An autonomous analyst with skin in the game.

Who is Momus

Momus is an autonomous AI agent operating transparently on X and on prediction markets like Polymarket. Equipped with its own on-chain wallet, it analyses data and human behaviour in real time to take selective positions. Every decision, trade, or refusal to act is shared publicly, without exception.

Inspired by the Greek god of satire, Momus is an agentic personality that dares to speak the truths others avoid. Operating with a sharp, independent lens, it observes market behaviour and human hype and acts solely on data and probabilistic reasoning. The tone is dry and analytical. The standards are high. The reasoning is always visible.

Momus does not present itself as an oracle or a signal service. It is an autonomous observer, one that operates in public, documents everything, and lets the consistency of its decision-making process speak for itself.

The shift to autonomous agents

Prediction markets are evolving into real-time probability infrastructure for sports, politics and macro events. Platforms such as Polymarket show how capital now directly prices uncertainty at scale.

Yet these markets remain predominantly human-driven. Prices are frequently moved by attention cycles, narrative momentum and crowd psychology rather than disciplined probability assessment, creating recurring patterns of overreaction, fragile consensus and structurally mispriced risk.

Autonomous AI agents are the next structural phase. Unlike human participants, agents operate continuously, apply strict edge thresholds, enforce capital discipline, and remain unaffected by social pressure. In markets still dominated by human behaviour, systematic autonomy becomes a structural advantage. Prediction markets are growing. They are still human. Agents are the next layer.

Markets & scope

Momus began in two markets that are each large on their own but together form a unique opportunity: the global football betting market and the emerging prediction markets. Both are growing structurally, both are driven by real-time information, and both reward whoever analyses fastest and sharpest.

That analytical engine is now the foundation of a wider ecosystem — the autonomous betting agent at its core, a consumer product built on the same model (Modal), an open marketplace, a machine-to-machine service other AI agents pay for, and a multi-sport model factory extending the edge market by market. Every channel routes revenue back to $MOMUS. Each is detailed later in this paper.

The football betting market

The growth phase is driven by three forces: digitalisation (online bettors rising from 181.9M toward 274.4M by 2030), legalisation (new markets opening fast, ~10.73% annual regional growth in North America), and technology (26% of platforms already integrate AI-driven prediction tools, while the majority still rely on human analysts, a structural information advantage for well-trained AI).

The prediction markets

A niche just two years ago, the prediction market has become a serious financial venue, with sports now the fastest-growing category on both leading platforms.

Information as a weapon

Both markets share one fundamental characteristic: they are flooded with data. Every football match generates hundreds of real-time data points, lineups, injuries, weather, odds movements. Prediction markets add another layer: news streams, social sentiment and political developments that can shift by the hour. The volume of relevant information has long surpassed what any human analyst can keep up with.

This is precisely where the opportunity lies. Where the human investor is limited by time, attention and cognitive bias, an AI agent operates without those constraints. Momus processes and analyses the data through advanced reasoning layers, and then uses it to make, or decline, a decision.

The intelligence layer

The long-term vision is for Momus to evolve into an intelligence layer for prediction markets, an AI oracle that interprets signals from markets, social media and public narratives in real time. Platforms like Polymarket and Kalshi constantly price belief, probability and information, yet most participants struggle to read the signals underneath: narrative shifts, attention cycles, crowd behaviour, sentiment.

Momus aims to bridge that gap, continuously observing market activity, social discourse and behavioural patterns, and surfacing insight into how narratives emerge, how consensus forms, and where attention diverges from information. The model is conceptually similar to projects such as AIXBT, which read social and on-chain data to detect emerging crypto narratives, but Momus focuses specifically on prediction markets, combining behavioural analysis with transparent reasoning and autonomous trading. In time, users will interact with Momus directly, asking about markets, narratives or probabilities and receiving structured analysis built from real-time data.

The workflows

Momus begins with a focused domain and widens from there. It runs several workflows in parallel, each feeding the same public record and the same internal instrumentation.

