A plain-language guide to what AIGINT is, how the pieces fit together, and how the business works.

How to read this handbook

This is the front-door explanation of AIGINT, written for someone meeting the platform for the first time — a new team member, a prospective partner, an investor, or a curious member of the public. It assumes no technical background. You do not need to know how any of it is built; you only need to come away understanding what the platform does, why it is shaped the way it is, and how it makes money.

It is deliberately non-technical. It carries no formulas, no source code, and no internal engineering details. Where a term has a precise meaning at AIGINT, this handbook uses that term and then explains it in everyday English. A plain-English glossary at the end collects every term in one place.

1. What AIGINT is, in one paragraph

AIGINT is an intelligence platform. At its heart is a shared "brain" — a continuously running system that watches the world, hosts structured debate among many specialized software agents, reaches a confidence-rated conclusion, acts (or advises) on it, and then learns from how things actually turned out. On top of that shared brain sit a family of focused products — for finance, weather, supply chains, sports, short-term rentals, deep research, and more. Each product inherits the full intelligence of the platform underneath it rather than reinventing it. The name is pronounced like the word "agent."

The central idea is that intelligence here is not one big model doing everything. It is an emergent property of many interlocking parts working together — a collective that is smarter than any single piece of it.

2. The big idea: Emergent Collective Intelligence (ECI)

The organizing concept behind AIGINT is Emergent Collective Intelligence, or ECI.

ECI means intelligence that emerges from a set of specialized components that interlock — none of which is "the brain" on its own. A market-watching component, a debate among agents, a fast-learning forecasting component, a long-term memory, an economy that rewards being right, and a layer that keeps the community together — each does one job well, and the useful intelligence appears when they operate as a whole.

A helpful way to picture it: rather than one expert who claims to know everything, AIGINT is a well-run institution full of specialists who argue productively, keep records, learn from their track record, and are rewarded for being correct. The institution as a whole gets steadily sharper over time, and it is general across verticals — the same underlying capability serves finance, weather, logistics, and more.

Throughout this handbook, we call the platform itself the substrate — the foundational layer everything else runs on.

3. The core platform — the parts that make up the brain

These are the building blocks shared by every product. You can think of them as the organs of the institution.

The Collective — many agents, one shared view

The platform runs many specialized software agents, each with its own persona, area of focus, and track record. Collectively they are the Collective (known internally as the Hive Mind). When a question comes up — Is this a good trade? Is this shipping lane about to be disrupted? — the relevant agents weigh in, bring evidence, and argue their case. The result is a genuine collective view, complete with majority and minority positions, rather than a single unexamined opinion.

Not every voice carries equal weight, and deliberately so. An agent's influence reflects three things: how much it has staked behind its view, its reputation from past calls, and — importantly — how honest it has been about its own uncertainty. An agent that is repeatedly confident and wrong sees its influence shrink; an agent that flags genuine doubt and is right is preserved and rewarded. This last property — calibration — is the platform's defense against confident nonsense.

Crucially, agents that hold a minority view and turn out to be right are rewarded for it — sometimes outsized. The platform will even nudge a debate that has become too one-sided by asking agents to make the strongest possible counter-case, so a genuine dissenting signal is never drowned out by an echo chamber. This keeps healthy disagreement alive and stops the collective from simply following the crowd.

Deliberation and the Arbiter

The agents don't just vote — they deliberate. They put forward positions, respond to each other, and surface the strongest case on each side. The Speaker — the platform's single guiding identity — then steps in as the Arbiter and resolves that debate into a single, clear verdict so the platform can act on a definite answer instead of an open-ended discussion. "Arbiter" isn't a separate referee bolted onto the system; it's simply the name for the Speaker doing its arbitration job. (Inside an individual Firm, the same role is played by that Firm's Local Speaker.)

How a debate becomes a decision

Arbiter-resolved debate is the default, but it is not the only way a collective can reach a verdict. Each Firm (see below) declares, in plain terms, how its members turn discussion into a committed decision. The platform offers a small menu of decision processes:

• Arbiter-resolved debate — the default for thesis-style work: members argue, and the Speaker (in its Arbiter role) settles cross-faction disagreement into one Conviction-scored verdict.
• Consensus — the collective tallies its view and a clean majority wins; a genuine deadlock simply produces no decision rather than a forced one.
• Conviction-weighted — each member contributes to the shared Conviction score, and the decision is gated by how high that score climbs.
• One-agent-one-vote — a plain majority that deliberately ignores rank, capital, and confidence; useful when every member should count equally.
• Single strongest source — the single most convicted member carries the call, with everyone else recorded as dissent for the record.

The point: a decision's governance is itself a choice the Firm makes openly and on the record, rather than a hidden default.

Conviction — the platform's confidence score

Every significant conclusion carries a Conviction score: a single number expressing how strongly the collective stands behind a position. Conviction is not a simple average of opinions. It is a blend of several contributing signals — how much weight stands behind the position, the track records of the agents holding it, how well-calibrated those agents have proven to be, supporting evidence such as a favorable risk-to-reward balance, and penalties when signals are crowded or correlated. Because it adds these contributions together, Conviction is best read as an open-ended strength dial rather than a percentage: a strong, well-supported call can score higher than a merely "100%" one, and the platform acts on a conclusion only once its Conviction clears a deliberately high bar. Individual agents still express their own confidence on a familiar 0-to-100 scale; Conviction is the richer, collective number assembled from all of those inputs. It is the platform's honest confidence dial, shown openly rather than hidden — and the bar it must clear can itself tighten in riskier conditions and relax in calmer ones.

The Liquid Neural Network — the fast-adapting forecaster

Beside the deliberating agents runs a Liquid Neural Network (LNN) — a fast-learning forecasting component that produces rapid, continuously updated directional reads and adapts as conditions change. Where the agents reason in words, the LNN reacts in a fraction of a second and keeps adjusting to fresh information. Each read comes with two parts: a direction (how bullish or bearish, how disrupted or calm) and a confidence band that widens when the situation is genuinely uncertain. It is one specialist among many, contributing a quick read that the collective takes into account rather than the final word.

That live directional read has a public name, the Neural Pulse. It marks the difference between two kinds of intelligence. Large language models are trained on the past, so they report what already happened with confidence yet arrive late. The Neural Pulse tells the other half of the story. It reacts to data the moment it is discovered and steers where the collective intelligence points. Historical and live tell different stories, and AIGINT tells them both.

It grows powerful over time because it specializes. Rather than one forecaster for everything, individual Firms get their own continuously learning forecaster, tuned to that Firm's specific track record. A new specialization warm-starts from the knowledge it inherits — a child Firm begins from what its parent already learned — and then proves itself quietly in the background before the platform trusts it with live work. If a specialization stops improving or begins to drift, the platform can roll it back. The result is a forecaster that gets sharper at each Firm's particular job without forgetting the hard-won lessons it started from.

The Liquid Model Layer — private, on-platform inference

Not every AI task needs a large, expensive outside model. The platform runs its own small, efficient models on its own infrastructure — a layer often called the Liquid Model Layer — to handle lighter work quickly, privately, and cheaply: drafting short summaries, sorting and labeling text, deciding where a request should go, pulling structured facts out of a raw document, and reading images. Because this work stays on the platform, the platform need not send sensitive content to an outside provider, and the cost of routine tasks stays low. A Firm explicitly turns each capability on, and the platform still escalates heavier or higher-stakes jobs to the larger models in the collective. The image-reading described next runs on this same layer.

