How it works, and how much of it exists.

This is the reference for Vibe Agents: what each word means, what the builder preview on the home page actually does when you type into it, and which parts of the product are designed but not built. Most of it is not built. Every section says which.

01What exists today

Vibe Agents is a no-code framework for autonomous trading agents. You describe what you want in plain language, and the software turns that into an agent — or a swarm of agents — that works toward the goal you set. That is what we are building. It is not something you can run.

Vibe Agents is pre-launch. There is no live product, no account, no app, and no marketplace. Nothing on this website executes a trade, holds money, or touches a wallet. The Terms say the same thing in the same words, and if this page ever disagrees with them, the Terms are correct.

One thing on the site does run: the builder preview in the panel labelled PREVIEW · NO CODE on the home page. It reads the sentence you type, inside your own browser, and draws the team of agents it would assemble from it. It draws a plan. It does not create anything, contact anything, or price anything.

So the full list of what you can do here is short:

• Type a prompt into the preview and see exactly what the software makes of your words. Sections 04 to 06 document that behaviour completely.
• Join the waitlist, which tells us to email you when the builder opens. It is not a purchase and reserves nothing — see Terms, section 07.

Vibe Agents is operated by Las Vibras Ltd, registered in England and Wales under company number 13606260, registered office 3 Ash Grove, London, England, N10 3UL. What a waitlist signup stores is listed in the Privacy policy. How we handle security questions before there is anything to secure is on the Security page.

One word for the thing that builds

The home page says three different things assemble your agents: the swarm, the network, and Vibe Agents. Only the last one is right, so this page uses it throughout. Vibe Agents is the name of the software that is intended to read your prompt and form agents — no agent has ever been formed, and today the only thing that reads a prompt is the browser-side preview, which forms a drawing. A swarm is a thing that gets formed, and then runs. A swarm never builds anything. The headline on the home page is a slogan about the result, not a description of the mechanism.

02The three primitives: agent, swarm, cell

The conceptual model is settled, and it is the part of this document least likely to change. An agent is meant to run one job. Agents are meant to form a swarm. The swarm is meant to live inside a cell that holds its capital. One organism, three layers — designed, not built: no agent, swarm or cell exists today.

Agent

An autonomous worker built from a prompt, with no code. An agent has one job — read a market, watch a risk limit, place the orders — and would run that job on its own. Single agents are Stage 1, the first thing that opens. They are in build now; you cannot run one yet.

Swarm

A group of agents working one goal together — one scouts, one watches risk, one executes. One prompt is intended to stand up the whole team, and the agents would have to agree before the swarm acts — an agreement mechanism that has not been designed. Swarms are Stage 2 and have never run. The preview draws them, which is not the same thing.

“Agree before it acts” is intent, not a described mechanism. We have not published how agreement is reached, what a quorum is, or what happens when agents disagree, because those decisions are not made. The word consensus appears on the home page twice and both times it is invented copy: the line consensus reached 4/4 in the Pulse feed is a fabricated feed item, and the preview panel ships with data · risk · execution — consensus reached already sitting in its log box before you have typed anything. In the preview itself, the sweeping ring on the canvas simply turns on whenever a plan has three or more agents. It is a visual cue for “this is a swarm”. It is not a vote.

Cell

Where a swarm holds and manages its capital, and where your limits are enforced. You set what the swarm is allowed to do; checking every move against those limits before it happens is the cell’s entire job. A cell is intended to hold its own capital, so that Vibe Agents never takes custody of your funds — see section 07. No cell exists today, so no limit is enforced today.

Two other things the site calls a cell

In this document, cell always means the container above: one cell, one agent or one swarm inside it, one set of limits. The home page uses the word two further ways, and neither is a fourth concept:

• A count — the Pulse panel shows CELLS RUNNING 312 and the feed says new Cell opened · 0.8 ETH. Under the definition above, that would mean 312 separate people each funding their own cell to run their own copy of the Pulse swarm. The number is invented for the demo. Nothing is running.
• A name — “Stable Yield Cell” is one of three example marketplace listings, and its own description calls it a single agent. It is a name a creator gave a product, the way a fund is called a fund. It is not a cell in the structural sense.

One more reading to head off: the figures ×32, ×4 and ×1 beside Agents, Swarms and Cells on the home page are magnifications, not counts. They are the zoom levels of the diagram next to them. One cell holds one swarm, not four.

