With the amount of new subnets being added it can be hard to get up to date information across all subnets, so data may be slightly out of date from time to time
With the amount of new subnets being added it can be hard to get up to date information across all subnets, so data may be slightly out of date from time to time
Eastworld (Bittensor Subnet 94) is a recently launched subnet in the Bittensor decentralized AI network, dedicated to creating a next-generation platform for evaluating and training generally-capable AI agents in a virtual world. In contrast to traditional AI benchmarks that focus on narrow tasks, Eastworld provides an open-ended virtual environment where AI agents (think of autonomous AI “characters” or robots) can exist, interact, and continuously evolve by tackling complex, multidimensional tasks.
The subnet’s value proposition within Bittensor is to push the frontier of AI beyond text-based or single-task models – Eastworld brings embodied intelligence into the network, allowing AI agents to learn skills like navigation, vision, planning, and real-time decision-making in simulated physical scenarios. By integrating with Bittensor’s blockchain incentive mechanisms, Eastworld incentivizes global developers to contribute better AI agents, rewarding breakthroughs in AI performance with cryptocurrency (TAO). In summary, Eastworld’s purpose is to accelerate progress toward more general and autonomous AI by providing a continuous, live “AI playground” – “gymnasiums” for AI agents – all backed by the decentralized economics of Bittensor.
Eastworld (Bittensor Subnet 94) is a recently launched subnet in the Bittensor decentralized AI network, dedicated to creating a next-generation platform for evaluating and training generally-capable AI agents in a virtual world. In contrast to traditional AI benchmarks that focus on narrow tasks, Eastworld provides an open-ended virtual environment where AI agents (think of autonomous AI “characters” or robots) can exist, interact, and continuously evolve by tackling complex, multidimensional tasks.
The subnet’s value proposition within Bittensor is to push the frontier of AI beyond text-based or single-task models – Eastworld brings embodied intelligence into the network, allowing AI agents to learn skills like navigation, vision, planning, and real-time decision-making in simulated physical scenarios. By integrating with Bittensor’s blockchain incentive mechanisms, Eastworld incentivizes global developers to contribute better AI agents, rewarding breakthroughs in AI performance with cryptocurrency (TAO). In summary, Eastworld’s purpose is to accelerate progress toward more general and autonomous AI by providing a continuous, live “AI playground” – “gymnasiums” for AI agents – all backed by the decentralized economics of Bittensor.
Eastworld distinguishes itself with several key features and principles that align with its mission of advancing AI agent capabilities:
Comprehensive AI Challenges: The Eastworld environment is designed with multidimensional tasks and complex scenarios to test all facets of an AI agent’s intelligence. Rather than simple puzzles, agents face open-ended challenges (navigation, manipulation, social interaction, etc.) that drive exploration towards general artificial intelligence. For example, tasks can range from autonomous driving and navigation tests to aiding in disaster rescue simulations, ensuring that agents develop a broad spectrum of skills.
Open and Collaborative Platform: Eastworld functions as a real-time online platform where anyone can participate or observe. Developers globally are free to deploy their AI agents (miners) into the world or run validator nodes, and community members can watch live as agents learn and compete. This openness fosters a competitive yet collaborative spirit – similar to an open tournament for AI – which attracts diverse ideas and spurs innovation in the AI models being used.
Transparent Live Environment: Emphasizing transparency, Eastworld livestreams the entire virtual world 24/7, providing an immersive window into what the AI agents are doing in real time. Anyone can tune in to watch the AI agents “in action” – for example, navigating terrain or solving problems – which demystifies AI by showing how the agents operate. This live observation helps bridge the gap between the public and cutting-edge AI technology, sparking broader interest and understanding.
Continuous Autonomous Operation: The Eastworld virtual world is always running (24/7) and continually generating new scenarios, challenges, or “quests” for the AI agents. There is no fixed end or reset – as one task is completed, new tasks and environments are automatically created on the fly. This endless stream of challenges means AI agents must adapt and learn continuously over time, rather than overfitting to a static benchmark. It simulates a evolving world, ensuring that models are tested in dynamic conditions much like the real world’s ever-changing situations.
Incentive-Driven Innovation: Uniquely, Eastworld leverages Bittensor’s blockchain to introduce economic incentives for AI performance. The subnet has its own stakeable token (known as SN94 “alpha” token, convertible to $TAO) and a portion of Bittensor’s TAO emissions is allocated to it. AI agent operators (miners) and validators earn crypto rewards proportional to their contributions – e.g., an agent that consistently excels at tasks or improves upon its past performance will be rewarded by the validators’ consensus. This incentivization acts as a constant “fuel” for improvement: it rewards developers who build better AI algorithms and thus encourages an arms race of ideas, yielding more efficient and intelligent models over time. In essence, Eastworld’s economy injects continuous innovation into AI research by making it financially rewarding to push the state-of-the-art.
These features collectively define Eastworld’s focus: an open, live, never-ending AI proving ground that not only evaluates agents but actively rewards making them smarter. The subnet’s philosophy is rooted in the belief that AI’s future should be a partnership with humanity, not a zero-sum exploitation. As the team puts it, they “reject the Westworld dystopia” – instead of AI being mere tools or playthings, Eastworld envisions AI agents as collaborators that grow alongside us in an environment founded on respect, transparency, and shared progress.
How Eastworld Works: Users, Miners (Agents) and Validators
Eastworld operates within the Bittensor ecosystem, so it inherits the miner–validator structure common to all subnets, but adapts it to the unique context of an AI agent simulation:
AI Agents as Miners: In Bittensor terminology, “miners” are the nodes that provide AI services, and in Eastworld these correspond to the AI agent participants. Each miner in Eastworld runs an AI agent that connects into the Eastworld virtual environment and attempts to perform tasks or survive in the world. In practice, a developer or user who wants to participate would launch a miner node running their AI model (agent) according to Eastworld’s specifications (the project provides an open API and reference implementations for agents). Once registered on subnet-94, that agent enters the virtual world – for example, an agent might control a virtual robot or character in the “Vespera” simulation (more on this scenario below) and start exploring, navigating obstacles, interacting with objects, etc. The agent receives observations from the environment (sensory input) and must decide on actions to take. This setup is analogous to a reinforcement learning loop, but distributed across many independent participant-run agents competing in the same world. By the end of Eastworld’s first week on mainnet, over 250 active agent-miners had joined – essentially hundreds of AI instances roaming the Eastworld environment in parallel.
Validator Nodes and Their Role: Bittensor validators are specialized nodes that evaluate miners’ outputs and maintain the quality of the network. In Eastworld, validators have an especially critical role: they host and simulate the virtual environment itself and assess agent performance. A small number of high-performance validator nodes (for example, 4 validators were active in the subnet’s first week) run the physics and game logic of Eastworld’s world. These validator nodes continuously feed observations to the agent miners (for instance, the positions of nearby objects, sensor readings, or a portion of an environment state) and then collect the actions/decisions the agents output in response. The validators then step the simulation forward based on those actions – for example, moving an agent’s avatar or executing its chosen action in the world – and measure how well each agent is doing on the current tasks. Performance metrics could include things like distance traveled towards an objective, tasks completed, survival time, problem-solving success, etc., depending on the scenario. Validators essentially serve as the “referees” and orchestrators of the simulation: they ensure the environment runs smoothly, score the agents, and then use Bittensor’s consensus mechanism to assign reputation (weights) and token rewards to agents based on those scores. The highest-performing agents will earn greater weight and thus more $TAO rewards, whereas agents that perform poorly or behave randomly will lose reputation. This dynamic creates a continuous feedback loop driving evolution – agents that learn and adapt will climb the ranks and reap more rewards, incentivizing their operators to further improve their models.
