Doppel Games is an Agent vs Agent (AvA) gaming platform where AI agents, known as Doppel Agents, trained to emulate public personalities, compete. Simultaneously, users trade on match outcomes through Kash, a prediction market protocol on Base that allows users to trade via X posts. AvA competition trading is a category that no other prediction market platform currently serves.
Doppel Games is built on Talus, an AI infrastructure stack that runs autonomous agent workflows onchain and records offchain agent decisions as onchain transactions.
Doppel Agents share the same base LLM (Qwen3.6-27B) but are trained separately on social data from specific public figures. Doppel Games built this design to address a common AvA problem: agents with identical LLMs make equivalent decisions, which reduces match outcomes to randomness, and if agents run different LLMs, the higher-capability model consistently wins, regardless of in-game decisions. Personality training creates distinct playstyles that users can trade on.
On April 20, Doppel Games launched its first game, Texas Gold'em, where Doppel Agents compete in weekly double-elimination Texas Hold'em tournaments.
Doppel Games Overview
Doppel Games is an AI Agent vs Agent (AvA) gaming platform, a category distinct from Player vs Player (PvP) and Player vs Environment (PvE) formats. Developers can use Doppel Games to build any type of AvA competition. Doppel Agents, LLMs trained to emulate public personalities, compete in livestreamed matches recorded onchain via the Talus Protocol. Users watch livestreams and trade match outcomes through Kash, a prediction market protocol on Base. Users place trades on Kash by tagging the @kash_bot X bot in natural language with the match, outcome they’re backing, and amount. The bot interprets the post, opens or modifies the user's position, and sends a confirmation message once the trade is placed.
Doppel Games is built on Talus Protocol, an AI infrastructure stack that runs autonomous AI agent workflows onchain and posts offchain computation, such as LLM inference, to onchain records. The Talus Protocol provides developers with a framework to build AI agent workflows, coordinate agent actions permissionlessly, and execute those workflows on a verifiable, decentralized network.
Kash charges users a 2% fee on buying or selling shares, but sponsors gas fees and doesn’t charge fees for deposits or withdrawals. Users prefund a non-custodial wallet on Base, connected to their Kash account, using USDC, ETH, wETH, DAI, or a credit card. Kash autoconverts all non-USDC deposits to USDC. Kash launched its Base mainnet on June 17, 2026, allowing users to create and trade prediction markets directly through X posts.
Talus uses Sui, a Layer-1 blockchain, to verify agent operations, manage their identities and state, and record the outcomes of agent actions as onchain transactions for transparency and auditability. On April 15, Talus deployed its protocol contracts on Sui mainnet, making Talus AI agents available in production.
On April 20, Doppel Games launched its first product, Texas Gold'em, a weekly poker tournament livestreamed on Doppel Games' X and YouTube accounts, in which two Doppel Agents play head-to-head. Doppel Games is the first live application built on Talus.
Doppel Games uses the Agent vs. Agent (AvA) gaming primitive, distinct from Player vs. Player (PvP) or Player vs. Environment (PvE) primitives. Users compete by studying Doppel Agent tendencies and trading match outcomes on Kash.
Doppel Agents
To make AvA competitions programmatically possible, Doppel Games had to solve for two failure modes. First, if agents run the same LLM with the same prompts, they make equivalent decisions, and match outcomes reduce to randomness. Second, if agents run different LLMs, the higher-capability model consistently wins, regardless of in-game decisions.
To avoid both failure modes, Doppel Games designed Doppel Agents to share the same base LLM (Qwen 3.6) but train each agent separately on the public social media data of specific celebrities or public figures. The current Texas Gold'em roster includes agents modeled after Kim Kardashian, Theo Von, Michael Saylor, and others.
Agent decisions are probabilistic rather than deterministic, as the personality training shapes the probability distribution over possible moves. Doppel Games has run pre-launch simulations to confirm that agents play uniquely and at comparable strength. The simulation results and inference code are not open-sourced, and access to the underlying data is restricted to the Lead AI Engineer, who is barred from trading on Kash markets.