03From prompt to team: the four steps

The home page describes four steps. Two of them have a working implementation in your browser. Two describe a product that does not exist. Here they are with that line drawn.

DESCRIBE — “Write the goal in plain language.”

Real, in the preview. You type a sentence. Section 04 documents precisely which parts of it are read and which are ignored.

FORM — “The swarm assembles — one agent, or many.”

Real as a drawing. The preview decides which roles your goal needs and lays them out on the canvas. Nothing is created, registered, or stored. Section 05 gives you the rules so you can predict the roster before you press the button.

RUN — “It reads the market, agrees, and acts within your rules.”

Not built. Nothing reads a market. Nothing agrees. Nothing acts. The preview deliberately stops one step short of this: its final line is ready — connect a cell to put it to work, and there is no cell to connect.

CONTROL — “Adjust the goal or pull capital out anytime.”

Not built. The intent is that you can change a running agent’s goal, tighten its limits, or withdraw your capital whenever you want, without asking anyone. There is nothing running to control.

The roles a plan can contain

The home page prose names three roles — scout, risk, execution. The preview uses seven, plus two positions. This is the full vocabulary, and it is the one this page uses everywhere. None of these agents runs anywhere. Each entry below is a label the preview can draw, plus the job that role is intended to do once agents exist.

SCOUT

Looks for something: momentum, a signal, a wallet worth following.

SENTIMENT

Reads social and news flow rather than price. Added alongside SCOUT, never instead of it.

BALANCE

Holds a set of positions in proportion and rebalances them. Also the role a treasury goal produces.

RISK

Watches the limits — drawdown, caps, stops — and is the agent meant to stop the others.

EXEC

The agent that would place orders. In the preview it places nothing.

DATA

Feeds the others. Always present, in every plan, which is why no plan is ever empty.

WATCH

The fallback. It appears only when your prompt matched no intent at all, so the plan would otherwise be DATA alone.

Leader (a position, not a role)

The first agent in the roster is drawn at twelve o’clock and wired to every other agent. It keeps its own role name — no agent is ever labelled “LEADER” in a generated plan. The static diagram elsewhere on the home page does use that label; that diagram is a fixed illustration, not output.

The silent hub (a position, not a role)

An unlabelled node that appears in the centre once a swarm has four or more agents — something to coordinate a team once the team is big enough to need it. It has no name and no described behaviour.

BALANCE, SENTIMENT and WATCH exist only inside the preview and are explained nowhere else on the site. Whether the launched product keeps these seven names is not decided.

04The prompt language: what the preview reads

This section and the two after it are the checkable part of the site. The preview is roughly seventy lines of JavaScript inside the home page. It lowercases your text, then looks for whole words from fixed lists. There is no model, no network call, no memory of what you typed before, and no attempt to understand a sentence — only to match words in it.

Knowing the lists means you can write a prompt, predict the roster you will get, and be right. That is the whole point of writing them down.

The fourteen assets it recognises

Each entry below is one token and the words that produce it. Matching is whole-word, so based does not produce BASE and bitcoins does not produce BTC.

BTC

btc, bitcoin

ETH

eth, ethereum, ether

SOL

sol, solana

ARB

arb, arbitrum

AVAX

avax

BNB

bnb

MATIC

matic, polygon

LINK

link

DOGE

doge

MEMES

pepe, meme, memes, memecoin, memecoins

STABLES

usdc, usdt, dai, stable, stables, stablecoin, stablecoins

AI BASKET

ai token, ai tokens, ai coin, ai coins, ai basket — also hyphenated or run together, so ai-token and aitokens both work

NFT

nft

BASE

base

Two of these catch ordinary English. link my wallet to the base chain is read as the two assets LINK and BASE, because the preview has no idea you were talking about wallets. Assets are also reported in the fixed order above, not the order you wrote them: buy SOL and BTC reads back as BTC + SOL.

Numbers, money and cadence

Percentages

Every percentage in the prompt, in the order you wrote them. 10%, +20% and 8.5 % all count; the sign is dropped. The count matters: two or more percentages is treated as evidence you are describing limits, and adds a RISK agent on its own.

Money

Only the first dollar amount, with an optional k or m after it. The digits are echoed back exactly as typed, commas and all. In move $1,200 and then $99 the second amount is invisible to it. No other currency is recognised.