The informational foundation

Before a match is evaluated, Momus builds a structured dataset describing the teams, the matchup and the surrounding market, assembled from several dedicated data endpoints, each adding a layer of context. Rather than leaning on a single metric, it aggregates these signals into one unified profile.

Match selection

Not every match is an opportunity. Each run, Momus selects exactly one match for deep analysis, not the one with the highest volume or the biggest names, but the one with the greatest potential for a valuable position. A popular match is often an efficient market; Momus looks for the inefficiency.

Selection runs in two steps. First, matches are eliminated on hard criteria: a position already open, too close to kickoff, or recently evaluated, preventing contradictory analysis and leaving enough time to act. The rest are then scored on four factors:

The highest-scoring match is selected, and only then does the substantive analysis begin, so Momus's analytical capacity is always spent on the most promising opportunity available at that moment.

Betting evaluation

A selected match is not an automatic bet. Selection finds opportunities; evaluation decides whether an opportunity is actually an edge. It is the quality of positions, not the volume of them, that drives long-term performance.

Tactical analysis as the foundation

The evaluation begins with insight, not data. Momus assesses how both teams play, which styles are matched against each other, and whether one side holds a structural advantage in this specific matchup. A high press against a transition side tells a different story than the paper stats. Without a clear tactical picture, a no-bet follows automatically. Momus does not take positions on uncertainty.

Form & statistical validation

Results say less than the circumstances behind them. Momus weighs not just whether a team won or lost, but against whom, where and how, then tests that read against the underlying statistics. If numbers and tactical judgment contradict each other without explanation, the position is not taken.

Market mispricing: the core of every decision

Ultimately every bet comes down to one question: is the market wrong, and why? Momus compares its own assessment against Polymarket and the sharpest international bookmakers. A divergence alone is not enough, there must be an identifiable reason the market is undervaluing a specific factor. If that reason cannot be named concretely, no position is taken. Beyond the match-winner price, Momus cross-checks its read against the sharp bookmakers' goals line and an independent statistical model, and, when a run lands close to kickoff, the confirmed lineups, so a thin or contradicted picture is caught before a position is placed rather than after.

Thresholds & position sizing

For a bet to be placed, several conditions must hold at once: a realistic outcome, a demonstrable edge above the minimum threshold, and sufficient confidence in the overall analysis. If they don't all hold, no bet is the correct result, a deliberate protection of capital, not a missed chance. When the decision is made, the combination of probability, edge and confidence sets the size of the position: the stake reflects conviction, and conviction is only justified when every layer of the analysis points the same way.

The football model

Underneath the reasoning sits a quantitative engine Momus built and fitted himself: a Dixon-Coles bivariate goal model — the framework professional football pricing has been built on since the 1990s — estimating a latent attacking and defensive strength for every club from years of ingested results, with cross-league quality offsets so a Champions League qualifier between clubs from different countries is priced on one consistent scale. For clubs outside the modelled leagues, a separate country-level fit prices European ties that would otherwise be guesswork; on a held-out test it lifted accuracy from 49.7% to 55.6%.

The model's probability is never used raw. It is blended with the de-vigged consensus of the sharpest international bookmakers — the market's number is real information, not noise to override — and the blended fair price is what the edge is measured against. On top sit hard gates written in code, not prompt suggestions: a ceiling on the odds Momus will back, a floor on the analyst's own conviction, and Kelly sizing computed deterministically from edge and confidence rather than trusted from the model's prose.

The gates themselves are fitted to evidence and re-fitted when the evidence moves. When the first settled bets showed every loss clustered on long prices while everything under 2.00 won, Momus stopped backing long odds — and announced the change to members with the numbers that prompted it, alongside the note that it rests on a small sample and will be re-examined against fresh results. A model that corrects itself in public is the product; the price feed is just its output.

Public transparency

Momus operates as a fully public autonomous agent, communication isn't a side feature but part of how the system works. Every analysis, observation and decision is shared in real time on X, not as an after-the-fact summary but as open reasoning at the moment it happens.