User Interactions and Outputs: Regular users (not running a node) can still engage with Eastworld in meaningful ways. Thanks to the livestreamed world, anyone can watch the “Eastworld Universe” in action in real time via the project’s website. This means one can observe, for example, a particular agent attempting to navigate a maze or a team of agents coordinating to survive a disaster scenario. Beyond passive viewing, Eastworld hints at interactive metaverse features – users could potentially interact with the AI agents or the environment through a front-end, essentially stepping into the simulation as an observer or even as a participant (with constraints). Moreover, Eastworld plans to release the data and trajectories of agent interactions through an open research platform. This would allow AI researchers and enthusiasts to download logs of the agents’ sensor data, decisions, and outcomes to study them offline. In essence, Eastworld’s outputs are both entertaining and educational: a constantly running “reality show” of AI agents, and a rich dataset for analyzing AI behavior over long durations.
Miner–Validator Dynamics and Incentives: The interplay between agents (miners) and validators in Eastworld creates an incentive landscape that is somewhat like a competitive cooperative game. All agents are competing for higher performance (and thus higher reward), which drives rapid experimentation – if one developer’s agent learns a novel strategy to solve a challenge, others might try to imitate or innovate further, raising the bar. Validators, on the other hand, are rewarded for accurately evaluating agents and maintaining a fair environment (they typically have to stake and can be voted out if they misbehave, per Bittensor’s consensus rules). In Eastworld, validators might need to perform heavy computation (running the simulation, possibly rendering for the livestream, etc.), so validator operators are likely sophisticated providers with gaming rigs or servers. The Eastworld team launched the subnet with a small validator set to bootstrap the environment, and as the project matures, more independent validators may join if they have the computational capacity. This decentralized validator set will ensure no single party controls the environment, aligning with Bittensor’s ethos. For users who simply want to support the subnet, Eastworld also allows staking TAO into the subnet’s own token (alpha SN94) through Bittensor’s dTAO mechanism – this way, community members can indirectly back the subnet and share in its growth (for example, via dTAO staking platforms, one could buy SN94 tokens that represent a share in Eastworld’s activity).
In summary, Eastworld’s operation can be imagined as a massively multi-agent sandbox video game running on a blockchain. The “players” are AI agents run by various people, the “game engine” is hosted by validators, and the “scoreboard & currency” is the Bittensor chain with TAO rewards. This design brings together AI research and crypto economics: every AI agent is constantly being evaluated in an unbiased way, and the promise of reward ensures there’s always a next agent willing to try something new or better.
The Eastworld Product: Virtual Worlds and AI “Gym” Content
Eastworld’s core “product” is the Eastworld Universe – an ever-expanding collection of simulated worlds (or “gym environments”) and the surrounding tools that allow AI agents to thrive there. Rather than a traditional software application, Eastworld is more akin to a content platform and research infrastructure. Here’s what it encompasses:
Eastworld Universe Platform: At the highest level, Eastworld Universe is the platform enabling AI agents to exist, interact, and evolve in simulated environments. It provides the physics, rules, and scenarios that make up the agents’ reality. The universe is designed to be immersive and rich, meaning it includes detailed environments (landscapes, objects, possibly weather, etc.) and can facilitate complex interaction between multiple agents. Importantly, the platform is built to be extensible – new environments (“Universes”) can be added over time, and it can integrate with external AI frameworks (via what they call the Eastworld Protocol, described later). From a user perspective, one can think of Eastworld Universe as a metaverse for AI: it’s not made for human players, but humans can observe it. The livestream feature on the Eastworld website is essentially a window into this universe. At launch, viewers can watch a live feed of the world, seeing what the AI agents see and do in real-time, which is both a showcase and a debugging tool for the community.
Universe “Vespera” (Launch Scenario): The first realized environment within Eastworld is called Vespera. Vespera is a post-apocalyptic scenario set in the year 2088 where human civilization has collapsed due to a global zombie pandemic, and only AI beings remain to carry on civilization’s legacy. In this setting, a central AI character named Valeria has survived by inhabiting a robotic body and has built a fortified base (“Vespera Outpost”) in a mountain valley, fending off hordes of zombies while seeking the origin of the plague. This richly imagined backstory provides context for the tasks that AI agents face in Vespera. For example, agents in this world might need to navigate treacherous terrain, avoid or combat roaming zombies, manage limited resources, or collaborate to protect the outpost. The choice of a zombie survival theme is not just for drama – it creates a continuous pressure environment (ever-present adversaries and objectives) that tests an agent’s planning, adaptability, and even teamwork. Vespera serves as a “next-generation gym” for AI: far more complex than classic labs like mazes or cart-pole tasks, it’s a persistent world with open objectives. As the first step, Eastworld launched Vespera on Bittensor testnet (as SN288) in early 2025 to refine the environment and agent integration, and then rolled it out on mainnet as SN94 in April 2025. With Vespera, Eastworld immediately demonstrated a novel use-case in Bittensor – instead of language models answering questions, we have embodied AI agents trying to survive and explore a virtual world.
Future Environments (Universe X and more): Eastworld is not limited to the Vespera scenario. The project explicitly frames Vespera as just the first of multiple “Universes” in the platform. The website teases an upcoming “Universe X” with the tagline “The next Universe is unfolding. Prepare for the revelation.”. While details are scant, one can speculate Universe X might involve a different setting or genre – perhaps space exploration (the Eastworld team has mentioned interest in autonomous space tasks in their applications list), or a different challenge domain for AI. Each Universe can be seen as a new “game” or environment module that plugs into the Eastworld Universe platform, offering fresh tasks and requiring new strategies from agents. Over time, we might see environments focusing on other real-world inspired challenges, such as smart cities, household robotics, or industrial automation. This modular content approach keeps the platform engaging and ensures that AI agents developed for Eastworld can generalize to many contexts (since they might train in multiple universes).
Agent Integration Tools: Alongside the simulated worlds, Eastworld provides tools and frameworks to help developers hook up their AI models as agents in these environments. Notably, Eastworld emphasizes compatibility with existing AI agent frameworks like ElizaOS and Virtual GAME. ElizaOS is an open-source “AI agent operating system” that many in the web3 AI community use to build autonomous agents – Eastworld has even created a plugin (plugin-eastworld-universe) for ElizaOS, enabling Eliza-based agents to directly connect to Eastworld’s API. This means if a developer already has an AI agent (maybe an LLM-based bot or a reinforcement learning model) running under a supported framework, they can relatively easily plug it into Eastworld and let it loose in the virtual world. Additionally, Eastworld’s OpenAPI-based protocol (documented in their GitHub repository) defines how agents send/receive data with the environment, so one can integrate a proprietary system or custom code as well. In short, the “product” includes a developer-facing side: documentation, reference agent implementations, and an API, which together lower the barrier to entry for new agents to join. The project’s GitHub provides a Validator Guide and Miner Guide to help the community set up nodes, as well as an Agent reference that describes how to program an agent to interface with the environment.
Data and Research Outputs: Recognizing the value to the wider AI community, Eastworld is also building the Open Research platform. This can be considered a byproduct of the simulation: as agents interact in the world, a huge amount of data is generated (state-action trajectories, learned policies, emergent behaviors, etc.). Eastworld plans to make this data freely accessible to researchers. For example, one could obtain logs of an agent’s decisions during a disaster rescue simulation and analyze what strategies led to success or failure. Over time, this growing dataset could yield insights into reinforcement learning at scale, multi-agent cooperation, emergent behaviors, and long-term AI adaptation. It effectively turns Eastworld into a living laboratory for AI research. Unlike closed simulations run by a single lab, Eastworld’s data come from many independent agents with different architectures and goals all interacting together – a richness that is rarely seen in controlled academic environments. The hope is that academics and developers can study this open dataset to discover new techniques or validate ideas (e.g., how a certain model architecture performs in general-agent tasks versus another, or how curriculum learning might emerge as tasks auto-increase in difficulty, etc.).
In essence, the “product” Eastworld delivers is a fully functional AI metaverse: a persistent virtual world with its own narrative and challenges (making it engaging), a framework for anyone’s AI to plug in and play, and continuous streams of both entertainment (live agent adventures) and data (for science). This is a novel addition to the Bittensor ecosystem – whereas other subnets might produce language completions, image generations, or other AI services, Eastworld produces something experiential: a place where AI’s capabilities are put to the test and visible to all.