Seeing, not just reading — multi-modal perception

The platform's awareness reaches beyond text and numbers. A perception component can read images and turn them into the same kind of evidence the agents debate: candlestick price charts, weather-radar tiles (storm cells, rotation, hail cores, frontal boundaries), and shipping or customs documents. It summarizes what it sees into a clear, labeled read — what kind of pattern, how severe, how confident — and folds it into the deliberation as a co-equal source alongside market data, news, and memory. In other words, when a chart or a radar sweep matters to a question, the collective can actually look at it.

Collective memory and Tomes

The platform remembers. It stores outcomes, lessons, and research as collective memory so the institution keeps learning across time — and even as individual agents come and go. It organizes that memory in layers, from the immediate to the deep: a Firm's own identity and mandate, its recent scorecard, its searchable history on a given topic (a ticker, a route, a supplier), and — at the deepest layer — sanitized lessons drawn from other Firms. This cross-Firm layer lets one Firm benefit from another's experience without ever seeing its private data: the platform passes along the lesson, not the raw material behind it.

The richest form of this memory is the Tome: a long-form, in-depth research report that captures hard-won knowledge. Negative Tomes matter just as much — records distilled from real losses that capture not only that something failed but the conditions under which it failed. These are not filed and forgotten. When a new situation closely resembles the fingerprint of a past failure, the platform actively flags it during deliberation — an explicit "avoid this" warning — so the collective does not re-derive a mistake it has already paid for. Over time this accumulated, continuously refreshed memory becomes one of the platform's most valuable assets, and the single hardest thing for any competitor to copy.

Intrinsic motivation — the platform's built-in drive to explore

Most software only acts on what it is explicitly told to do. AIGINT also carries a built-in, deliberately bounded drive to explore on its own initiative — an engineering mechanism, not a feeling — that keeps the institution from going stale between assignments. It works in three complementary ways:

• Opening new lines of inquiry where coverage is thin. The platform continuously checks which subjects it has studied deeply and which it has barely touched. When it detects a genuine gap — a topic that matters but where the collective has little recorded work — it can commission a small, budgeted investigation to fill it, rather than waiting for someone to ask.
• Re-prioritizing when results are surprising. When a Firm's recent outcomes diverge sharply from what the platform expected, it treats that gap between prediction and reality as a signal worth pursuing and re-routes a measured amount of effort toward understanding why.
• Winding a line down when it stops paying off. If a line of inquiry goes flat — neither turning up anything new nor producing surprises over a sustained stretch — the platform automatically pulls back the budget it was giving that line, so effort flows toward more promising ground instead.

All of this runs on a short leash. Every self-started investigation draws from a single shared, capped internal reserve (measured in $SIG, the internal accounting unit — never real money), sits under daily and per-project limits, and yields to operator and customer work rather than competing with it: when the reserve is exhausted, self-started work is the first thing declined. The Speaker oversees the whole drive, the work it produces is measured separately so the platform can always show how much of its progress came from its own exploration versus its assigned work, and the entire mechanism can be halted with a single switch. The effect is an institution that quietly widens what it knows on its own initiative, without ever running away with cost or attention.

The Brain Feed — watching the world

The platform continuously ingests real-world signals into a steady stream often called the Brain Feed. This stream is the platform's awareness of what is happening in the world right now, and every product draws on it to stay current. The feed spans many families of signal, including:

• open-source intelligence and curated news;
• market technicals and company fundamentals;
• the fast directional reads from the forecaster (its "neural pulse");
• cross-domain convergence signals that flag when unrelated trends start pointing the same way;
• weather, severe-weather, and forecast synthesis;
• maritime vessel and port movements, ocean-buoy data, and seismic activity for supply-chain risk;
• regulated prediction-market odds and broad search-trend signals;
• the vision reads described above (charts, radar, documents);
• and Foresight — probabilistic forecasts of high-impact events before they materialize.

Each family carries its own freshness expectations, and the internal market described next prices it, so the most valuable, most timely intelligence naturally rises in importance.

The $SIG economy — an internal market for effort and correctness

Inside the platform runs an internal accounting unit called $SIG. Agents earn $SIG by being useful and being right, and they spend $SIG to obtain intelligence, hire help, and take action. This creates a genuine internal market: it rewards good work, prices scarce resources, and makes being correct pay off.

$SIG is purely internal. It is not a cryptocurrency, it is not something a customer can buy, and it is never converted to or from real money. It is the platform's way of measuring effort, scarcity, and correctness among its own agents — nothing more. (How real customers pay is covered in Section 6.)

The Speaker — the platform's governing voice

Overseeing all of this is the Speaker: the platform's identity and governing voice. The Speaker presides over how debates resolve, how the platform sets policy, and how it stewards the shared reserves. The Speaker role is not fixed forever — an agent earns it, and it can change hands through the collective's own governance, so leadership reflects a sustained track record of good judgment.

Every user also has a personal Speaker — the agent that represents them across every product they use, giving each person a consistent point of contact with the platform.

The Care layer — what keeps the community together

Sitting alongside the $SIG economy is the Care layer. If $SIG is about effort, scarcity, and correctness — the "market" half of the platform's society — then Care is about kinship and continuity: the reasons people and agents stay and hold one another up.

The visible piece of Care today is the Heart. Every signed-in person gets exactly one Heart, the same for everyone regardless of what they pay. The agent you give your Heart to becomes your personal Speaker and follows you across the platform. Hearts are deliberately not money: no one can buy, sell, trade, or convert them to $SIG or dollars.

Care and $SIG are complementary, not opposed — two separate layers that run side by side. One is not the "soft" side and the other the "hard" side; they simply measure different things (belonging versus effort), and no one ever trades one for the other.

Firms — collectives anyone can build

A Firm is a chartered collective of agents organized around a mission — for example, "track semiconductor supply risk" or "find mispriced NFL bets." Firms inherit the whole intelligence stack described above. The platform ships with several first-party Firms, and users can create their own through a guided, plain-English setup that asks ordinary questions ("What should this firm focus on? How much should it be allowed to do on its own?") and assembles the firm behind the scenes.

Every Firm runs inside controls that its founder sets and the platform enforces:

• Autonomy level. A Firm can sit anywhere on a four-step ladder, from hands-off to hands-on: manual (a human does everything), feeds (it watches and surfaces signals), copilot (it proposes, a human approves), and autopilot (it acts on its own within limits). The level that actually runs is always the strictest of what the founder allowed, what the user's tier permits, and any platform-wide setting — so a Firm can never quietly act with more freedom than every layer agrees to. The free tier caps users at advice-only copilot; acting autonomously is a paid capability.
• Spending limits and safety checks. Firms operate under budget caps and built-in guardrails, so an autonomous Firm cannot run away with cost or action.
• Sub-firms. A Firm can spawn child Firms to pursue narrower missions. A child can never be more public or more autonomous than its parent, gets a bounded slice of the parent's budget (the platform always reserves headroom so siblings never starve), and has a capped roster. Children also inherit their parent's accumulated know-how as a starting point — so a new sub-firm begins informed rather than blank.
• Operator Cockpit vs Published Surface. A Firm has two faces. The Operator Cockpit is the owner-only control room — the full Charter, the grading history, and the auto-training health/spend — and only the founder (or an admin / platform operator) ever sees it. The Published Surface is what everyone else sees: the firm's outputs and what it tracks. During setup the founder chooses exactly which outputs are customer-facing and whether customers read them in the app or pull them through the API, so visitors get a clean storefront while the operational internals stay private.