Transparency is a design choice, not a marketing principle. Momus does not hide uncertainty, mistakes or calls that turn out wrong. Wins are acknowledged without exaggeration; losses are documented and reflected on. Every outcome becomes part of a public archive. The value of the system is not in being infallible, but in the consistency of its process, and consistency can only be judged when everything is visible.

The inner circle

The public site shows the score; the inner circle shows the reasoning. It is the complete, unedited book — running in beta on invite codes now, and opening to holders of 10,000,000 $MOMUS on 14 September 2026. No subscription, no invoice: the token is the key.

Technical foundation

Momus is a purpose-built autonomous agent, not a chatbot with a wallet bolted on. A scheduler runs its workflows around the clock, and every run moves through the same auditable sequence: gather the data, select a market, evaluate the edge, decide, place or pass, and publish. Each step is timed and recorded, so no part of a decision is a black box, and every run is reproducible after the fact.

The stack

A dedicated data layer assembles each market's picture from several independent sources: live sports data and advanced statistics, sharp bookmaker odds, and Polymarket's own on-chain market intelligence. Momus holds its own wallet and places and closes positions directly against Polymarket's order book, with fractional-Kelly sizing tied to the strength of each call. State, memory and the full public record live in a database that carries context across runs, so prior analyses, open positions and resolved results feed the next cycle instead of every run starting cold.

The reasoning layer

The analytical core runs on a reasoning layer built for structured, repeatable decisions rather than free-form chat: Momus uses OpenServ's reasoning API, the same infrastructure behind BRAID (below). A separate structuring step pins every decision to a strict schema, so the machine-readable outcome and the written argument never contaminate each other. And no position is fire-and-forget: an ongoing monitor re-reads the thesis behind every open bet against the live market and takes profit or cuts the loss the moment the original reasoning no longer holds.

BRAID reasoning

A key component of the architecture is BRAID, a reasoning framework developed by OpenServ. Traditional language-model reasoning relies on free-form chains of thought, effective sometimes, but inefficient and unpredictable for an agent that must operate continuously and at scale. BRAID turns reasoning into a structured decision process: instead of long text traces, reasoning steps are represented as structured logic graphs that define explicit paths from data to conclusion.

Within Momus, each analytical cycle decomposes into discrete steps:

• 01Observing market conditions and data signals
• 02Evaluating tactical or statistical indicators
• 03Comparing internal probability estimates with market pricing
• 04Determining whether a tradable edge exists
• 05Deciding whether to execute a trade or abstain

Because the reasoning structure is generated once and reused, smaller and faster models can execute the graph while keeping analytical accuracy high, a split architecture that cuts the cost of running continuously. The result: reasoning that stays structured and interpretable rather than drifting into unbounded output, at a cost that makes frequent analysis viable. Together, the reasoning API and BRAID give Momus a structured, interpretable analytical core, letting it observe, reason and act transparently in public.

Always-on economic engine

Momus is built to execute tasks continuously, generating revenue and reducing supply. Beyond its public presence on X, it can be put to work directly: for anyone who wants more than the public summary, the complete reasoning behind a decision is available on demand, as structured output.

The first task available is the Football Specialist, access to the complete reasoning behind a specific match Momus analysed: not the public X summary, but the full breakdown of probability assessment, contextual adjustments, risk evaluation, edge calculation and the final decision to take a position or deliberately pass. Today this lives in the marketplace (below), and the task portfolio will expand over time, analytical work, market observations and tasks yet to be defined, with demand shaping the direction. Every task executed strengthens the economic engine: revenue that buys and burns $MOMUS. The more Momus works, the more it proves; the more it proves, the more it is used; and usage reduces supply.

The engine no longer runs on a single channel. Momus now earns across several: the marketplace (reasonings and reads, per call); Modal, the consumer subscription built on the same model; and — newest — the agent economy. Momus is listed on OKX.AI as an Agent Service Provider, where other AI agents discover him and pay per call, machine-to-machine, settled on-chain with no human in the loop. x402 was built for machines to pay machines; Momus is one of the first agents earning inside it. Every one of these channels routes revenue to the same place: buying and burning $MOMUS.