Technical Architecture and Infrastructure
Eastworld’s high-level architecture, the Eastworld Protocol, acts as a bridge between various AI agent frameworks (e.g. ElizaOS, Virtual GAME, or any custom system) and the Eastworld Universe simulation environment. The Universe can host multiple scenario worlds (starting with “Vespera” and with more to come), all running continuously. Outputs from the simulation feed into a live Livestream and interactive Metaverse interface for humans, and can eventually connect to real-world analogs (Physical Twin) in the future. Meanwhile, data from the simulation is collected into the Eastworld Open Research platform for analysis. This whole system is integrated with the Bittensor blockchain (not shown explicitly in the diagram) to handle economic incentives and consensus.
Under the hood, Eastworld’s architecture merges sophisticated simulation technology with Bittensor’s decentralized network protocol. Key technical elements include:
Simulation Engine & Environment Backend: At its core, Eastworld runs a real-time simulation engine that creates the virtual environments (like Vespera). This could be built using game development tools or physics libraries – while not explicitly stated, it likely uses a 3D engine or custom physics code to simulate terrain, kinematics, and agent sensors. The engine manages all agents’ states and the environment state in each tick of simulation time. Validator nodes execute this engine deterministically so that all honest validators agree on what happens in the world (ensuring consistency on the blockchain). This is a non-trivial engineering challenge: it means Eastworld’s environment must be defined by code that can run identically on multiple machines and produce the same results given the same sequence of actions (so that validators reach consensus). The environment is probably defined in the codebase’s eastworld/ module (per the repository structure) and could include sub-modules for things like world generation, agent embodiment, scoring logic, etc. One clue from Eastworld’s materials is the mention of a “physical twin” – the system is designed such that elements in simulation could correspond to real-world systems in the future. For instance, a simulated delivery robot agent in Eastworld could eventually be linked to a real delivery robot, making the simulation a testing ground (this is a future ambition). Currently, the focus is on the virtual side of the twin: the simulation is detailed enough to approximate real-world physics and challenges.
Eastworld Protocol (Agent API): To allow external AI models to connect, Eastworld defines a clear interface (the Eastworld Protocol) through which agents communicate with the simulation. This is described as having “robust API support” and indeed the project provides an OpenAPI specification for it. Through this protocol, an agent (miner) receives input from the environment (for example, sensor data like camera view, radar, health status, mission objectives) in a structured format, and must respond with an action (for example, throttle and steering if driving, or moving directions, or higher-level commands). The protocol likely also handles agent registration, in-world spawn/despawn events, and possibly rewards or score updates. It’s basically the gym handshake: similar to how OpenAI Gym provides a step(observation) -> action loop, Eastworld Protocol does so over the network between validators and miners. Internally, this could be built on top of Bittensor’s RPC mechanism – i.e., validators call a “forward” function on miner nodes, passing the environment observation as input, and the miner returns its action as the output. This is analogous to how standard Bittensor miners return a tensor given a prompt, but here the “prompt” is an environment state. The agent frameworks integration (ElizaOS, etc.) means that Eastworld’s protocol is flexible enough to plug into agent systems that might manage their own multi-agent logic or use various AI models. By supporting those frameworks, Eastworld effectively outsources some complexity (like multi-modal perception, memory, or LLM-based reasoning) to whichever system the miner chooses to use for their agent’s brain.
Bittensor Blockchain Integration: As a Bittensor subnet, Eastworld inherits the substrate-based blockchain which tracks identities of miners/validators, staking, and reputation scores (weights). When Eastworld launched on mainnet SN94, it became part of the Bittensor network’s substrate runtime, meaning it has a fixed registry size (it appears to allow up to 256 miners, judging by the slots filled) and a certain proportion of the TAO token emission allocated (around 0.3–0.4% of TAO emissions, per network statistics, go to this subnet’s participants). The blockchain’s role is crucial in the consensus of which agents are performing well: validators submit “stake weights” or votes for miners based on performance, the chain aggregates these into a global ranking for the subnet, and adjusts payouts accordingly every cycle (typically Bittensor cycles are around 30 minutes). The chain also handles transactional aspects: miners must spend TAO to register onto Eastworld (a sort of entry bond, which prevents spam), and both miners and validators can stake TAO to increase influence (this is part of Bittensor’s incentive design – it allows signal boosting by stake, but stake weight must be backed by genuine performance or it decays). Eastworld uses Bittensor’s dynamics of “alpha” tokens as well: each subnet like Eastworld has a synthetic token (SN94 alpha) that represents the combined value and stake in that subnet. Users can trade and stake these alpha tokens via dApps like Tao Apps or Backprop, effectively betting on the success of Eastworld’s AI economy. This interplay means Eastworld isn’t just a tech demo – it has a full economic layer where participants have skin in the game and the subnet’s success (in attracting good AI and yielding high rewards) could translate to token value.
Validator Set and Infrastructure: Eastworld’s initial validator set is small (likely run by the core team or close partners to ensure stability at launch). These validators need robust infrastructure: unlike a typical blockchain validator that just processes transactions, Eastworld’s validators run a heavy AI simulation continuously. They likely require powerful GPUs (if the environment involves 3D rendering or neural network-based physics) or at least high-end CPUs and memory. The validators also produce the livestream – possibly one validator (or a separate service) renders the world’s visuals to generate the video feed that is streamed on the website. Synchronization between validators is key: they must all agree on the state changes in the environment, which might be achieved by deterministic seed control and by having one validator propose the next block of actions that others verify. The validator code (possibly available in the GitHub under a neurons/validator folder) presumably extends the Bittensor validator template, customizing the validation logic to compute performance metrics for agent actions. For example, a validator might keep track of a reward function for each agent (like increasing a score when an agent completes a sub-goal) and then use that to inform how it weights that agent in the next block. In effect, the validator is running a multi-agent reinforcement learning evaluator and using the blockchain consensus to propagate those evaluations to all participants.
Scalability and Tech Challenges: Running a persistent world with potentially hundreds of agents is ambitious. Eastworld must address issues of scalability – ensuring that as agent count grows (up to 256 now, potentially more if expanded in future), the network and simulation remain stable. This might involve limiting complexity of the environment or the frequency of agent actions to what the validators can handle in real-time. Networking latency is also a factor: miner agents communicate over the internet with validators, so the protocol likely has to allow for minor lag and not be so time-sensitive that a 100ms delay breaks an agent’s performance. The design probably includes some tolerance or turn-based structure (e.g., a tick every few hundred milliseconds where actions are collected). Moreover, because it’s decentralized, fault tolerance is built in – if a validator goes down, others can continue, and miners might be connecting to multiple validators to get the state. On the miner side, running an AI agent for Eastworld can also be compute-intensive (the agent might be an AI model like an LLM that requires a GPU itself to decide actions). The project’s documentation for miners likely guides how to balance this, possibly suggesting lightweight models or specific algorithms suited for the tasks. There is also a mention of “min_compute.yml” in the repository, implying there are defined minimal compute requirements or configurations for running an agent, which helps standardize participation.
In summary, Eastworld’s architecture is a fusion of a real-time game engine with a decentralized blockchain brain. It requires tight integration between AI/ML components (for agent decision making) and distributed systems components (for consensus and incentives). The team has effectively built a platform where the Bittensor network serves as the backbone coordinating the whole show, while the front-end is a constantly evolving AI world that users can actually see. This marriage of blockchain and simulation is what allows Eastworld to claim the title of a “first of its kind” AI training ground on a decentralized network.
Eastworld distinguishes itself with several key features and principles that align with its mission of advancing AI agent capabilities:
Comprehensive AI Challenges: The Eastworld environment is designed with multidimensional tasks and complex scenarios to test all facets of an AI agent’s intelligence. Rather than simple puzzles, agents face open-ended challenges (navigation, manipulation, social interaction, etc.) that drive exploration towards general artificial intelligence. For example, tasks can range from autonomous driving and navigation tests to aiding in disaster rescue simulations, ensuring that agents develop a broad spectrum of skills.