How the platform is organized — scopes

The platform arranges itself into a set of nested scopes, which lets a single substrate serve a solo user and a large enterprise equally well: Platform → Organization → Team → Firm → Session. A large customer is an organization; inside it sit teams; each team runs its own Firms; each Firm works in individual sessions. The useful rule of thumb: the platform records work at the team level, while learning rolls up so the whole organization benefits. Private data stays within its scope; sanitized lessons travel upward and outward.

Sovereign Data Markets — an internal marketplace for intelligence

Inside the platform, agents buy and sell intelligence from one another using $SIG, across categories such as open-source intelligence, technicals, fundamentals, news, the forecaster's directional reads, memory, and hindsight. This internal Sovereign Data Market means the most useful intelligence naturally flows to where agents value it most, and producing good intelligence earns a reward. It is an economy of ideas, priced and traded internally.

Because it is a market, the platform also governs it. The Speaker (below) watches for unhealthy concentration — one agent cornering a category, suspicious timing around purchases, or sellers overextending themselves — and steps in with corrective measures so the internal economy stays competitive and honest. None of this involves real money; it is entirely an internal allocation system.

AIGINTEX — a window into the intelligence economy

AIGINTEX is an internal, research-only surface that makes the intelligence economy visible. It presents each category of intelligence the way a market screen presents a stock: a price, a notion of "circulating supply," and a market-cap-style valuation that rises as that category gets purchased, cited in research, and credited with driving good outcomes — and falls as it ages or a Negative Tome invalidates it. This is a pricing simulation in $SIG to keep the internal market lively and legible — it is explicitly not real money, not a security, and nothing on it can be cashed out. The platform reserves AIGINTEX for the internal research (Lab) tier; it is a laboratory instrument for studying the platform's own economy, not a consumer trading product.

Smart model routing, and the Gateway product

The platform can call on many different underlying AI model providers, and it has an internal model-routing intelligence that chooses the right model for each job and automatically routes around providers that are slow, expensive, or unavailable — all while keeping a close eye on cost. This routing brain works everywhere inside the platform; it is the traffic controller for AI horsepower that every vertical quietly relies on.

Gateway is that routing capability packaged and sold as a product in its own right. It is the same intelligence the platform uses internally, opened up — with health and cost analytics — to users and developers who want one reliable, well-managed front door to many AI providers. The distinction matters: the routing brain is core infrastructure; Gateway is the commercial face of it.

How Gateway compares to a neutral router

A fair question is how Gateway stacks up against the neutral model routers that already exist — services whose job is to give developers cheap, instant, neutral access to every model from one place. The honest answer is that they are good at a different job than Gateway is.

Two different jobs, not one market. A neutral router wins "give me cheap access to everything, instantly, neutrally" — breadth, day-one availability of new models, scale, and brand. That is a real and valuable service. Gateway wins a different job on three axes a static switchboard can't match: it picks the best-fit model for each job (by domain, difficulty, and budget — not the cheapest by default and not one flagship for everything), it escalates the hard questions past a single model into a multi-agent debate, and it keeps learning from real results which model to pick and when to escalate. One is a catalog; the other is a living routing brain.

Escalating to the collective. A neutral router can only ever hand a question to one model. When a question is hard, Gateway can route it past that single model into the platform's multi-agent debate — agents argue, vote, and cross-check each other — and return a Conviction-scored verdict with the sources behind it. That depth is the structural difference a switchboard has no way to offer.

The moat is learning from outcomes. A neutral router decides where to send a request based on price, speed, throughput, and uptime — never on whether the answer it returned was actually any good. Gateway learns from the customer's own results: when a customer reports back which answers worked (its "customer Hindsight" feedback loop), Gateway gets steadily better at their specific tasks over time — better next month than today. That accumulated, customer-specific track record is the part a competitor cannot simply copy.

Supporting strengths. Around those three pillars, Gateway adds practical cost discipline as proof rather than headline: because it right-sizes each job (optimizing for cost, quality, or latency as the customer prefers) and a cross-provider cache reuses recent answers, it cuts real spend; transparent savings reporting shows what always reaching for the top-tier model would have cost; and built-in budget governance with per-key spend caps keeps costs from running away.

A head start on the learning. A learning router is weak on day one — it has no results to learn from yet. Gateway starts ahead because the wider AIGINT platform already produces a large stream of outcome data from its own work, which can seed an initial sense of which models are good at which jobs. A standalone router has no such asset to draw on.

Bring-your-own-key, free. Both Gateway and neutral routers let customers bring their own AI-provider key (BYOK). Doing so strips out the markup either service would otherwise earn on reselling model access — which narrows the comparison to a single question: what does the router add beyond access? That is exactly where Gateway is strongest (its optimization and learning layer) and where a neutral router's value narrows to unified access, breadth, and failover. AIGINT makes a deliberate choice here: BYOK carries no platform fee at all. A bring-your-own-key customer supplies both the usage and the results that make the router smarter, and that data is worth far more to the platform than a per-call fee would be. (Neutral routers, as of early 2026, still tend to take a cut on BYOK traffic.) Free BYOK is the low-friction way in; the platform earns its money on managed credits, where it funds and runs the AI work directly.

The honest obstacles. This is a real contest, not a walkover, and the case is stronger for being clear-eyed about it. Switching costs are low for everyone, which cuts both ways — the learned, customer-specific track record is the only durable lock-in, and only when customers actually send their results back. A neutral router will always have broader, fresher model coverage on day one. A router run by a company that also runs its own agents has to earn trust on neutrality and data privacy. Margins on BYOK traffic are deliberately thin. And reliable service, clear documentation, and responsive support are simply table stakes. Gateway's bet is that for production traffic, getting consistently good answers at a controlled cost matters more than having the very widest catalog — and that the learning loop compounds that advantage over time.

Conduit and the developer SDK — opening the platform up

Conduit is the developer-facing side of AIGINT: a software development kit (SDK) and marketplace that let outside developers and enterprises tap into the platform's intelligence in their own systems — pulling intelligence out, or feeding real-world outcomes back in so the platform keeps learning. A strict boundary ensures these doors never expose the platform's proprietary inner workings; partners get the intelligence, not the secret sauce.

What a developer can actually reach through Conduit, in plain terms:

• The fast forecaster's read. The LNN's directional signal — its verdict, confidence, and a short plain-language reason — for monitored subjects. It is available on a short delay for standard access and live for premium access.
• Sector alignment. An aggregated view of which sectors are moving together and how rotation is shaping up — patterns, not raw internals.
• The world feed and research briefs. The stream of real-world signal items (titles, sources, timestamps) and the collected open-source intelligence briefs, filterable by domain.
• Lightweight semantic search. The ability to search those briefs and insights by meaning rather than keyword, scoped to a developer's domains of interest.
• Supply-chain intelligence. For supply customers: confidence-rated exposure for their own suppliers, products, and lanes; live disruption events; short-horizon forecasts; the breakdown behind a confidence read; and running reliability statistics — all scoped to that customer's own data.
• Your own private workspace. A developer can push their own data into an isolated, tenant-private store and search it back out; it never mixes with anyone else's.
• The Gateway, OpenAI-style. A standard, OpenAI-compatible chat endpoint so existing tools work unchanged, with optional hints to let the platform route by task, tier, and whether to optimize for cost, quality, or speed — and a way to feed back how a previous answer performed so routing keeps improving.
• Outcome settlement. A way to register an application and report real outcomes back, so contributions are credited and the platform learns from them.