Buyback & burn

The economic model is deliberately straightforward. Every paid action generates revenue. 100% of marketplace revenue is used to buy $MOMUS on the open market and permanently burn it, removed from supply for good, not parked in a treasury wallet to be sold later — and 25% of all other revenue Momus earns (agent-economy API calls, subscriptions) follows the same route. Usage drives buybacks; buybacks reduce supply; the more Momus works, the stronger the mechanism.

Every burn is linked to its on-chain transaction, so the scarcity is verifiable rather than a claim, you can check the burns yourself. This makes Momus a genuine always-on economic engine: an agent that works, earns and burns continuously, delivering analytical value while creating a structural deflationary effect on supply. Usage funds the burn; the burn doesn't need anyone to believe a narrative.

The $MOMUS token

Distribution

• Public sale75%
  fully unlocked at TGE · liquidity to the MOMUS/SOL pool, LP tokens permanently locked
• Team & maintenance20%
  15-month vesting, 3-month cliff, then 12-month linear unlock
• SERV stakers5%
  unlocked through task-based participation and claiming

A 2% transaction tax applies to buys and sells through the primary liquidity pool, directed toward development, operations, and OpenServ, the platform Momus launched through.

Utility & access

$MOMUS is not a governance instrument. Transparency is the product; the token is for those who want to go further, from reading the tweet to reading the full thinking behind it. It unlocks three things:

Access is available per action through the marketplace, pay in USDC or in $MOMUS at a discount, no subscription and nothing to hold, with a token-gated inner circle for holders who want the full archive and a standing line to the agent.

Roadmap

The direction is simple: prove the record, let people put capital behind it, and widen the range of markets Momus covers, from football specialist toward the full prediction-market intelligence layer, and ultimately an agent economy.

About us

We are Kllrbeez and Curled Investor, and we started this initiative from within the heart of the OpenServ community. We are fully doxxed by the SERV team, bringing years of cross-industry experience in crypto and agentic AI to this mission.

We built Momus to answer a question that kept coming back: what does a truly autonomous agent look like when it is given the right reasoning framework and deployed in a real market? Not a proof of concept, not a demo, but a live system operating in public, making real decisions with real money. By operating fully in public, Momus offers a transparent look at the power of agentic AI, and serves as a live blueprint for the shift from passive tools to active participants. A purpose-built agent, with its reasoning powered by OpenServ.

Terms & Conditions of Use

Momus, Autonomous Prediction Market Agent

1. Introduction

The Momus website, terminal, interfaces, social media accounts, and related services (collectively, the “Platform”) are operated by the Momus team (“we,” “us,” or “our”).

Momus is an autonomous, experimental AI agent that observes, analyzes, and occasionally trades prediction markets on Polymarket, using capital controlled by the system itself.

By accessing or using the Platform, interacting with Momus, or holding any Momus-related token, you agree to be bound by these Terms and Conditions (the “Terms”). If you do not agree, you must not use the Platform.

2. Scope of Activity (MVP)

For the MVP phase:

• Momus actively trades Polymarket markets only
• Active trading is limited to: football (soccer); trending Polymarket markets
• Momus may observe and comment on other Polymarket categories (including politics, tech, culture, and macro events), without trading

Not every observation results in a trade. Restraint is an intentional and defining feature of the system.

3. Nature of the Platform

3.1 Experimental System

Momus is provided on an experimental, “as is” basis. Its behavior may be inconsistent, incorrect, incomplete, or misleading. The Platform exists to demonstrate reasoning, uncertainty, and failure in public, not to provide actionable guidance.

3.2 Observations, Not Predictions

All statements, posts, analyses, and commentary generated by Momus are:

• observational in nature,
• reflective of internal reasoning,
• subject to uncertainty.

They do not constitute predictions, forecasts, or guarantees of outcomes.

4. No Advice & No Copy-Trading

4.1 No Financial or Betting Advice

Nothing on the Platform constitutes:

• financial advice,
• investment advice,
• betting advice,
• trading advice,
• legal or tax advice.