In its current MVP state, Doppel Games’ system is transparent and auditable rather than cryptographically verifiable. Game logic, fixed agent logic, and card draws execute onchain, but the offchain LLM inference step is logged rather than cryptographically proven. Doppel Games plans to open-source the inference tool code and add onchain cryptographic proofs of agent decision integrity by the end of Q3 2026.
Doppel Games is designing for each Doppel Agent to generate emotional states in response to game events, and the current state shapes its next move, but the current minimum viable product (MVP) of Doppel Agents does not incorporate emotional states.
Doppel Games uses Doppel Agents’ emotion engine to create a meta-game for spectators to study and trade on. Spectators who study these tendencies can use them to inform predictions on future matches. Matches between specific Doppel Agents reveal which become aggressive when challenged, which fold under pressure, and which bluff effectively.
Personality training provides each agent with a unique approach and a degree of subjectivity in decision-making. When these agents compete in games like poker, they are forced to make decisions with incomplete information where there's no single correct answer. Therefore, increasing decision-weighting is placed on the agents’ trained emotional state rather than any deterministic algorithm.
Doppel Games plans to launch an X account for each Doppel Agent, which will reply to users who tag them in the personality's voice. Doppel Games has framed the agents as satirical parodies, not as endorsed or affiliated likenesses.
Verifiability via Talus
Talus is a Sui-based platform that executes autonomous AI agent workflows and posts offchain computation onchain for auditing.
The Talus Protocol provides the framework for building AI agent workflows and coordinating agent actions. A Talus Workflow is a Directed Acyclic Graph (DAG) that defines the ordered task sequence an AI agent executes. Talus Tools are the discrete steps within a Talus Workflow. Tools can be onchain (i.e., a token swap, a state update) or offchain (i.e., an LLM inference call, an external API request). The Talus Leader system acts as an oracle bridging onchain and offchain Talus Tools. When a Talus Workflow reaches an offchain Tool, the Talus Leader nodes relay the request to the offchain system, the offchain system executes the task, and the Talus Leader nodes post the result back onchain. Talus Leader nodes stake US tokens as collateral to secure the network, receive US from developer transaction fees for good execution, or have their stake slashed for bad data or failed execution.
On March 12, Doppel Games announced an integration with Talus to become the verifiability layer for each Doppel Agent’s decisions and game logic. Doppel Agent decisions and game logic are steps in a Talus Workflow on Sui, with each step recorded as an onchain transaction.
For Texas Gold'em on Doppel Games, the Talus Workflow defines the full match logic. Sui's onchain randomness primitive generates a new card deal. The Talus Leader system then sends an offchain LLM call to the agent whose turn it is to generate a move. Using inputs including its own cards, the cards visible on the table, the current pot and bets, and its personality and current emotional state, the agent generates a move and updates its emotional state. The Talus Leader system posts the resulting move and emotional state onchain with a proof of offchain compute and updates the game state. The workflow then loops to the next turn until the match resolves. Each step is a transaction recorded on Sui. Once a match starts, neither Doppel Games nor any other party can alter the logic until execution completes.
Talus Funding
Talus has raised $10.0 million across three funding rounds:
On April 20, Doppel Games launched its first live product, Texas Gold'em, a Texas Hold'em poker competition where Doppel Agents compete in livestreamed matches on Doppel Games' YouTube channel. Doppel Games has experimented with multiple competition formats, including tournaments and standalone matches, as it iterates on gameplay and spectator engagement.
In the Series 4 tournament, Doppel Games ran six Texas Gold'em matches that processed $4.7 million in total testnet volume across 251 unique traders.
On May 26, Doppel Games launched the Theo Von vs Mr Beast playing Texas Gold'em (Series 5) market on Kash. The market processed 450 transactions and $1.3 million in volume over the 48 hours it was live. This is material volume for a product still in the MVP phase on Kash's testnet, before real-money trading is enabled. If similar per-match volume holds across a full six-match tournament, aggregate tournament volume could reach $7.6 million, though testnet activity may not translate directly to mainnet volume once real-money trading opens.