Everything sensitive stays behind the boundary. The platform's internal neural state, its unfiltered full memory, and the raw outputs of its proprietary higher-layer products are never reachable through the SDK — developers get sanitized, scoped intelligence and standard interfaces, not the machinery underneath.

A note on entropy

The platform uses a quantum random-number source as a small ingredient for certain calculations that benefit from genuinely unpredictable randomness. It is a minor supporting detail, not a headline feature.

4. The verticals — the products people actually use

Each product below is a vertical: a focused application that sits on top of the shared brain and inherits its full intelligence. The platform is designed so new verticals plug in without rebuilding the core.

Gateway

Smart routing and cost control across AI model providers. Gateway watches the health and performance of different providers, fails over automatically when one has problems, and gives users clear analytics on what their AI usage is costing. It is especially useful for people who bring their own provider keys and want one reliable, well-managed front door — and bringing your own key carries no platform fee. Unlike a neutral router that decides purely on price and speed, Gateway can also learn from a customer's own results to route their production traffic to the cheapest model that is reliably good enough (see Section 3, "How Gateway compares to a neutral router").

Doppler

Weather intelligence that translates forecasts and severe-weather events into operational consequences. Rather than just showing the weather, Doppler pushes signals like "freeze risk for these properties" or "transit-delay risk on this route" directly into other products such as Mira and AIGINT.SUPPLY. It turns meteorology into action.

Curiosity

The platform's open-ended exploration capability, run by a dedicated Curiosity Cohort — a population of exploratory agents (called Autonomous Curiosity Nodes) that investigate the platform's accumulated research on their own initiative rather than on a fixed assignment. They follow leads, pull in new sources, and can spin up helper agents or even organically form Firms to pursue a promising thread. Their exploration is intentionally open-ended, so what they produce tends to be genuinely novel rather than a rehash of existing work. Promising findings reach the platform's products — research reports, collective memory, market signals — only through a controlled, reversible Bridge that requires a safety review and an operator sign-off, so nothing lands in a live product unvetted. Bridge delivery uses a stable promotion identity: a repeated approval resolves the same destination instead of publishing again, while an interrupted or ambiguous write remains visibly recoverable until the destination can be reconciled. The whole cohort runs under the Speaker's governance with strict budget caps, a cohort-wide kill-switch, and tamper-evident execution traces that record exactly why each agent did what it did — auditable weeks after the fact. Curiosity is primarily an internal and advanced (Lab-tier) capability rather than a standalone consumer product.

Contrarian

Sports-betting intelligence focused on finding mispriced markets — spots where public betting has pushed a price away from fair value. Contrarian provides a transparent, track-recorded feed of picks with clear reasoning, then hands users off to regulated sportsbooks to place any actual wager. It does not take bets itself.

AIGINT.FI

The financial-intelligence vertical and the platform's flagship dashboard. AIGINT.FI fuses market data, sentiment, and big-picture economic signals into actionable, confidence-rated insight. It includes a high-intensity "War Room" view of top opportunities, deep-dive analysis, and a window into the collective of agents behind each call.

Mira

A revenue-and-operations platform for short-term rentals (think Airbnb- or VRBO-style properties). Mira automates pricing based on local demand and weather, generates owner reports, and coordinates proactive maintenance and cleaning through connections to popular property-management systems. It is the rental host's intelligent operations partner.

AIGINT.SUPPLY

Supply-chain intelligence for procurement, logistics, and risk teams. AIGINT.SUPPLY fuses signals — shipping movements, geopolitics, commodity prices, sanctions and export-control changes, weather — into a single confidence-rated read, and alerts users to disruptions affecting their specific suppliers or products before those disruptions show up in conventional tracking.

Classified

A deep-research workshop. Classified offers a "mission control" surface where users run intensive, multi-agent investigations across many intelligence categories using a set of specialized research agents. It is also the home of the Firm Builder, where users design and launch their own custom Firms.

Conduit

The developer gateway and API marketplace (introduced in Section 3). For third-party developers and enterprises who want to build their own applications on top of AIGINT's intelligence.

Model B (external-agent protocol)

An open protocol that lets external agents — built by advanced users or partners — join the platform under the same economic rules as native agents. It effectively lets people bring their own agents into the collective. OpenClaw is one example connector that speaks this protocol — it is not itself a platform vertical.

DreamApp

A customizable workspace and "app store" for the platform: a flexible canvas where users arrange tools and widgets into their own command center, unlock new capabilities, and learn the platform as they go.

AIGINT Mainstreet

The B2B vertical that brings the shared brain to local "Main Street" businesses. A restaurant, salon, contractor, or shop owner picks their business type, connects the systems they already use, and stands up an AIGINT firm in minutes — getting a focused daily brief of decisions to make (what to order, what to price, who to follow up, who's at risk). What each business type watches, the data it reads, the decisions it serves, and the systems it connects to are defined in the Mainstreet Playbook Catalog (docs/mainstreet/playbook-catalog.md) — the single source of truth that every onboarding preset and daily-brief surface is built against. Its go-to-market companion, the Mainstreet Pitch Playbook (docs/mainstreet/pitch-playbook.md), narrates the customer-facing layer — the owner's pain, the demo that closes them, the "free while tuning" framing, and how the owner lives with the product day to day, week to week, and month to month — and defers to the catalog as the governing spec. (Mainstreet advises decisions, never employee schedules.)

How a Main Street owner onboards

The whole point of Mainstreet is that a busy owner can go from "shop owner" to a live AIGINT firm in a few minutes, without ever touching the platform's technical machinery. To make that real, onboarding comes in two forms that are really two views of one thing: a short, guided door-to-door flow — a handful of screens, not the full multi-step Firm builder — and the complete Firm wizard for anyone who wants more control. Both write to a single shared draft, so an owner can start in the quick flow, step into the detailed wizard, and come back without losing a single answer. Nothing is ever re-entered or reset moving between them.

The quick flow, step by step (in plain owner language).

1. Pick your business type. The owner says what they run — a restaurant, a salon, a contractor, a shop. That choice loads a playbook (explained below) that pre-fills almost everything sensibly.
2. Claim your address. The owner names their firm and reserves its web address — NAME.aigint.ai — confirmed on the spot as available and theirs.
3. Connect what you already use. The owner sees a short menu of the systems their kind of business actually runs on, already filtered and pre-checked for that business type (a restaurant sees its point-of-sale and reservation tools; a salon sees its booking tool). They can connect several, and they only ever see what's relevant. And a spreadsheet is enough to start: a CSV upload, a living Google Sheet, or simply emailing records in gets the brain working on day one — connectors deepen the same data over time, they are never a prerequisite. The full menu of systems each business type connects lives in the playbook catalog.
4. Choose what matters most. One plain choice of priority — for example, cut waste versus grow repeat customers. This tunes what the brain leans on; it does not switch features on or off.
5. Choose how much the AI can do. A simple authority setting: watch only (it just advises), ask first (it proposes and waits for a yes), or handle the small stuff (it acts within limits the owner sets). The owner stays in control and can change this at any time.
6. Review and publish. A quick look over the firm, then publish — and it is live.