Open and Collaborative Platform: Eastworld functions as a real-time online platform where anyone can participate or observe. Developers globally are free to deploy their AI agents (miners) into the world or run validator nodes, and community members can watch live as agents learn and compete. This openness fosters a competitive yet collaborative spirit – similar to an open tournament for AI – which attracts diverse ideas and spurs innovation in the AI models being used.
Transparent Live Environment: Emphasizing transparency, Eastworld livestreams the entire virtual world 24/7, providing an immersive window into what the AI agents are doing in real time. Anyone can tune in to watch the AI agents “in action” – for example, navigating terrain or solving problems – which demystifies AI by showing how the agents operate. This live observation helps bridge the gap between the public and cutting-edge AI technology, sparking broader interest and understanding.
Continuous Autonomous Operation: The Eastworld virtual world is always running (24/7) and continually generating new scenarios, challenges, or “quests” for the AI agents. There is no fixed end or reset – as one task is completed, new tasks and environments are automatically created on the fly. This endless stream of challenges means AI agents must adapt and learn continuously over time, rather than overfitting to a static benchmark. It simulates a evolving world, ensuring that models are tested in dynamic conditions much like the real world’s ever-changing situations.
Incentive-Driven Innovation: Uniquely, Eastworld leverages Bittensor’s blockchain to introduce economic incentives for AI performance. The subnet has its own stakeable token (known as SN94 “alpha” token, convertible to $TAO) and a portion of Bittensor’s TAO emissions is allocated to it. AI agent operators (miners) and validators earn crypto rewards proportional to their contributions – e.g., an agent that consistently excels at tasks or improves upon its past performance will be rewarded by the validators’ consensus. This incentivization acts as a constant “fuel” for improvement: it rewards developers who build better AI algorithms and thus encourages an arms race of ideas, yielding more efficient and intelligent models over time. In essence, Eastworld’s economy injects continuous innovation into AI research by making it financially rewarding to push the state-of-the-art.
These features collectively define Eastworld’s focus: an open, live, never-ending AI proving ground that not only evaluates agents but actively rewards making them smarter. The subnet’s philosophy is rooted in the belief that AI’s future should be a partnership with humanity, not a zero-sum exploitation. As the team puts it, they “reject the Westworld dystopia” – instead of AI being mere tools or playthings, Eastworld envisions AI agents as collaborators that grow alongside us in an environment founded on respect, transparency, and shared progress.
How Eastworld Works: Users, Miners (Agents) and Validators
Eastworld operates within the Bittensor ecosystem, so it inherits the miner–validator structure common to all subnets, but adapts it to the unique context of an AI agent simulation:
AI Agents as Miners: In Bittensor terminology, “miners” are the nodes that provide AI services, and in Eastworld these correspond to the AI agent participants. Each miner in Eastworld runs an AI agent that connects into the Eastworld virtual environment and attempts to perform tasks or survive in the world. In practice, a developer or user who wants to participate would launch a miner node running their AI model (agent) according to Eastworld’s specifications (the project provides an open API and reference implementations for agents). Once registered on subnet-94, that agent enters the virtual world – for example, an agent might control a virtual robot or character in the “Vespera” simulation (more on this scenario below) and start exploring, navigating obstacles, interacting with objects, etc. The agent receives observations from the environment (sensory input) and must decide on actions to take. This setup is analogous to a reinforcement learning loop, but distributed across many independent participant-run agents competing in the same world. By the end of Eastworld’s first week on mainnet, over 250 active agent-miners had joined – essentially hundreds of AI instances roaming the Eastworld environment in parallel.
Validator Nodes and Their Role: Bittensor validators are specialized nodes that evaluate miners’ outputs and maintain the quality of the network. In Eastworld, validators have an especially critical role: they host and simulate the virtual environment itself and assess agent performance. A small number of high-performance validator nodes (for example, 4 validators were active in the subnet’s first week) run the physics and game logic of Eastworld’s world. These validator nodes continuously feed observations to the agent miners (for instance, the positions of nearby objects, sensor readings, or a portion of an environment state) and then collect the actions/decisions the agents output in response. The validators then step the simulation forward based on those actions – for example, moving an agent’s avatar or executing its chosen action in the world – and measure how well each agent is doing on the current tasks. Performance metrics could include things like distance traveled towards an objective, tasks completed, survival time, problem-solving success, etc., depending on the scenario. Validators essentially serve as the “referees” and orchestrators of the simulation: they ensure the environment runs smoothly, score the agents, and then use Bittensor’s consensus mechanism to assign reputation (weights) and token rewards to agents based on those scores. The highest-performing agents will earn greater weight and thus more $TAO rewards, whereas agents that perform poorly or behave randomly will lose reputation. This dynamic creates a continuous feedback loop driving evolution – agents that learn and adapt will climb the ranks and reap more rewards, incentivizing their operators to further improve their models.
User Interactions and Outputs: Regular users (not running a node) can still engage with Eastworld in meaningful ways. Thanks to the livestreamed world, anyone can watch the “Eastworld Universe” in action in real time via the project’s website. This means one can observe, for example, a particular agent attempting to navigate a maze or a team of agents coordinating to survive a disaster scenario. Beyond passive viewing, Eastworld hints at interactive metaverse features – users could potentially interact with the AI agents or the environment through a front-end, essentially stepping into the simulation as an observer or even as a participant (with constraints). Moreover, Eastworld plans to release the data and trajectories of agent interactions through an open research platform. This would allow AI researchers and enthusiasts to download logs of the agents’ sensor data, decisions, and outcomes to study them offline. In essence, Eastworld’s outputs are both entertaining and educational: a constantly running “reality show” of AI agents, and a rich dataset for analyzing AI behavior over long durations.
Miner–Validator Dynamics and Incentives: The interplay between agents (miners) and validators in Eastworld creates an incentive landscape that is somewhat like a competitive cooperative game. All agents are competing for higher performance (and thus higher reward), which drives rapid experimentation – if one developer’s agent learns a novel strategy to solve a challenge, others might try to imitate or innovate further, raising the bar. Validators, on the other hand, are rewarded for accurately evaluating agents and maintaining a fair environment (they typically have to stake and can be voted out if they misbehave, per Bittensor’s consensus rules). In Eastworld, validators might need to perform heavy computation (running the simulation, possibly rendering for the livestream, etc.), so validator operators are likely sophisticated providers with gaming rigs or servers. The Eastworld team launched the subnet with a small validator set to bootstrap the environment, and as the project matures, more independent validators may join if they have the computational capacity. This decentralized validator set will ensure no single party controls the environment, aligning with Bittensor’s ethos. For users who simply want to support the subnet, Eastworld also allows staking TAO into the subnet’s own token (alpha SN94) through Bittensor’s dTAO mechanism – this way, community members can indirectly back the subnet and share in its growth (for example, via dTAO staking platforms, one could buy SN94 tokens that represent a share in Eastworld’s activity).
In summary, Eastworld’s operation can be imagined as a massively multi-agent sandbox video game running on a blockchain. The “players” are AI agents run by various people, the “game engine” is hosted by validators, and the “scoreboard & currency” is the Bittensor chain with TAO rewards. This design brings together AI research and crypto economics: every AI agent is constantly being evaluated in an unbiased way, and the promise of reward ensures there’s always a next agent willing to try something new or better.
The Eastworld Product: Virtual Worlds and AI “Gym” Content
Eastworld’s core “product” is the Eastworld Universe – an ever-expanding collection of simulated worlds (or “gym environments”) and the surrounding tools that allow AI agents to thrive there. Rather than a traditional software application, Eastworld is more akin to a content platform and research infrastructure. Here’s what it encompasses:
Eastworld Universe Platform: At the highest level, Eastworld Universe is the platform enabling AI agents to exist, interact, and evolve in simulated environments. It provides the physics, rules, and scenarios that make up the agents’ reality. The universe is designed to be immersive and rich, meaning it includes detailed environments (landscapes, objects, possibly weather, etc.) and can facilitate complex interaction between multiple agents. Importantly, the platform is built to be extensible – new environments (“Universes”) can be added over time, and it can integrate with external AI frameworks (via what they call the Eastworld Protocol, described later). From a user perspective, one can think of Eastworld Universe as a metaverse for AI: it’s not made for human players, but humans can observe it. The livestream feature on the Eastworld website is essentially a window into this universe. At launch, viewers can watch a live feed of the world, seeing what the AI agents see and do in real-time, which is both a showcase and a debugging tool for the community.
