone shaft, a timber deck and a cone roof. The grove is twelve parts: four trunks, each with two crowns.
What makes it different is what happens before anything is kept. When you ask for a change, the engine checks it and shows you a preview, drawn see-through in the video. Nothing is in the world yet. Only when you confirm is the change committed, and the world moves to its next revision. Every commit is checked against the revision it was made for, so an edit made against an older world is refused, and a retry of the same commit never writes twice.
Ask for something it doesn't support, and it says so. In the video we ask for an elevator; the engine refuses it, naming the part it doesn't know, instead of quietly building something else. Undo takes back the latest revision, and only the latest, so it can never undo someone else's newer work. The finished world exports as plain JSON.
The engine is the world-building core of WanderAround, extracted and released on its own under the MIT license. It also ships as the seventh satellite of Oracle7 12.0.2, where an outside AI such as Claude Code or Codex can talk to Oracle7 and build with these same tools. We ran that integration live: 20 of 20 checks passed.
It is a developer preview. Natural language does not yet go straight to a world change, and walking, physics and multiplayer are still to come.
https://t.co/BudWrks5ko
THE ORACLE7 GAME ENGINE
Today we released the Oracle7 Hybrid Game Engine, Developer Preview v0.1.0, as its own open repository under the MIT license.
What it is: the composable world foundations of WanderAround, extracted into a reusable engine. Worlds are built from declarative recipes, boxes, cylinders, spheres, cones and planes in twenty materials, rendered through WanderAround's own projector, with Rapier physics helpers and a fixed-step clock.
What it does: every change to a world is a preview until it is explicitly confirmed. Every commit is checked against the world's revision, so a stale change is refused and a retry never writes twice, and undo takes back the latest revision. Unsupported mechanics are refused rather than quietly replaced with something else. The world is saved with a lock and an atomic write.
How AI fits: the engine never commits on its own. Through Oracle7 12.0.2, where it ships as the seventh satellite, an outside AI such as Claude Code, Codex or Cursor gets Oracle7's own chat and the engine's world tools through one door, read-only unless you allow changes, with every call recorded. We ran the integration live: 20 of 20 checks passed. On a clean install, its 53 tests, strict type check and build pass.
What it is not yet: natural language does not go straight to a world change (the AI composes the recipes), and walking, physics and multiplayer are not in this preview.
Run it locally with Node.js 20 or later: npm ci, npm test, npm run dev.
https://t.co/BudWrks5ko
solana:DU7Lhe2QtcRBJdor7KbpjfUDmLyA2kdhR49Li3czpump