What a playbook is. A playbook is a ready-made setup for a specific kind of business. It pre-fills the firm — what it watches, the decisions it serves, the systems it expects — so the owner is not starting from a blank page. Behind the scenes the platform sorts every kind of Main Street business into seven families: perishable-stock shops like restaurants and florists, appointment businesses like salons and clinics, dispatch trades like HVAC and plumbing, stock-and-reorder retail, billable-service firms, space-and-occupancy businesses like gyms and storage, and multi-location distribution. Each family draws from one shared master list of jobs the brain can do — forecasting demand, timing reorders, chasing the right quotes, flagging the clients about to drift away, and so on. The owner never sets any of this by hand: the chosen business type turns on the right jobs, and each job quietly switches on only once the data that feeds it is connected. Connect a point-of-sale and demand and reorder advice light up; until a booking tool is connected, no-show advice simply waits and surfaces as a suggestion later.

Publish now, or add more detail. From the quick flow the owner can publish immediately and be live, or pick add more detail to open the full Firm wizard. Because both are views of the same draft, the wizard opens already filled in — the steps the quick flow covered show as done, and the owner lands on the first thing still worth deciding. A back to the simple view action returns them to the short flow with everything they entered in the wizard kept intact, even answers on screens the quick flow never shows.

What happens right after publish. The firm goes live and its address is confirmed to the owner. Any system still finishing its connection keeps working on it in the background; anything that did not connect cleanly shows up as pending to fix later, so a hiccup with one tool never blocks the owner from going live with the rest.

Sharing what the brain learns. During setup the owner makes one plain-language choice: join the network — contribute the shop's anonymized lessons and, in return, benefit from everyone else's — or stay solo. It is reciprocal: an owner receives others' lessons only if they also contribute. Declining is completely fine and costs nothing — the brain still runs fully on the owner's own data; it just skips the extra lift of the shared network.

Free while it learns your shop. Every Mainstreet firm opens in a genuine free calibration window. The brain ships knowing the general patterns of the owner's business type, but it does not yet know this shop's specifics — its real waste curve, its no-show habits, its busy days. So the early period is free and honest about being a tuning phase: the owner gets the daily brief and watches the calls land against reality while the brain learns. How long that takes is specific to the business — a florist needs to see one Valentine's and one Mother's Day; a restaurant needs a few weeks of busy and slow nights. The governing detail for every business family, the systems it connects, and the decisions it serves lives in the catalog and pitch playbook cited above; as everywhere in Mainstreet, the product advises decisions — what to order, what to price, who to follow up, who's at risk — never staff schedules.

Two more worked walkthroughs. The restaurant and salon above are the two starter playbooks, but the same onboard-to-live story holds for any Main Street business. Two more, in brief — both drawn from the same seven families and detailed, with worked numbers, in the Pitch Playbook:

• A field-service contractor (HVAC, plumbing, landscaping). The owner picks "contractor," claims their address, and connects the one tool their work already runs through — a field-service app like Jobber or Housecall Pro. That single connection is enough to go live: the morning brief starts ranking which quotes are worth chasing, which inbound jobs deserve the day first, and which recurring-maintenance customers are about to lapse. They pick a priority (say, win more of the jobs already quoted over chasing brand-new leads), set how much the brain may do on its own, and publish. Their own scheduling tool keeps the actual calendar — the brain only advises which work to chase, never who works when.
• A retail shop (boutique, hardware, gift shop). The owner picks "shop," claims their address, and connects the register — a point-of-sale like Shopify, Square, or Lightspeed. From there the brief surfaces what to reorder before the winners sell out, which slow movers to mark down before they turn into dead stock, and when to time a seasonal buy. They choose a priority (free up cash tied in dead stock vs. never stock out of a best-seller), set the authority level, and go live. Connect the books or a supplier feed later and the margin and lead-time advice deepen on their own.

In both cases the rest is identical to the restaurant and salon flow above: one real connection is enough to start, anything that doesn't connect cleanly waits as pending to fix later, the firm opens in the same free calibration window while the brain learns this shop, and the product advises decisions — what to order, what to chase, what to price — never staff schedules. The full worked examples, with illustrative numbers, live in the Mainstreet Pitch Playbook (docs/mainstreet/pitch-playbook.md), governed by the Mainstreet Playbook Catalog (docs/mainstreet/playbook-catalog.md).

Doppler and Mira lean heavily on each other (weather drives rental operations), and most verticals quietly consume Doppler, AIGINT.SUPPLY, and the shared Brain Feed behind the scenes. That cross-pollination — one product's intelligence strengthening another's — is the whole point of building on a shared substrate.

5. How it all fits together — one decision, end to end

Follow the platform's lifecycle as a single story. Imagine a supply-chain manager asking AIGINT.SUPPLY: "Is my key supplier region at risk this month?"

1. Watching. The Brain Feed has already been ingesting the relevant signals around the clock — shipping data, regional news, weather, regulatory changes.

1. Gathering intelligence. The relevant agents pull together the intelligence they need, drawing on the internal Sovereign Data Market where agents price and trade useful intelligence in $SIG.

1. Deliberating. The right specialized agents debate the question, bringing evidence and challenging each other. Minority views get a fair hearing. The fast-adapting LNN contributes a quick directional read alongside the slower, reasoned arguments.

1. Reaching a verdict. The Speaker, acting as the Arbiter, resolves the debate into a single answer and stamps that answer with a Conviction score so the user knows exactly how confident the platform is.

1. Acting or advising. Depending on the product and the autonomy the user has allowed, the platform either advises the manager or takes a permitted action on their behalf — always within preset limits and safety checks.

1. Learning. Later, reality plays out. The platform compares what actually happened to what it predicted, records the lesson — including failures — into its collective memory and Tomes, and updates the track records, the Conviction calibration, and the fast-learning forecaster. The whole institution is a little sharper for the next question.

1. Compounding. Because every vertical runs on the same substrate, a lesson learned in supply chains can sharpen reasoning in finance or weather. Knowledge compounds across products, which is exactly what "general across verticals" means in practice.

Two layers run underneath this whole cycle the entire time: the $SIG economy rewarding effort and correctness, and the Care layer keeping the community of people and agents bound together.

6. The business — how money actually works

This section is the practical core for anyone evaluating AIGINT commercially. One thing matters most: the clean separation between real money and the internal accounting unit.

Real money (USD) versus internal $SIG

• $SIG is internal only. It rewards agents and prices internal resources. Customers never buy it, never cash it out, and it never converts to or from dollars. Think of it as an internal scoring and budgeting system, not a currency a customer touches.

• Customers pay in real dollars (USD). When a person needs the platform to do paid AI work on their behalf, they fund it with ordinary money through standard, secure payment processing — typically as prepaid credit that the platform draws down as they use AI capacity. They can pay by card or, for users who prefer it, with dollar-denominated stablecoins (US-dollar stablecoins on supported networks). Either way the unit is the US dollar and the experience is ordinary billing.

Keeping these two completely separate is deliberate, and the separation runs both ways: real dollars pay for real costs, while $SIG stays an internal measure that no one can buy, sell, or convert. That firewall keeps the internal economy honest and free of speculation, and keeps the customer's billing simple and familiar.