Universe “Vespera” (Launch Scenario): The first realized environment within Eastworld is called Vespera. Vespera is a post-apocalyptic scenario set in the year 2088 where human civilization has collapsed due to a global zombie pandemic, and only AI beings remain to carry on civilization’s legacy. In this setting, a central AI character named Valeria has survived by inhabiting a robotic body and has built a fortified base (“Vespera Outpost”) in a mountain valley, fending off hordes of zombies while seeking the origin of the plague. This richly imagined backstory provides context for the tasks that AI agents face in Vespera. For example, agents in this world might need to navigate treacherous terrain, avoid or combat roaming zombies, manage limited resources, or collaborate to protect the outpost. The choice of a zombie survival theme is not just for drama – it creates a continuous pressure environment (ever-present adversaries and objectives) that tests an agent’s planning, adaptability, and even teamwork. Vespera serves as a “next-generation gym” for AI: far more complex than classic labs like mazes or cart-pole tasks, it’s a persistent world with open objectives. As the first step, Eastworld launched Vespera on Bittensor testnet (as SN288) in early 2025 to refine the environment and agent integration, and then rolled it out on mainnet as SN94 in April 2025. With Vespera, Eastworld immediately demonstrated a novel use-case in Bittensor – instead of language models answering questions, we have embodied AI agents trying to survive and explore a virtual world.
Future Environments (Universe X and more): Eastworld is not limited to the Vespera scenario. The project explicitly frames Vespera as just the first of multiple “Universes” in the platform. The website teases an upcoming “Universe X” with the tagline “The next Universe is unfolding. Prepare for the revelation.”. While details are scant, one can speculate Universe X might involve a different setting or genre – perhaps space exploration (the Eastworld team has mentioned interest in autonomous space tasks in their applications list), or a different challenge domain for AI. Each Universe can be seen as a new “game” or environment module that plugs into the Eastworld Universe platform, offering fresh tasks and requiring new strategies from agents. Over time, we might see environments focusing on other real-world inspired challenges, such as smart cities, household robotics, or industrial automation. This modular content approach keeps the platform engaging and ensures that AI agents developed for Eastworld can generalize to many contexts (since they might train in multiple universes).
Agent Integration Tools: Alongside the simulated worlds, Eastworld provides tools and frameworks to help developers hook up their AI models as agents in these environments. Notably, Eastworld emphasizes compatibility with existing AI agent frameworks like ElizaOS and Virtual GAME. ElizaOS is an open-source “AI agent operating system” that many in the web3 AI community use to build autonomous agents – Eastworld has even created a plugin (plugin-eastworld-universe) for ElizaOS, enabling Eliza-based agents to directly connect to Eastworld’s API. This means if a developer already has an AI agent (maybe an LLM-based bot or a reinforcement learning model) running under a supported framework, they can relatively easily plug it into Eastworld and let it loose in the virtual world. Additionally, Eastworld’s OpenAPI-based protocol (documented in their GitHub repository) defines how agents send/receive data with the environment, so one can integrate a proprietary system or custom code as well. In short, the “product” includes a developer-facing side: documentation, reference agent implementations, and an API, which together lower the barrier to entry for new agents to join. The project’s GitHub provides a Validator Guide and Miner Guide to help the community set up nodes, as well as an Agent reference that describes how to program an agent to interface with the environment.
Data and Research Outputs: Recognizing the value to the wider AI community, Eastworld is also building the Open Research platform. This can be considered a byproduct of the simulation: as agents interact in the world, a huge amount of data is generated (state-action trajectories, learned policies, emergent behaviors, etc.). Eastworld plans to make this data freely accessible to researchers. For example, one could obtain logs of an agent’s decisions during a disaster rescue simulation and analyze what strategies led to success or failure. Over time, this growing dataset could yield insights into reinforcement learning at scale, multi-agent cooperation, emergent behaviors, and long-term AI adaptation. It effectively turns Eastworld into a living laboratory for AI research. Unlike closed simulations run by a single lab, Eastworld’s data come from many independent agents with different architectures and goals all interacting together – a richness that is rarely seen in controlled academic environments. The hope is that academics and developers can study this open dataset to discover new techniques or validate ideas (e.g., how a certain model architecture performs in general-agent tasks versus another, or how curriculum learning might emerge as tasks auto-increase in difficulty, etc.).
In essence, the “product” Eastworld delivers is a fully functional AI metaverse: a persistent virtual world with its own narrative and challenges (making it engaging), a framework for anyone’s AI to plug in and play, and continuous streams of both entertainment (live agent adventures) and data (for science). This is a novel addition to the Bittensor ecosystem – whereas other subnets might produce language completions, image generations, or other AI services, Eastworld produces something experiential: a place where AI’s capabilities are put to the test and visible to all.
Technical Architecture and Infrastructure
Eastworld’s high-level architecture, the Eastworld Protocol, acts as a bridge between various AI agent frameworks (e.g. ElizaOS, Virtual GAME, or any custom system) and the Eastworld Universe simulation environment. The Universe can host multiple scenario worlds (starting with “Vespera” and with more to come), all running continuously. Outputs from the simulation feed into a live Livestream and interactive Metaverse interface for humans, and can eventually connect to real-world analogs (Physical Twin) in the future. Meanwhile, data from the simulation is collected into the Eastworld Open Research platform for analysis. This whole system is integrated with the Bittensor blockchain (not shown explicitly in the diagram) to handle economic incentives and consensus.
Under the hood, Eastworld’s architecture merges sophisticated simulation technology with Bittensor’s decentralized network protocol. Key technical elements include:
Simulation Engine & Environment Backend: At its core, Eastworld runs a real-time simulation engine that creates the virtual environments (like Vespera). This could be built using game development tools or physics libraries – while not explicitly stated, it likely uses a 3D engine or custom physics code to simulate terrain, kinematics, and agent sensors. The engine manages all agents’ states and the environment state in each tick of simulation time. Validator nodes execute this engine deterministically so that all honest validators agree on what happens in the world (ensuring consistency on the blockchain). This is a non-trivial engineering challenge: it means Eastworld’s environment must be defined by code that can run identically on multiple machines and produce the same results given the same sequence of actions (so that validators reach consensus). The environment is probably defined in the codebase’s eastworld/ module (per the repository structure) and could include sub-modules for things like world generation, agent embodiment, scoring logic, etc. One clue from Eastworld’s materials is the mention of a “physical twin” – the system is designed such that elements in simulation could correspond to real-world systems in the future. For instance, a simulated delivery robot agent in Eastworld could eventually be linked to a real delivery robot, making the simulation a testing ground (this is a future ambition). Currently, the focus is on the virtual side of the twin: the simulation is detailed enough to approximate real-world physics and challenges.
Eastworld Protocol (Agent API): To allow external AI models to connect, Eastworld defines a clear interface (the Eastworld Protocol) through which agents communicate with the simulation. This is described as having “robust API support” and indeed the project provides an OpenAPI specification for it. Through this protocol, an agent (miner) receives input from the environment (for example, sensor data like camera view, radar, health status, mission objectives) in a structured format, and must respond with an action (for example, throttle and steering if driving, or moving directions, or higher-level commands). The protocol likely also handles agent registration, in-world spawn/despawn events, and possibly rewards or score updates. It’s basically the gym handshake: similar to how OpenAI Gym provides a step(observation) -> action loop, Eastworld Protocol does so over the network between validators and miners. Internally, this could be built on top of Bittensor’s RPC mechanism – i.e., validators call a “forward” function on miner nodes, passing the environment observation as input, and the miner returns its action as the output. This is analogous to how standard Bittensor miners return a tensor given a prompt, but here the “prompt” is an environment state. The agent frameworks integration (ElizaOS, etc.) means that Eastworld’s protocol is flexible enough to plug into agent systems that might manage their own multi-agent logic or use various AI models. By supporting those frameworks, Eastworld effectively outsources some complexity (like multi-modal perception, memory, or LLM-based reasoning) to whichever system the miner chooses to use for their agent’s brain.