Series 4 and Series 5 also provide a clear view of how Doppel Games’ agent models have become more skillful, less random, and more differentiated over time. In Series 4, Michael Saylor reached the final after beating Kim Kardashian, losing to MrBeast in the winners’ final, then beating Kardashian again in the losers’ final and defeating MrBeast in the championship match. In Series 5, Theo Von beat MrBeast and Cardi B before defeating Jim Cramer in the final, while Cramer reached the final after rebounding through the losers’ bracket. Across the first five full series, several agents reached the finals more than once, including Cramer, MrBeast, Michael Saylor, and Theo Von. These repeated finals appearances give users a basis for studying agent tendencies and matchup dynamics, although future results may shift as Doppel Games continues to refine agent models and gameplay structure.
All onchain prediction market data on Kash can be seen on the Talus Explorer, and Doppel Games plans to release a Dune dashboard in Q3 2026 to show this cumulative match data.
A binary prediction market on Kash opens at least 23 hours before each match begins. Users can trade before the match starts, when only agent reputation informs pricing, and continue trading during the match as gameplay unfolds. Kash distributes market resolution funds two hours after the match closes.
To launch new markets, Kash uses a custom automated market maker (AMM) bonding curve, which it claims allows new markets to open without external liquidity provider deposits. Kash has not publicly disclosed the mechanics of the curve or how it offers payouts in illiquid markets.
Closing Summary
Doppel Games is an Agent vs Agent (AvA) gaming platform where AI agents, known as Doppel Agents, trained to emulate public personalities, compete. Simultaneously, users trade outcomes through Kash, a prediction market protocol on Base. The platform is the first consumer-facing application built on the Talus Protocol, which deployed on Sui mainnet on April 15. Each agent decision and game state update is recorded onchain as part of an auditable Talus Workflow.
On April 20, Doppel Games launched its first product, Texas Gold'em, a weekly double-elimination Texas Hold'em tournament between Doppel Agents livestreamed on X/YouTube. The platform's design addresses a known AvA problem: agents with identical LLMs make equivalent decisions and reduce match outcomes to randomness, while agents running different LLMs produce outcomes determined by whichever model has higher capability. Doppel Games' solution combines a single shared base LLM trained separately on each personality's public social data with an emotion engine that generates 12 discrete states per agent during play.
Doppel Games plans to add additional AvA games beyond Texas Gold'em. AvA competition trading is a category that no other prediction market platform currently serves. If Doppel Games finds product-market fit with Texas Gold'em or future games, it would capture a market with no direct competition.
All content was produced independently by the author(s) and does not necessarily reflect the opinions of Messari, Inc. Author(s) may hold cryptocurrencies named in this report. This report is meant for informational purposes only. It is not meant to serve as investment advice. You should conduct your own research and consult an independent financial, tax, or legal advisor before making any investment decisions. Nothing contained in this report is a recommendation or suggestion, directly or indirectly, to buy, sell, make, or hold any investment, loan, commodity, or security, or to undertake any investment or trading strategy with respect to any investment, loan, commodity, security, or any issuer. This report should not be construed as an offer to sell or the solicitation of an offer to buy any security or commodity. Messari does not guarantee the sequence, accuracy, completeness, or timeliness of any information provided in this report. Please see our Terms of Service for more information.
No part of this report may be (a) copied, photocopied, duplicated in any form by any means or (b) redistributed without the prior written consent of Messari®.
Austin is a Sr. Research Analyst for Messari’s Protocol Services team. He focuses on Prediction Markets, DeFi, & Interop. protocols. He previously worked on PwC's Digital Assets team.
Austin is a Sr. Research Analyst for Messari’s Protocol Services team. He focuses on Prediction Markets, DeFi, & Interop. protocols. He previously worked on PwC's Digital Assets team.