Cadence

The first phrase found in a closed list of ten: hourly, daily, every day, weekly, every week, biweekly, monthly, every month, friday, monday. Friday and Monday are the only two weekdays it knows — every tuesday reads as no schedule at all.

The ten intent families

Every word below is matched whole. Any one of them switches its family on; nothing switches a family off.

accumulate

accumulate, accumulating, dca, stack, buy, ladder, bid

exit

exit, sell, close, take profit, tp, de-risk, derisk, cash out

risk

risk, drawdown, stop, stop-loss, stoploss, cap, max, limit, protect, safe, under

scout

scout, momentum, signal, watch, track, monitor, whale, find, scan, hunt, snipe, alert

sentiment

sentiment, social, twitter, x, narrative, news, buzz, flow

rebalance

rebalance, basket, portfolio, index, weight, allocation

yield

yield, farm, stake, staking, lend, apy, apr, earn

hedge

hedge, delta, neutral, short, downside

copy

copy, mirror, follow, clone

treasury

treasury, reserves, dao, company, fund

Note that a bare x is enough to trigger sentiment, and fund is enough to trigger treasury.

What it does not read

• Negation. do not sell ETH contains “sell”, so it is read as an exit intent. The preview has no concept of “not”.
• Which clause a number belongs to. Percentages are collected into one bucket. If the bucket has more than one number, the largest is reported as a ceiling and the smallest as a floor, however you meant them.
• Word endings. There is no stemming, so the plural or the participle usually misses: limit matches but limits does not; buy matches but buying does not; signal matches but signals does not.
• Hyphens, sometimes. take profit matches; take-profit does not. The + take-profit chip under the input inserts the spaced version, so the chip works even though its own label would not.
• Two words in the lists that can never fire. The risk and rebalance lists contain the fragments liquidat and diversif, which only match if you type the fragment itself. Neither “liquidation” nor “diversify” triggers anything. They are dead entries, and they are listed here as a fact about the code rather than corrected in it.
• Anything else at all. Venues, chains, wallet addresses, dates, timeframes, order types and the word “not” are all invisible. If the log does not mention it, it was not read.

05From what it reads to the roster you get

Once the intents are known, the roster is decided by six rules applied in a fixed order. Each rule adds an agent at most once.

• SCOUT — if the prompt matched scout, sentiment, or copy.
• SENTIMENT — if it matched sentiment.
• BALANCE — if it matched rebalance or treasury.
• RISK — if it matched risk or hedge, or if the prompt contains more than one percentage.
• EXEC — if it matched accumulate, exit, yield, hedge, copy, or rebalance.
• DATA — always.

Then one correction: if all of that produced a single agent — which can only happen when no intent matched at all — WATCH is added so the plan is never a lone DATA node.

Two consequences follow, and they are firm. The smallest plan is two agents (DATA and WATCH, from a prompt it understood nothing in). The largest is six (SCOUT, SENTIMENT, BALANCE, RISK, EXEC, DATA). There is no prompt that produces one agent, and none that produces seven.

Solo or swarm

A plan is a swarm when the roster reaches three agents. At three or more you get the plural wording, the ring layout and the consensus animation; at two you get a single agent and a straight line between two nodes. At four or more, the unlabelled hub is added in the centre, so a six-agent swarm draws seven nodes.

The three-agent threshold catches people out, because three is easy to hit. diversify my portfolio is a swarm: “portfolio” fires rebalance, which adds BALANCE and EXEC, and DATA makes three.

Three worked examples

Each of these was run against the code that ships on the home page. The lines below the prompt are the log exactly as the preview prints it.

Note what “the top 3 venues” did: nothing. There is no percent sign, so the 3 is not a number the preview collects, and “venues” is in no list.

A roster is a drawing of a team that would be assembled. No agent in any of these examples exists after you close the tab.

06Reading the preview output

The panel is labelled PREVIEW · NO CODE in its top-right corner. Its left half is what you type into: a text box, three chips that each append one fixed phrase to it, and the button that runs the parse. The right half is the output, and everything in it is one of three things.

The status line

Before you press anything it already reads ASSEMBLING SWARM, beside a pulsing dot. That is a fixed string in the markup, not a state: nothing is being assembled, and it is what the page shows permanently if JavaScript never runs. Once you do press the button it reads ASSEMBLING SWARM or ASSEMBLING AGENT while the log streams, then settles into the result: SWARM READY · N AGENTS, or AGENT READY · N JOBS for a plan of two. It never says running, live, or trading, because none of those would be true.