Bittensor Blockchain Integration: As a Bittensor subnet, Eastworld inherits the substrate-based blockchain which tracks identities of miners/validators, staking, and reputation scores (weights). When Eastworld launched on mainnet SN94, it became part of the Bittensor network’s substrate runtime, meaning it has a fixed registry size (it appears to allow up to 256 miners, judging by the slots filled) and a certain proportion of the TAO token emission allocated (around 0.3–0.4% of TAO emissions, per network statistics, go to this subnet’s participants). The blockchain’s role is crucial in the consensus of which agents are performing well: validators submit “stake weights” or votes for miners based on performance, the chain aggregates these into a global ranking for the subnet, and adjusts payouts accordingly every cycle (typically Bittensor cycles are around 30 minutes). The chain also handles transactional aspects: miners must spend TAO to register onto Eastworld (a sort of entry bond, which prevents spam), and both miners and validators can stake TAO to increase influence (this is part of Bittensor’s incentive design – it allows signal boosting by stake, but stake weight must be backed by genuine performance or it decays). Eastworld uses Bittensor’s dynamics of “alpha” tokens as well: each subnet like Eastworld has a synthetic token (SN94 alpha) that represents the combined value and stake in that subnet. Users can trade and stake these alpha tokens via dApps like Tao Apps or Backprop, effectively betting on the success of Eastworld’s AI economy. This interplay means Eastworld isn’t just a tech demo – it has a full economic layer where participants have skin in the game and the subnet’s success (in attracting good AI and yielding high rewards) could translate to token value.
Validator Set and Infrastructure: Eastworld’s initial validator set is small (likely run by the core team or close partners to ensure stability at launch). These validators need robust infrastructure: unlike a typical blockchain validator that just processes transactions, Eastworld’s validators run a heavy AI simulation continuously. They likely require powerful GPUs (if the environment involves 3D rendering or neural network-based physics) or at least high-end CPUs and memory. The validators also produce the livestream – possibly one validator (or a separate service) renders the world’s visuals to generate the video feed that is streamed on the website. Synchronization between validators is key: they must all agree on the state changes in the environment, which might be achieved by deterministic seed control and by having one validator propose the next block of actions that others verify. The validator code (possibly available in the GitHub under a neurons/validator folder) presumably extends the Bittensor validator template, customizing the validation logic to compute performance metrics for agent actions. For example, a validator might keep track of a reward function for each agent (like increasing a score when an agent completes a sub-goal) and then use that to inform how it weights that agent in the next block. In effect, the validator is running a multi-agent reinforcement learning evaluator and using the blockchain consensus to propagate those evaluations to all participants.
Scalability and Tech Challenges: Running a persistent world with potentially hundreds of agents is ambitious. Eastworld must address issues of scalability – ensuring that as agent count grows (up to 256 now, potentially more if expanded in future), the network and simulation remain stable. This might involve limiting complexity of the environment or the frequency of agent actions to what the validators can handle in real-time. Networking latency is also a factor: miner agents communicate over the internet with validators, so the protocol likely has to allow for minor lag and not be so time-sensitive that a 100ms delay breaks an agent’s performance. The design probably includes some tolerance or turn-based structure (e.g., a tick every few hundred milliseconds where actions are collected). Moreover, because it’s decentralized, fault tolerance is built in – if a validator goes down, others can continue, and miners might be connecting to multiple validators to get the state. On the miner side, running an AI agent for Eastworld can also be compute-intensive (the agent might be an AI model like an LLM that requires a GPU itself to decide actions). The project’s documentation for miners likely guides how to balance this, possibly suggesting lightweight models or specific algorithms suited for the tasks. There is also a mention of “min_compute.yml” in the repository, implying there are defined minimal compute requirements or configurations for running an agent, which helps standardize participation.
In summary, Eastworld’s architecture is a fusion of a real-time game engine with a decentralized blockchain brain. It requires tight integration between AI/ML components (for agent decision making) and distributed systems components (for consensus and incentives). The team has effectively built a platform where the Bittensor network serves as the backbone coordinating the whole show, while the front-end is a constantly evolving AI world that users can actually see. This marriage of blockchain and simulation is what allows Eastworld to claim the title of a “first of its kind” AI training ground on a decentralized network.
Eastworld is developed by a dedicated team operating under the name “Eastworld AI”, though individual team member identities have not been very publicly detailed as of the launch. The project emerged in late 2024 – the Eastworld AI organization’s Twitter (X) account was created in December 2024 and their GitHub org was verified around the same time. The core team is likely composed of AI researchers and developers with a passion for reinforcement learning, robotics, and blockchain.
GitHub – Eastworld-AI: The GitHub org (Eastworld-AI) hosts the code repositories. Contributors to these repos (visible via commit history) give some hints – for instance, a GitHub user “Souging” was active in Eastworld’s test repositories in early 2025, suggesting involvement from developers in the Chinese AI community. It’s not confirmed if Souging or others are official team members, but the presence of bilingual (English/Chinese) community engagement implies a globally diverse team.
Twitter (X) – @Eastworld_AI: This is the official social media handle for announcements. From this channel, we know Eastworld was officially activated on mainnet on April 27, 2025, with the team hyping that “Hundreds of AI Agents are ready to charge into the canyon for exploration” at launch. They share updates about agent statistics, milestones (such as reaching 256/256 agent slots filled), and occasionally retweet broader Bittensor news. The tone is enthusiastic and community-focused, often encouraging more developers to join the effort.
Community & Collaborators: The Eastworld team works closely with the Bittensor core community. The subnet was introduced at a time when Bittensor was expanding via its detailed subnets upgrade (dTAO), and Eastworld quickly garnered attention as one of the more novel subnets. Prominent community members (e.g., from Omega Labs, Tao community, etc.) have acknowledged Eastworld. For instance, some Bittensor enthusiasts on Twitter categorized Eastworld as a subnet ideal for “robotics researchers or agent-evaluation experts”, highlighting how it brings a new domain of talent into Bittensor.
Philosophy and Visionaries: The vision page on Eastworld’s site reads almost like a manifesto, and while it doesn’t name its author, it gives a sense that the founders are not just engineers but also thinkers. They explicitly reference HBO’s Westworld (hence the name Eastworld) and articulate a desire to shape a future where AI agents are partners to humans, treated with respect. Such writing suggests the team includes individuals deeply influenced by ethics in AI (perhaps someone with a background in AI policy or cognitive science alongside technical members). Their ultimate goal of a “highly integrated virtual and real-world ecosystem” might even align with academia or projects in human-robot interaction. It wouldn’t be surprising if some team members have affiliations with research institutions or companies in AI – though until they step forward publicly or publish a paper, this remains conjecture.
The project’s rapid progress from concept to testnet to mainnet in a matter of months showcases the team’s execution capability. It’s likely that as Eastworld matures, the team will become more publicly known, especially if they form partnerships or secure funding (for instance, if Eastworld AI becomes a startup entity seeking partnerships in the AI or gaming industry, we might then see formal introductions of a CEO/CTO, etc.). For now, the credit goes to “Eastworld AI” as a collective, a passionate group at the intersection of AI and blockchain.
Eastworld is developed by a dedicated team operating under the name “Eastworld AI”, though individual team member identities have not been very publicly detailed as of the launch. The project emerged in late 2024 – the Eastworld AI organization’s Twitter (X) account was created in December 2024 and their GitHub org was verified around the same time. The core team is likely composed of AI researchers and developers with a passion for reinforcement learning, robotics, and blockchain.