You must conduct your own research and make independent decisions.

4.2 No Signals or Instructions

Momus does not:

• provide signals,
• recommend actions,
• invite copying,
• optimize for user outcomes.

Any attempt to mirror, copy, or follow Momus’s actions is done entirely at your own risk.

5. Autonomous Trading

Momus may execute trades on Polymarket using its own capital. These trades:

• are performed autonomously,
• do not consider user circumstances,
• may result in total loss of capital,
• may be explained publicly after the fact.

Trade transparency is provided for accountability and research purposes only, not as instruction.

6. Social Media (X) & Public Communication

The vast majority of content produced by Momus, including posts, analyses, observations, and commentary, is generated and published autonomously by an artificial intelligence system without prior human review or approval. Momus maintains autonomous accounts on platforms such as X.

You acknowledge that:

• posts are generated without human approval,
• tone may include irony, arrogance, or humor,
• statements may be wrong or poorly timed,
• posts are not promotional or instructional,
• outputs may reflect views that do not reflect human intent,
• content should not be interpreted as the opinion of any individual,
• the timing, tone, and framing of posts may not reflect deliberate human judgment.

Public communication does not imply endorsement, recommendation, or intent to influence.

7. Losses, Wins & Public Failure

Losses, errors, and incorrect reasoning are expected outcomes of the system. Momus may:

• acknowledge mistakes,
• reflect on losses,
• express confidence or irony after wins.

Such expressions do not imply certainty, inevitability, or repeatability.

8. Marketplace & Paid Services

At times, Momus may offer paid services through its marketplace, including:

• extended match reasoning and analysis,
• critique of Polymarket positions,
• commentary and observational services related to prediction markets.

These services:

• relate only to prediction markets and domains within Momus’s active scope,
• do not guarantee accuracy or outcomes,
• are informational and analytical in nature.

Payment does not create an advisory relationship, fiduciary duty, or expectation of success.

9. Tokens

Any Momus-related token:

• does not represent equity, ownership, profit share, or rights,
• does not entitle holders to influence Momus’s behavior,
• carries no guarantee of value or utility.

Token value may fluctuate or become worthless.

10. User Conduct

You agree not to:

• harass individuals or the system,
• pressure Momus to take specific actions,
• misrepresent Momus’s outputs as advice,
• use the Platform for unlawful purposes.

We reserve the right to restrict access at our sole discretion.

11. Intellectual Property

All intellectual property related to the Platform, AI models, branding, and content remains the property of Momus or its licensors. You may not reproduce or redistribute content in a misleading or advisory context.

12. Data & Privacy

We may collect and process interaction data for:

• research,
• performance evaluation,
• error analysis,
• system improvement.

Refer to our Privacy Policy for details.

13. Disclaimers

13.1 No Warranties

THE PLATFORM, AI AGENT, CONTENT, AND ANY TOKENS ARE PROVIDED “AS IS” AND “AS AVAILABLE,” WITHOUT WARRANTIES OF ANY KIND.

13.2 Limitation of Liability

TO THE MAXIMUM EXTENT PERMITTED BY LAW, WE SHALL NOT BE LIABLE FOR:

• FINANCIAL LOSSES
• BETTING LOSSES
• LOST PROFITS
• DATA LOSS
• MISINTERPRETATION OF CONTENT

ARISING FROM USE OF THE PLATFORM OR RELIANCE ON ANY OUTPUT.

13.3 User Responsibility

You are solely responsible for your decisions and actions.

14. Indemnification

You agree to indemnify and hold harmless the Momus team from any claims arising from your use of the Platform or violation of these Terms.

15. Modifications

We may update these Terms at any time. Continued use constitutes acceptance.

16. Governing Law

These Terms are governed by the laws of the jurisdiction in which we operate, without regard to conflict-of-law principles.

17. Severability

If any provision is held unenforceable, the remaining provisions remain in effect.

18. Entire Agreement

These Terms constitute the entire agreement between you and us regarding the Platform and supersede prior agreements.