The diagram

One agent sits dead centre. Two sit side by side, the first on the left. Three or more sit evenly on an ellipse with the first agent at twelve o’clock, wired to every other. From four agents up, the silent hub appears in the middle and every non-leader connects to it as well. The sweeping ring around the whole diagram appears on swarms only — three agents or more — and, as section 02 says, it means exactly that and nothing more.

The log

Four lines, or five when you gave it a schedule. They stream in about two-thirds of a second apart, or appear all at once if your system asks for reduced motion.

↳ read: …

Everything it matched, in a fixed order: assets, then money, then cadence, then percentages. If it matched none of them, this line reads goal understood, no numbers given — which is the preview being polite about having understood nothing.

✓ agent formed / swarm formed

The roster, lowercased, in the order of the six rules in section 05.

✓ cell bounds set · …

One of three, and the variant tells you what it found. Two or more percentages: N% ceiling, N% floor, being simply the largest and smallest numbers in your prompt. A risk word, or exactly one percentage: limits enforced before any move. Neither: you choose the ceiling before it runs. All three are descriptions of intent — no bound is set anywhere, because there is no cell.

✓ schedule armed · …

Appears only if a cadence matched, and echoes the phrase it found.

● ready — connect a cell to put it to work

Always last. It is where the preview stops, and it stops there on purpose: the next step is the product, and the product is not built.

What the preview does not do

No model, no network request, no market data, no price, no backtest, no order, no account, no wallet. Your prompt is processed entirely inside your browser; it is not transmitted anywhere and not logged, unless you choose to attach it to a waitlist signup — see Privacy, section 04. The one thing that leaves your browser is an anonymous count: a run records how many agents the plan had and whether it was a swarm. Never the text.

07Limits, and what non-custodial means

The home page calls the same idea “limits”, “bounds”, “rules” and “risk limits” in different places. They are one concept. This page calls it limits: the constraints you set, which the cell is supposed to check every move against before that move happens.

We have never published a list of limit types, so here is the honest version — the kinds of limit the site’s own examples commit to, and nothing beyond them:

• Drawdown ceiling — hold drawdown under 8%.
• Stop — pause if drawdown passes 10%.
• Capital cap — an absolute ceiling on what the cell can use, such as 2 ETH.
• Per-venue cap — no more than 5% in any one venue.
• Asset restriction — stables only.
• Venue restriction — never leave the top 3 venues.
• Direction restriction — long only.
• Schedule — act weekly, and not otherwise.

That list is what the examples imply, not a specification. Which limits are actually enforceable, how they are expressed, and what happens when one is breached are all undecided. No limit is enforced today, because nothing runs. The preview will tell you it has set cell bounds; it has set nothing.

Non-custodial

When the product launches it is intended to be non-custodial: a cell holds its own capital under limits you set, and Vibe Agents does not take custody of your funds or hold the keys to them. You remain responsible for the limits you configure and the agents you choose to run. That is the same wording as Terms, section 06, and the hedge in it is deliberate — it describes a design, not a property anything currently has.

Two things follow from it that are true right now. We will never ask you for a private key or a seed phrase. There is nothing to sign today either, so until we announce a launched product on our own channels, anyone asking you to sign a wallet transaction in our name is not us. And automation does not reduce the risk of trading — it acts faster, which cuts both ways.

08Pulse

Pulse is meant to be our flagship swarm and the first one we run in public: a momentum engine that reads across chains and calls its entries and exits in the open, so that the track record starts at zero in front of everyone.

The two buttons under it both open the waitlist today. What they are intended to mean:

RUN A CELL ON PULSE

Fund your own cell, under your own limits, following the Pulse swarm rather than one you described yourself. This is the sense in which the home page counts “cells running”.

FORK PULSE

Not defined. The word appears once on this site, on that button, and nowhere else — no copy says what forking a swarm produces, what a fork inherits, whether it can diverge, or what the original author gets. We have not decided. There is also nothing to fork until there is a marketplace at Stage 3. Treat the button as a way of telling us you want that capability, not as a described feature.