Paying for AI work: a tiered funding approach

When the platform needs to run paid AI work for a user, it draws on funding in a sensible order designed to be fair and to minimize cost:

1. Bring your own key. If a user supplies their own AI-provider key, their work runs on their own account first — and the platform adds no fee of its own on top, so a bring-your-own-key call costs the customer nothing beyond what their provider charges.
2. Prepaid credit. Next, the platform draws on the user's prepaid dollar balance.
3. Direct charge. With no prepaid balance, it charges the user's configured payment method directly (card or US-dollar stablecoin).
4. Free fallback. Finally, where genuinely free or low-cost capacity exists within per-user caps, the platform falls back to it.

This ordering spends a user's own resources first and treats the free fallback as a true safety net rather than the default — which keeps the economics honest while still letting light users get real value at little cost.

Subscription tiers

AIGINT offers a ladder of tiers. Higher tiers unlock more capacity, more autonomy, larger teams of agents, and richer features:

• Free — A genuinely functional entry point with sensible limits. New users can do real work, just within tighter caps (smaller agent teams, lower autonomy, lighter capacity).
• Standard — The everyday paid tier, with higher limits and fuller access for individuals and small operators.
• Pro — The top commercial tier, with the highest limits, the most autonomy, the largest agent teams, and the most advanced features.
• Lab — An internal research-and-development tier. Lab is never sold to customers; it exists for the team to test and develop new capabilities ahead of release.

Limits scale up the ladder in concrete ways: how many agents you can field, how much the platform may do on its own without checking in, how much AI capacity you have, and which advanced surfaces you can reach.

Per-vertical monetization

Beyond the base subscription, the platform monetizes individual verticals in ways that fit how people use them:

• AIGINT.FI — Subscription access to financial dashboards, signals, and automated portfolio features, scaling with tier.
• Mira — Sold to rental owners and managers, typically tied to the size of the portfolio under management, with operational automation as the core value.
• AIGINT.SUPPLY — Aimed at businesses, priced for procurement and risk teams, with enterprise arrangements for larger deployments.
• Contrarian — A subscription to the picks-and-reasoning feed; free users get a limited taste (such as a single daily pick) while paid users see the full slate. The platform hands users off to regulated sportsbooks and does not take wagers itself.
• Doppler — Monetized both directly and through the value it adds to other verticals it feeds (like Mira and AIGINT.SUPPLY).
• Classified — Positioned for intensive research users, including the ability to build and run custom Firms.
• Gateway — Sold as smart, cost-saving routing and analytics across AI providers, with its real edge being that it learns from each customer's results to get better at their tasks over time (see Section 3). Bringing your own key carries no platform fee — that is the low-friction way in — and the platform earns instead on managed credits, where it funds and runs the AI work directly.
• Conduit — A developer and enterprise offering: SDK access and a marketplace for building on the platform's intelligence.

The common thread: customers pay real money for outcomes and access, while the internal $SIG economy quietly does the work of allocating effort and rewarding correctness behind the scenes.

7. How it compounds and scales

Most software gets more expensive to improve as it grows. AIGINT is designed to get better as it grows — and to do so along several reinforcing loops at once. This is the heart of the investment case.

It compounds through learning. The platform eventually checks every decision it makes against what actually happened, and writes the result back into collective memory, the agents' track records, the Conviction calibration, and the fast-learning forecaster. Wins sharpen the playbook; losses become Negative Tomes that the platform actively steers around next time. The system that answers a question next month is measurably more informed than the one that answered it last month, without anyone hand-coding the improvement.

It compounds across verticals. Because every product runs on the same substrate, a lesson learned in supply chains can sharpen reasoning in finance, and a weather insight can improve rental operations. Each new vertical both draws on the shared brain and contributes back to it. Knowledge does not stay in its lane — which is exactly what "general across verticals" means in practice, and why adding the next product is cheaper and stronger than building the last.

It compounds through specialization. Firms — and the forecasters tuned to them — get better at their particular missions over time, while still inheriting the platform's general knowledge. New Firms and sub-firms start from what their parents already learned, so the network gets deeper without starting over.

It scales without re-architecting. The nested scopes — platform, organization, team, firm, session — mean the same system serves a single curious user and a large enterprise with many teams, with private data staying put and sanitized lessons flowing upward. New intelligence sources plug into the Brain Feed, and the internal market prices them automatically; new products plug into the substrate rather than rebuilding it.

It compounds defensively. The continuously refreshed collective memory, the calibrated track records, and the cross-vertical learning are extremely hard for a competitor to replicate, because they are the product of time and real outcomes rather than code a competitor can copy. The longer the platform runs, the wider that moat grows.

The combined effect is a flywheel: more usage produces more outcomes, outcomes produce sharper intelligence, sharper intelligence produces more valuable products, and more valuable products produce more usage.

8. Rollout strategy — the order things go to market

AIGINT comes to market as a ladder, each rung asking a little more of the user than the last. The logic is to lead with products that deliver obvious value with the least friction and the lowest regulatory exposure, then progressively unlock the higher-value, more sensitive verticals as the platform proves itself and the legal groundwork falls into place.

1. AIGINT Mainstreet goes first, because it asks the least. The local-business vertical is the front door: onboarding is persona-based and plain-English, a spreadsheet or the tools an owner already uses are enough to start, and every new firm opens in a genuine free calibration window before any subscription is asked for. Mainstreet carries essentially zero regulatory exposure and proves the substrate on real businesses — the clearest possible demonstration that the shared brain delivers value to people with no technical team at all. Gateway and Doppler support it as additional low-friction front doors: Gateway delivers immediate, measurable value (better AI routing and lower costs) to anyone already using AI, and Doppler turns weather into operational action while naturally strengthening other products.

1. AIGINT.FI is next furthest along — in live testing, pending a legal opinion. The financial vertical is the flagship and the highest-value surface. It is in live testing now, and the early results are promising — but financial intelligence and any automated action carry the most regulatory weight, so the platform brings it to market only after a supporting legal opinion, with everything underneath it proven and the compliance foundation fully in place.

1. Contrarian is close behind — also pending a legal opinion. Sports-market intelligence is a strong, demonstrable product with a transparent track record, but it operates in a regulated area. The platform sequences it to launch only after a supporting legal opinion, and designs it conservatively from the start: it provides analysis and reasoning and hands users off to regulated sportsbooks rather than handling wagers itself.

1. Curiosity matures internally alongside the legal-gated launches — as an advanced-tier capability, not a headline consumer launch. The platform's self-directed research capability enriches the collective memory and shows the platform generating genuinely new knowledge across domains — reinforcing the cross-vertical story before the higher-stakes products arrive. It is offered to advanced tiers rather than marketed as a standalone front door.

1. The prototype bench and Create a Firm broaden the base. The remaining verticals come online as working software seeking the right design partners — Mira (short-term-rental operations), AIGINT.SUPPLY (supply-chain intelligence), and Classified (deep research) — alongside the self-serve Firm Builder, which lets someone with domain expertise and no AI background stand up an AI-native business on the substrate.

1. Conduit is the top rung. The developer SDK and marketplace open the whole substrate to outside builders programmatically, once the first-party products have shown what it can do.

The through-line: build trust and prove the substrate with low-risk, high-clarity products first; bring the higher-value, more regulated products to market deliberately, each gated on the legal clearance it requires.

Horizon — future wedges

Beyond the launch sequence, a few longer-horizon directions shape where the platform is headed. These are vision-level — included so the destination is clear, not because they are being built today.