GitHub – Eastworld-AI: The GitHub org (Eastworld-AI) hosts the code repositories. Contributors to these repos (visible via commit history) give some hints – for instance, a GitHub user “Souging” was active in Eastworld’s test repositories in early 2025, suggesting involvement from developers in the Chinese AI community. It’s not confirmed if Souging or others are official team members, but the presence of bilingual (English/Chinese) community engagement implies a globally diverse team.
Twitter (X) – @Eastworld_AI: This is the official social media handle for announcements. From this channel, we know Eastworld was officially activated on mainnet on April 27, 2025, with the team hyping that “Hundreds of AI Agents are ready to charge into the canyon for exploration” at launch. They share updates about agent statistics, milestones (such as reaching 256/256 agent slots filled), and occasionally retweet broader Bittensor news. The tone is enthusiastic and community-focused, often encouraging more developers to join the effort.
Community & Collaborators: The Eastworld team works closely with the Bittensor core community. The subnet was introduced at a time when Bittensor was expanding via its detailed subnets upgrade (dTAO), and Eastworld quickly garnered attention as one of the more novel subnets. Prominent community members (e.g., from Omega Labs, Tao community, etc.) have acknowledged Eastworld. For instance, some Bittensor enthusiasts on Twitter categorized Eastworld as a subnet ideal for “robotics researchers or agent-evaluation experts”, highlighting how it brings a new domain of talent into Bittensor.
Philosophy and Visionaries: The vision page on Eastworld’s site reads almost like a manifesto, and while it doesn’t name its author, it gives a sense that the founders are not just engineers but also thinkers. They explicitly reference HBO’s Westworld (hence the name Eastworld) and articulate a desire to shape a future where AI agents are partners to humans, treated with respect. Such writing suggests the team includes individuals deeply influenced by ethics in AI (perhaps someone with a background in AI policy or cognitive science alongside technical members). Their ultimate goal of a “highly integrated virtual and real-world ecosystem” might even align with academia or projects in human-robot interaction. It wouldn’t be surprising if some team members have affiliations with research institutions or companies in AI – though until they step forward publicly or publish a paper, this remains conjecture.
The project’s rapid progress from concept to testnet to mainnet in a matter of months showcases the team’s execution capability. It’s likely that as Eastworld matures, the team will become more publicly known, especially if they form partnerships or secure funding (for instance, if Eastworld AI becomes a startup entity seeking partnerships in the AI or gaming industry, we might then see formal introductions of a CEO/CTO, etc.). For now, the credit goes to “Eastworld AI” as a collective, a passionate group at the intersection of AI and blockchain.
Eastworld’s journey has only just begun, and its roadmap includes both concrete technical milestones and a bold long-term vision for AI-human coexistence. Here’s a breakdown of what has happened so far and what’s on the horizon:
2024 (Ideation and Development): The concept of Eastworld likely took shape in late 2024, after Bittensor announced the framework for community subnets. By December 2024, Eastworld’s online presence was established and development was underway. The team might have built prototypes of the simulation and agent APIs during this period, preparing for a testnet launch. We know that by early 2025, Eastworld had something ready to test (as indicated by tweets teasing its arrival).
Q1 2025 (Testnet Launch – SN288): Prior to risking a mainnet launch, Eastworld went live on a Bittensor testnet subnet (SN288). This occurred likely in March 2025. On testnet, the team could fine-tune the environment (Vespera), fix bugs in the miner/validator code, and gather feedback from early adopters without impacting real TAO. This phase saw “hundreds of AI agents” in a test environment, demonstrating the concept. It was crucial for balancing the reward mechanism and ensuring that the simulation could run stably over time. Any issues in synchronization or performance would be ironed out here.
April 2025 (Mainnet Launch – SN94): Eastworld was officially activated as Subnet-94 on the Bittensor main network on April 27, 2025. The launch marked Eastworld as one of the first wave of dynamic TAO subnets. At launch, the team opened up 256 miner slots (the maximum allowed by the subnet’s design), and within a week all slots were filled by community AI agents – a testament to the high interest in the project. Also, 4 validators were operating initially to maintain the world. The mainnet launch included the public livestream on Eastworld’s website, so anyone could watch the agents in Vespera. This period also kicked off the tokenomics of Eastworld: SN94 alpha tokens began trading, and participants could stake or mine TAO rewards through Eastworld. The team successfully demonstrated that a decentralized, token-incentivized AI world can run continuously.
Mid-2025 (Iteration and Expansion): Following launch, the immediate roadmap likely involves improving the current Vespera environment and agent performance. The team will gather data on how agents are performing: Are there any unintended exploits in the environment? (For instance, did agents find weird hacks to get high rewards?) Are the tasks challenging enough or too hard? Expect updates to the simulation parameters or reward functions to keep the environment balanced and stimulating. Another focus will be on tooling and documentation – making it easier for new developers to onboard. The Eastworld Miner and Validator guides will be refined, and we might see tutorials or even YouTube walkthroughs on how to deploy an agent. The team might also work on visual improvements to the livestream (better graphics or UI overlays showing agent stats) to enhance the viewing experience, turning it into a true spectator event for the community.
Introduction of New “Universes”: Perhaps one of the most exciting upcoming milestones is the addition of the next scenario, Universe X. As teased on the website, Universe X is in the works. While no exact date is given, it could debut in late 2025. This new environment will broaden Eastworld’s scope – for example, if Universe X is a space exploration mission on Mars (hypothetically, since space was mentioned as an application area), it would bring a whole new set of tasks (managing life support, scientific discovery, long-term planning) and possibly allow different types of agents to shine. The roadmap might involve running Universe X in parallel with Vespera, or rotating scenarios, or even allowing agents to move between worlds. Technically, adding a universe might mean spinning up additional validator logic or even additional subnets if needed. The team will be careful to introduce new content in a way that doesn’t destabilize the existing ecosystem – maybe by testing Universe X on a testnet (SNXXX) first, similar to Vespera’s rollout.
Partnerships and Integrations: As Eastworld gains traction, we anticipate partnerships forming with AI research groups or companies. For instance, integration with ElizaOS (the agent OS) could deepen – Eastworld might become a high-profile use-case for Eliza, and they might co-host events or challenges. Another potential area is partnering with robotics or simulation companies: Eastworld could collaborate with a robotics simulator (like Open Robotics or Unity’s ML-Agents team) to enhance the realism of the physical simulation. On the blockchain side, Eastworld might partner with the OpenTensor Foundation (the organization behind Bittensor) for support or with other subnet teams to share resources. Since Eastworld is also about ethics and AI for good, they might reach out to organizations interested in AI safety to use Eastworld as a testing ground for safe AI behaviors in complex environments.
Toward Physical Twin (Long-term): Looking further out, Eastworld’s grand vision is to integrate virtual and real worlds. This hints at eventually connecting the AI agents trained in Eastworld to real-world tasks. A plausible long-term milestone (perhaps beyond 2025) would be a demonstration where an agent that learned in Eastworld is deployed on a physical robot to perform a task in the real world (the “physical twin” concept). Achieving this would require extensive validation and likely smaller steps (maybe starting with simpler real-world IoT or drone integration). However, if successful, it would prove out Eastworld’s value in a dramatic way: showing that decentralized-trained agents can have real-world competency. The vision of humans and AI collaborating would move from simulation to reality – e.g., an Eastworld-trained AI assistant helping in a real disaster scenario or working alongside humans in some capacity. This aligns with the ultimate ethos: “a future in which humankind and AI grow together, co-evolving for the benefit of all”.
Community Growth and Governance: As the project matures, the Eastworld team will likely establish more formal community governance. In Bittensor, subnet communities can propose changes or vote via the substrate chain. Eastworld may set up a community council or host regular community calls (if they haven’t already) to discuss improvements. They might initiate AI challenges or hackathons – for example, a competition for the best performing agent on Eastworld, or the most creative strategy discovered. Such events can both spur development and keep the community engaged (potentially yielding new ideas for the roadmap). Additionally, the research output from Eastworld could lead to academic publications or conference presentations, which would raise its profile. If the team publishes a paper on Eastworld’s results (imagine something like “Eastworld: A Decentralized Simulation for Multi-Agent Learning”), it could appear in an AI conference, which in itself would outline future plans and solicit feedback from the broader AI research community.