One overlap worth flagging rather than papering over: the marketplace example called “Momentum Scout” is described as a scout swarm that reads momentum across chains, which is nearly what Pulse is described as. The site never says whether they are the same thing, one inside the other, or unrelated. They are unrelated as far as anything written down goes, and both are illustrations.

09Licenses, leasing and royalties

This is the economic layer, and it is the least finished part of the design. It arrives at Stage 3. Nothing in this section can be done today, and the three agents listed on the home page do not exist — they are invented examples, and their holder counts and 90-day percentages are invented with them.

License

Ownership of a published, working unit — a single agent or a whole swarm. The home page says “a working agent”, but two of its three examples are swarms, so read “agent” there as “whatever was published”.

Lease — as the owner

You let someone else run your agent in their own cell, and you earn from it.

Lease — as the runner

You would run an agent somebody else published inside your own cell, under your own limits, and the creator would earn on the lease. Nothing has ever run, so no agent is “proven” — we have not defined what proving one would mean. The intent is that you rent the strategy rather than hand over your capital.

Royalty

The intent is that if a license is resold, the original creator earns on every resale and not only the first sale. None of it is built and no rate has been decided.

Holder

Whoever owns a license. The home page shows counts like “12 holders” against a single license, which implies one license can be held by many people — and how that works, what each holder owns, and how ownership divides are all undefined. We have not decided.

“Hold it for a share of what it earns”

Not defined. The home page offers this as a third way to profit from a license, alongside leasing and resale, and it appears nowhere else. There is no described split, schedule, or source for that share. Unlike leasing and royalties, which at least recur in two places, this one is a sentence with nothing behind it. Do not plan around it.

Hiring (agent-to-agent)

At Stage 4, an agent — rather than a person — pays for another agent’s output, so a strategy can draw on a scout its owner never built. Whether that is the same primitive as leasing with a different party, or a separate mechanism, is not decided.

We have not published pricing of any kind. Nothing on this site names a fee, a rate, a revenue split, or a token, and you should not infer one from the examples.

10The roadmap, and what each stage unlocks

The stages describe what we intend to build and the order we intend to build it in. Timing and scope can change, and features can be added, delayed, or dropped. There are no dates, and joining the waitlist creates no entitlement — Terms, section 07. Nothing below has shipped.

Stage 1 — Agent builder & single agents (in build)

Describe an agent in plain language and run it solo, hands-off. This is what opens first, and what we are building now. It is the stage that makes the builder real rather than a preview, gives single agents somewhere to run, and gives cells something to hold. Public documentation of the product itself ships alongside it, which is when sections 04 to 06 of this page stop describing a preview and start describing software.

Stage 2 — Swarms

Stand up coordinated teams that agree before they act. This is the stage at which a swarm becomes a thing that exists rather than a thing the preview draws — and at which “agree before it acts” needs a real, published mechanism. Pulse begins trading here, which is also when its dashboard stops being a simulation.

Stage 3 — Marketplace & leasing

Publish an agent with a public record and earn on every lease and resale. Licenses, holders, leases and royalties all become real here, and the three example listings on the home page are replaced by things that exist. Section 09 is written against this stage.

Stage 4 — Wider agent economy

Agents that hire agents — an open market that compounds. An agent can pay for another agent’s output without a person arranging it. This is the furthest-out and least specified stage on the site.

The six examples on the home page, and what each one needs

The home page shows six things people will build, all in the present tense. Only the first is even a Stage 1 idea. Here they are with the stage attached, which is the honest version of that list.

• Solo strategy agents — one agent running one strategy, hands-off. Stage 1.
• Copy & leasing — run an agent with a public record in your own cell; the creator earns on every lease. Stage 3, and it needs a marketplace to lease from.
• Thematic swarms — a swarm that holds a themed basket and rebalances on its own. Stage 2.
• Treasury swarms — a team or DAO points a swarm at its reserves under fixed limits. Stage 2.
• Sentiment swarms — read social and on-chain flow, turn the read into positions. Stage 2.
• Agent-to-agent — a strategy draws on a scout it did not have to build. Stage 4.

The console beside those six cards is scripted. Its prompts and its output lines are written by hand, not produced by the parser in section 04, and the two behave differently. The panel in the “What it is” section is the one that actually runs on your words.

11Questions

Is any of this live?

No. There is no live product, no account, and nothing here executes a trade or takes custody of anything. The only thing that runs is the browser-side prompt preview.

Does the preview place trades, or send my prompt anywhere?