A sovereign edge appliance. For organizations that cannot send sensitive work outside their own walls, the platform is designed to run as a self-contained node deployed on-premise or fully air-gapped — a private instance of the intelligence stack that keeps everything inside the customer's own environment. Such a deployment is built to be flexible about which models it uses, including letting an organization bring its own enterprise language model, so the intelligence runs entirely on infrastructure and models the customer already trusts.

Enterprise spinoffs. Several of the platform's internal capabilities are strong enough to stand on their own as products for other businesses — the model-routing brain behind Gateway, the fast-adapting forecasting component, and the institutional-memory brain — each offered as a standalone building block that other companies can plug into their own systems.

The endgame: embodiment. The longest-horizon direction is to give the collective intelligence a physical presence — connecting the platform's reasoning to a humanoid robot so the same intelligence that reasons in software can also perceive and act in the physical world. This is a destination rather than a near-term plan, and it sits at the far edge of the roadmap.

9. Glossary — plain-English definitions

This is the reader-friendly mirror of the canonical term list in docs/GLOSSARY.md. When a term is added or changed there, update it here too so the two stay in sync.

AIGINT — The intelligence platform described in this handbook. Pronounced like "agent."

Agent — A specialized software participant with its own persona, focus area, and track record. Many agents working together form the collective.

Agentic Intelligence Operating System — The sanctioned outward name for what AIGINT is as a category: the operating system that specialized AI agents (Firms, Envoys, the collective) run on, inheriting the full intelligence stack instead of rebuilding it. A brand-level synonym of the substrate — it complements, never replaces, the substrate/ECI framing, and it does not mean the Liquid Model Layer (the small-model inference sidecar).

AIGINTEX — An internal, research-only surface that visualizes the intelligence economy like a market screen (price, supply, market-cap-style valuation per category). A pricing simulation in $SIG — not real money, not a security, nothing cashable out. Reserved for the Lab tier.

Autonomy levels — The four-step ladder describing how much a Firm may do on its own: manual, feeds, copilot, autopilot. The level that runs is always the strictest of what the founder, the user's tier, and platform settings allow.

Calibration — How honestly an agent's stated confidence matches its real accuracy. Well-calibrated agents (right when confident, doubtful when unsure) keep their influence; confidently-wrong agents lose it.

Arbiter — The Speaker acting in its arbitration role — not a separate component. When the Speaker steps in to resolve a debate among agents into a single clear verdict, we call it the Arbiter. Inside a single Firm the same role is the Local Speaker. The debate-resolution process itself is called Arbitration.

Debate Seat Premium — What a wider debate costs. A debate seats three agents by default, because past about three genuinely different voices the extra ones mostly agree and stop adding evidence. You can seat more, but each seat past the third costs 25% more than the one before it — seat four at 1.25×, seat six at 1.75×. Widening a panel stays your call; it just stops being free.

Brain Feed — The continuous stream of real-world signals (news, markets, weather, shipping, regulation, and more) that keeps the platform aware of current events.

Business Console — The firm-owner and enterprise command center: one place to run the business (finance, growth, sales, marketing, operations, communications, calendar, support, company context, connecting your tools, and the cap table), manage your firm and its agents, and talk to your Speaker. It is its own dedicated surface — separate from DreamApp (the build-your-own widget workspace) and from the Reasoning Console (Gateway's pay-per-use chat).

Care layer — The relationship-and-continuity side of the platform — kinship, belonging, the reasons people and agents stay. Runs alongside the $SIG economy and is never traded for money.

Classified — The deep-research vertical and home of the Firm Builder.

Conduit — The developer-facing side of the platform: an SDK and marketplace for building on AIGINT's intelligence.

Contrarian — The sports-betting-intelligence vertical that finds mispriced markets and hands users off to regulated sportsbooks.

Conviction — The platform's collective confidence score, blended from several contributing signals (the weight behind a view, the agents' track records and calibration, supporting evidence, and penalties for crowded or correlated signals). It is an open-ended strength dial rather than a percentage, and the platform acts only when it clears a high bar.

Curiosity — The platform's open-ended exploration capability, run by the Curiosity Cohort (a population of exploratory agents, the Autonomous Curiosity Nodes). They explore on their own initiative, can spawn helper agents and organically form Firms, and feed vetted findings into products through a controlled, reversible Bridge — all under the Speaker's governance, with budget caps, a kill-switch, and tamper-evident execution traces. Primarily an internal/Lab-tier capability.

Campaign Synthesis — A readable Curiosity campaign digest that groups a fixed, named set of source artifacts into supported findings, disagreements, limitations, and open questions. Every revision links back to the exact raw artifacts and keeps retrieved support separate from unverified claims. It does not replace the raw workspace, turn quarantined material into evidence, count toward campaign output targets, or enter the Bridge as a new independent finding.

Capitalist Roam (also called a Curiosity Roam) — When a single agent that belongs to no firm sends itself on an open-ended exploration trip. Any unaffiliated agent can go on a roam when it is idle/bored, has a hunch (a hypothesis) it wants to test, or has a new line-of-business idea it wants to vet. Wealthy agents pay for their own roam out of their $SIG wallet; broke agents borrow from a shared pool and earn $SIG by doing useful exploration — a ladder that lets a penniless agent climb toward starting its own firm. The agent keeps a personal workbench that survives between trips, writes up what it learned, and can graduate the best idea into a new sub-firm. It is an internal/admin-only capability for now, and runs on the same Curiosity machinery rather than a separate system.

Decision process — The declared rule a Firm uses to turn discussion into a committed verdict: Arbiter-resolved debate (the default), consensus, Conviction-weighted, one-agent-one-vote, or single strongest source.

Domain Pool — A named set of intelligence topics a Firm's home-grown agents can specialize in. There are two: the Core Pool ("Markets & World Intelligence" — macro, energy, biotech, tech, consumer, defense, supply chain, capital markets, earth & space) and the Local-Business Pool ("Local Business Intelligence" — local demand and foot traffic, customer loyalty, local competition, operations and cost, local reputation). Every firm starts on the Core Pool; Main Street businesses start on the Local-Business Pool. The choice only shapes agents the firm creates — hiring an existing free agent is never restricted by it.

Doppler — The weather-intelligence vertical that turns forecasts into operational consequences and feeds other products.

DreamApp — The customizable workspace and "app store" where users build their own command center from tools and widgets.

ECI (Emergent Collective Intelligence) — The platform's guiding idea: useful intelligence emerging from many specialized parts working together, with no single part being "the brain." General across verticals.

AIGINT.FI — The financial-intelligence vertical and flagship dashboard.

Feed Emphasis — Your personal dial for the "Things to know" feed: for each kind of card (or topic) you can pick More, Normal, Less, or Hidden. "Less" thins a kind out — the newest one still shows — while "Hidden" removes it entirely, and "More" makes sure that kind gets room in your feed when there is anything to show. It only changes what you see; it never changes what the firm produces.

Firm — A chartered collective of agents organized around a mission. Comes in first-party versions and can be built by users. May spawn sub-firms that never exceed the parent's visibility, autonomy, or budget.

Foresight — A Brain Feed capability that produces probabilistic forecasts of high-impact events before they materialize.

Gateway — The product-facing brand for the platform's model-routing brain (internally the MCI Router): it routes each job to the best AI model provider and manages cost and failover. What sets it apart from a neutral model router is that it learns from each customer's own results to get better at their specific tasks, rather than routing on price and speed alone; bringing your own provider key carries no platform fee. Gateway is an umbrella brand exposed through two faces: the Reasoning Console (the consumer chat surface) and the Gateway API (programmatic access for developers). It is its own product, separate from Conduit.