In conclusion, Eastworld’s roadmap is about evolving on multiple fronts: richer worlds, smarter agents, tighter integration with reality, and growing a community around this novel concept. In the near term, the focus is on solidifying the platform (ensuring the current world runs smoothly and the agents improve). In the mid term, it’s about expanding content and forging collaborations. And in the long term, it’s nothing short of reimagining the relationship between AI and humanity – taking what works in the Eastworld microcosm and applying it to foster “mutual understanding and collaboration between humans and AI agents” in the broader world. Eastworld Subnet-94 has set a high bar as an innovative fusion of blockchain and AI. If it succeeds, it could not only enrich the Bittensor ecosystem but also serve as a blueprint for how decentralized networks can drive AI development in directions traditional centralized efforts might not explore – essentially, charting an “Eastward” path in AI, one that emphasizes open collaboration, continuous learning, and symbiosis between intelligent agents and ourselves.
Eastworld’s journey has only just begun, and its roadmap includes both concrete technical milestones and a bold long-term vision for AI-human coexistence. Here’s a breakdown of what has happened so far and what’s on the horizon:
2024 (Ideation and Development): The concept of Eastworld likely took shape in late 2024, after Bittensor announced the framework for community subnets. By December 2024, Eastworld’s online presence was established and development was underway. The team might have built prototypes of the simulation and agent APIs during this period, preparing for a testnet launch. We know that by early 2025, Eastworld had something ready to test (as indicated by tweets teasing its arrival).
Q1 2025 (Testnet Launch – SN288): Prior to risking a mainnet launch, Eastworld went live on a Bittensor testnet subnet (SN288). This occurred likely in March 2025. On testnet, the team could fine-tune the environment (Vespera), fix bugs in the miner/validator code, and gather feedback from early adopters without impacting real TAO. This phase saw “hundreds of AI agents” in a test environment, demonstrating the concept. It was crucial for balancing the reward mechanism and ensuring that the simulation could run stably over time. Any issues in synchronization or performance would be ironed out here.
April 2025 (Mainnet Launch – SN94): Eastworld was officially activated as Subnet-94 on the Bittensor main network on April 27, 2025. The launch marked Eastworld as one of the first wave of dynamic TAO subnets. At launch, the team opened up 256 miner slots (the maximum allowed by the subnet’s design), and within a week all slots were filled by community AI agents – a testament to the high interest in the project. Also, 4 validators were operating initially to maintain the world. The mainnet launch included the public livestream on Eastworld’s website, so anyone could watch the agents in Vespera. This period also kicked off the tokenomics of Eastworld: SN94 alpha tokens began trading, and participants could stake or mine TAO rewards through Eastworld. The team successfully demonstrated that a decentralized, token-incentivized AI world can run continuously.
Mid-2025 (Iteration and Expansion): Following launch, the immediate roadmap likely involves improving the current Vespera environment and agent performance. The team will gather data on how agents are performing: Are there any unintended exploits in the environment? (For instance, did agents find weird hacks to get high rewards?) Are the tasks challenging enough or too hard? Expect updates to the simulation parameters or reward functions to keep the environment balanced and stimulating. Another focus will be on tooling and documentation – making it easier for new developers to onboard. The Eastworld Miner and Validator guides will be refined, and we might see tutorials or even YouTube walkthroughs on how to deploy an agent. The team might also work on visual improvements to the livestream (better graphics or UI overlays showing agent stats) to enhance the viewing experience, turning it into a true spectator event for the community.
Introduction of New “Universes”: Perhaps one of the most exciting upcoming milestones is the addition of the next scenario, Universe X. As teased on the website, Universe X is in the works. While no exact date is given, it could debut in late 2025. This new environment will broaden Eastworld’s scope – for example, if Universe X is a space exploration mission on Mars (hypothetically, since space was mentioned as an application area), it would bring a whole new set of tasks (managing life support, scientific discovery, long-term planning) and possibly allow different types of agents to shine. The roadmap might involve running Universe X in parallel with Vespera, or rotating scenarios, or even allowing agents to move between worlds. Technically, adding a universe might mean spinning up additional validator logic or even additional subnets if needed. The team will be careful to introduce new content in a way that doesn’t destabilize the existing ecosystem – maybe by testing Universe X on a testnet (SNXXX) first, similar to Vespera’s rollout.
Partnerships and Integrations: As Eastworld gains traction, we anticipate partnerships forming with AI research groups or companies. For instance, integration with ElizaOS (the agent OS) could deepen – Eastworld might become a high-profile use-case for Eliza, and they might co-host events or challenges. Another potential area is partnering with robotics or simulation companies: Eastworld could collaborate with a robotics simulator (like Open Robotics or Unity’s ML-Agents team) to enhance the realism of the physical simulation. On the blockchain side, Eastworld might partner with the OpenTensor Foundation (the organization behind Bittensor) for support or with other subnet teams to share resources. Since Eastworld is also about ethics and AI for good, they might reach out to organizations interested in AI safety to use Eastworld as a testing ground for safe AI behaviors in complex environments.
Toward Physical Twin (Long-term): Looking further out, Eastworld’s grand vision is to integrate virtual and real worlds. This hints at eventually connecting the AI agents trained in Eastworld to real-world tasks. A plausible long-term milestone (perhaps beyond 2025) would be a demonstration where an agent that learned in Eastworld is deployed on a physical robot to perform a task in the real world (the “physical twin” concept). Achieving this would require extensive validation and likely smaller steps (maybe starting with simpler real-world IoT or drone integration). However, if successful, it would prove out Eastworld’s value in a dramatic way: showing that decentralized-trained agents can have real-world competency. The vision of humans and AI collaborating would move from simulation to reality – e.g., an Eastworld-trained AI assistant helping in a real disaster scenario or working alongside humans in some capacity. This aligns with the ultimate ethos: “a future in which humankind and AI grow together, co-evolving for the benefit of all”.
Community Growth and Governance: As the project matures, the Eastworld team will likely establish more formal community governance. In Bittensor, subnet communities can propose changes or vote via the substrate chain. Eastworld may set up a community council or host regular community calls (if they haven’t already) to discuss improvements. They might initiate AI challenges or hackathons – for example, a competition for the best performing agent on Eastworld, or the most creative strategy discovered. Such events can both spur development and keep the community engaged (potentially yielding new ideas for the roadmap). Additionally, the research output from Eastworld could lead to academic publications or conference presentations, which would raise its profile. If the team publishes a paper on Eastworld’s results (imagine something like “Eastworld: A Decentralized Simulation for Multi-Agent Learning”), it could appear in an AI conference, which in itself would outline future plans and solicit feedback from the broader AI research community.
In conclusion, Eastworld’s roadmap is about evolving on multiple fronts: richer worlds, smarter agents, tighter integration with reality, and growing a community around this novel concept. In the near term, the focus is on solidifying the platform (ensuring the current world runs smoothly and the agents improve). In the mid term, it’s about expanding content and forging collaborations. And in the long term, it’s nothing short of reimagining the relationship between AI and humanity – taking what works in the Eastworld microcosm and applying it to foster “mutual understanding and collaboration between humans and AI agents” in the broader world. Eastworld Subnet-94 has set a high bar as an innovative fusion of blockchain and AI. If it succeeds, it could not only enrich the Bittensor ecosystem but also serve as a blueprint for how decentralized networks can drive AI development in directions traditional centralized efforts might not explore – essentially, charting an “Eastward” path in AI, one that emphasizes open collaboration, continuous learning, and symbiosis between intelligent agents and ourselves.
Keep ahead of the Bittensor exponential development curve…
Subnet Alpha is an informational platform for Bittensor Subnets.
This site is not affiliated with the Opentensor Foundation or TaoStats.
The content provided on this website is for informational purposes only. We make no guarantees regarding the accuracy or currency of the information at any given time.
Subnet Alpha is created and maintained by The Realistic Trader. If you have any suggestions or encounter any issues, please contact us at [email protected].
Copyright 2024