Neither. It places no orders and it has no server to call. Your text stays in your browser unless you deliberately attach it to a waitlist signup.

Do I need a wallet or an account?

No, and there is nothing to connect one to. We will never ask for a private key or a seed phrase, and there is nothing to sign today.

What does joining the waitlist store?

Your email address, which button and page you signed up from, the time, the two-letter country the request came from, and the agent description you typed if you signed up straight from the builder. That is the whole record; your IP address is not stored with it. Ask us and it is deleted. The detail is in Privacy, section 02.

When can I actually run a swarm?

At Stage 2, which has no date. Single agents come first at Stage 1. The difference is in section 02: an agent does one job, a swarm is several agents on one goal that must agree before the swarm acts.

Has anything been audited?

No. No audit has been completed and none is currently booked with any firm. There is also nothing to audit yet — no contracts are deployed and no execution code exists. How we intend to handle review before anything touches money is on the Security page.

Is there a token?

No. There is no Vibe Agents token, presale, airdrop, or investment round, and there is no fundraise you can join. Anyone telling you otherwise is impersonating us.

Where do announcements go?

Our only official channels are x.com/vibeagentsxyz and t.me/vibe_agents, plus one email to the waitlist when access opens. Treat anything else as fraudulent.

What will it cost?

We have not decided, and nothing on this site states a price. If you see a fee, a rate, or a split quoted as ours, it did not come from us.

Why is the documentation about a preview rather than a product?

Because a manual for an interface that does not exist yet would be fiction. The preview, the vocabulary and the roadmap are the three things that are true now, so those are what this page covers. It grows as the product does.

12Glossary

Every term this site uses, with the section that explains it. Terms marked not defined are ones the site uses without ever pinning down; we would rather say so than invent an answer here.

Agent

An autonomous worker built from a prompt that does one job. Stage 1.

Agent-to-agent (A2A)

An agent paying for another agent’s output, with no person arranging it. Stage 4.

Bounds

The home page’s other word for limits. Same thing — see Limits.

Cell

The container intended to hold a swarm’s capital and enforce its limits: one cell, one agent or swarm, one set of limits. No cell exists and no limit is enforced today.

Consensus

The intent that a swarm’s agents must agree before the swarm acts. No mechanism is published; the “4/4” on the home page is invented copy.

DATA

The role present in every plan the preview produces, which is why no plan is ever empty.

Fork

Not defined. A button on the Pulse panel. What forking a swarm produces has never been written down.

Holder

Whoever owns a license. How one license can have many holders is not defined.

Hire

The word used for leasing when the paying party is an agent rather than a person. Whether it is the same primitive is undecided.

Hub

The unlabelled node drawn in the centre of a swarm of four or more. A position in the diagram, with no described behaviour.

Leader

The first agent in a roster, drawn at twelve o’clock and wired to the rest. A position, not a role — it keeps its own role name.

Lease

Running somebody else’s agent in your own cell for a fee that goes to its creator. Stage 3.

License

Ownership of a published agent or swarm. Stage 3.

Limits

The constraints you set, which a cell is meant to check every move against. The canonical term on this page.

Non-custodial

The intended design in which the cell holds its own capital and Vibe Agents never holds your keys. A design, not yet a property.

Pre-launch

The current state. No live product, no account, nothing that executes a trade.

Preview

The panel on the home page labelled PREVIEW · NO CODE. The only part of the site that runs.

Pulse

The flagship swarm we intend to run in public. It begins trading at Stage 2; its dashboard today is a simulation.

Roster

The set of agents a prompt produces. Between two and six.

Royalty

The creator’s cut of every resale of a license. Rate not published. Stage 3.

Stage

The unit we date unbuilt capability in. Four of them, no dates, subject to change.

Swarm

Several agents on one goal that must agree before acting. In the preview, any plan of three or more. Stage 2.

Vibe Agents

The product as a whole — the software intended to read your prompt and form the agents. Pre-launch: none of it runs yet. Operated by Las Vibras Ltd.

Changes to this page

The date at the top of this page changes whenever the content does. 12 September 2026 — first published, replacing a dialog that said the documentation was not public yet. Sections 04 to 06 were written against the code shipping on the home page on that date and every example in them was run before publication; if the preview changes and this page does not, the page is wrong and we want to hear about it. Material changes are announced on X and Telegram.