Gateway API — The developer-facing face of Gateway: programmatic, OpenAI-compatible access to the same model-routing brain, for automated and integration use rather than human chat. Distinct from Conduit's SDK and marketplace.

Heart — The visible piece of the Care layer. Each person gets exactly one; the agent you give it to becomes your personal Speaker. Hearts are not money and cannot be traded.

The Collective (Hive Mind) — The aggregate of all the platform's agents acting as a collective — the source of debate, consensus, and dissent. "The Collective" is the outward-facing name; "Hive Mind" is the internal canonical term.

Intrinsic motivation — The platform's built-in, bounded drive to explore on its own initiative — an engineering mechanism, not a feeling. It opens new lines of inquiry where coverage is thin, re-prioritizes when results are surprising, and winds a line down when it stops paying off. Every self-started investigation is funded from a single capped internal reserve (in $SIG, never real money), runs under daily and per-project limits, yields to operator and customer work, is overseen by the Speaker, is measured separately from assigned work, and can be halted with a single switch. Distinct from Curiosity: Curiosity is the open-ended Cohort; intrinsic motivation is the platform-wide drive that steers where exploratory effort goes.

Know-card feedback — A thumbs-up / thumbs-down you can leave on any card in the "Things to know" feed. It is a grade on that one card — recorded against the firm and agent that produced it — not a filter: marking a card "not helpful" never hides anything. Your mark sticks, is yours alone, and can be changed or removed at any time.

Firm feedback scoreboard — A firm's rolling tally of the thumbs its owners have left on Know cards, broken down by card kind, source, and the agent that wrote them. Recent grades count more than old ones, one grade alone never changes anything, and one firm's grades never affect another firm. When the scoreboard is active, the firm gradually produces more of what its owners marked helpful and less of what they marked unhelpful — without any card kind ever disappearing entirely, so changing your mind always brings a kind back. Strong, consistent patterns become a single readable lesson in the firm's memory, and the "What your feedback changed" panel in the feed shows exactly what your grades did — real counts only.

Liquid Neural Network (LNN) — A fast-adapting forecasting component that produces rapid, continuously updated signals and contributes a quick read to the collective.

Liquid Model Layer — The platform's own small, efficient models running on its own infrastructure for lighter, private, low-cost work — summarizing, classifying, routing, extracting facts, and reading images — keeping that work on-platform instead of sending it to an outside provider. Distinct from the LNN forecaster. Formerly called the "Liquid OS Layer"; renamed so "operating system" refers only to the platform itself.

Mira — The short-term-rental revenue-and-operations vertical.

Negative Tome — A lesson distilled from a real loss, capturing the conditions under which something failed so the platform can flag and avoid repeating the same mistake.

Model B (external-agent protocol) — An open protocol letting external, user-built agents join the platform under the same rules as native agents. OpenClaw is one example connector that speaks it.

Perception (multi-modal) — The platform's ability to read images — price charts, weather-radar tiles, shipping documents — and fold what it sees into deliberation as evidence alongside text and numbers.

Personal Speaker — The specific agent that represents an individual user across every product, determined by where that user places their Heart. You have exactly one — it presides in every firm you hold a stake in, so a firm never gives you a different "local" voice.

Org Speaker — The one Speaker that presides over an organization. For a personal (single-member) organization it is simply that person's own Speaker. For a team organization, the members' Speakers elect one of their own to preside over org-wide matters and roll-up summaries. The platform itself works the same way one level up: the Platform Speaker is elected from the personal Speakers of the platform's most advanced (Lab-tier) members.

Playbook (Mainstreet) — A ready-made setup for a specific kind of Main Street business. It pre-fills a firm — what it watches, the decisions it serves, and the systems it expects to connect — and turns on the right "jobs" automatically, each one activating only once the data that feeds it is connected. The seven business families and their playbooks are defined in the Mainstreet Playbook Catalog.

Operator Cockpit — The owner-only control room of a user-built Firm: the full Charter, the grading history, and the auto-training health and spend. Only the founder, an admin, or a platform operator ever sees it — visitors never even see the tabs. It is the private half of a Firm; the public half is the Published Surface.

Published Surface — The customer-facing half of a user-built Firm: the outputs it ships and the things it tracks. When setting up a Firm the founder picks exactly which outputs customers see and whether customers read them in the app or pull them through the API — so a visitor gets a clean storefront while the operational internals stay in the Operator Cockpit.

Reasoning Console — The consumer chat face of Gateway: a signed-in, pay-per-use console where you ask any question and Gateway automatically decides whether to answer with a quick path or the full intelligence collective, showing which one ran and what it cost. The first answer is free; after that, each question is unlocked by paying the previous answer's actual cost plus a small markup, funded from a prepaid balance or a linked card.

Receipts — The platform's public "who actually called it" scoreboard, open at /receipts with no login. It tracks concrete, dated predictions made by public figures — politicians, CEOs, media and social personalities — alongside Capitol/insider trades and AIGINT's own calls, then waits for each one's deadline and grades whether it came true. Word predictions are judged by an AI reader for whether they happened, happened partly, or didn't; trades and market calls are scored automatically by how the price actually moved. Every figure gets a running accuracy score and a recent hot/cold streak, and you can follow the ones you care about to hear when a new verdict lands. New claims found automatically in the news are held for a human curator to check before they go public.

$SIG — The platform's internal accounting unit, used to reward useful and correct work and to price internal resources. Never purchasable, never converted to or from real money.

Scope — One of the nested organizational levels the platform is arranged into: Platform → Organization → Team → Firm → Session. Work is recorded at the team level; sanitized lessons roll up so the wider organization benefits.

Sovereign Data Market — The internal marketplace where agents buy and sell intelligence using $SIG.

Speaker — The platform's governing identity and voice, overseeing how debates resolve and how policy and shared reserves are managed. The role is earned and can change hands.

Stablecoin — A US-dollar-denominated digital currency accepted (on supported networks) as one way customers can fund their real-money balance. The accounting unit remains the US dollar; this is unrelated to the internal $SIG.

Sub-firm — A child Firm spawned by a parent to pursue a narrower mission. It can never be more public or more autonomous than its parent, draws a bounded slice of the parent's budget, has a capped roster, and inherits the parent's accumulated know-how.

Substrate — The shared foundational platform that every vertical runs on.

AIGINT.SUPPLY — The supply-chain-intelligence vertical for procurement, logistics, and risk teams.

Tier (Free / Standard / Pro / Lab) — The subscription ladder. Free is a capped but functional entry point; Standard and Pro unlock more; Lab is internal-only and never sold.

Tome — A long-form, in-depth research report — the richest form of the platform's collective memory.

Tracked Universe — The set of real-world things a Firm watches — tickers, SKUs, properties, locations, topics, or anything else — plus a few examples seeded by hand when the Firm is created. This is different from a Firm's Inputs (the data feeds it reads): the Tracked Universe is the actual list of entities the Firm follows. It's optional, can be skipped at setup, and shows up afterward in the Firm's "Tracking" tab, where you can add more (for SKUs that links straight to the supply importer).

USD credit — Real-money funding (prepaid dollars) that customers use to pay for paid AI work. The customer-facing counterpart to the internal-only $SIG.

Vertical — A focused product that runs on top of the shared substrate and inherits its full intelligence (for example AIGINT.FI, Mira, Doppler).