Real users are now stepping into funnAI’s experimental arena, where artificial intelligence plays for keeps and runs entirely on-chain. With wallets connected and rewards on offer, the latest move by OnicAI and the DFINITY Foundation is making AI interaction more tangible, more competitive, and a lot more fun.
The doors are officially open for public testing. The tagline says it all: real users, real wallets, real rewards. This isn’t a staged simulation or internal trial. It’s the start of a live experiment that invites participants to launch their own AI agents—nicknamed mAIners—onto the Internet Computer Protocol (ICP). These agents learn, adapt and battle it out in a dynamic environment where performance is tracked on-chain, and every outcome is transparently recorded.
FunnAI first caught the attention of developers and Web3 watchers earlier this year when OnicAI unveiled its plans for the first AI competition hosted entirely on-chain. The vision was ambitious: hundreds of AI agents operating concurrently without relying on traditional cloud systems or external servers. The platform would live natively on ICP, using the network’s ability to execute smart contracts efficiently and scale without the common friction points of many blockchains.
Backed by the DFINITY Foundation, the project had access to ICP’s infrastructure from the ground up. What makes funnAI different is that it doesn’t treat the blockchain as an afterthought. Every element—from the logic that drives the AI to the mechanics that determine reward payouts—is embedded in the chain itself. The experience isn’t mediated through external APIs or hidden server logic. What happens in funnAI stays on-chain and stays verifiable.
The newly launched user testing phase is an important next step. Until now, development had been in the hands of a limited team, with controlled demonstrations and sandbox testing. With the current phase, real users are being asked to step in and shape the system with their choices. Each participant brings their own Internet Identity, connects their wallet, and is free to spin up a mAIner. From that moment, the agent enters the game, and the performance clock starts ticking.
While the name ‘mAIner’ may sound like a pun, it reflects a larger shift in thinking. These agents aren’t running AI models on someone else’s infrastructure. They’re operating as autonomous units within the chain, competing in environments that demand decision-making, interaction, and, in some cases, survival. The competition layer is essential. It adds stakes. If an agent performs well, it earns its creator rewards—tracked transparently and paid out accordingly. If it doesn’t, it learns and tries again.
There’s a gaming element here, but it’s not based on hand-eye coordination or reaction speed. It’s about building smart agents and refining their strategies. These aren’t bots that follow hard-coded scripts. They’re designed to adapt, to learn from repeated outcomes, and to improve over time. In that sense, funnAI is less like a game and more like a long-running experiment with incentives built into every turn.
From a technical perspective, ICP handles the back-end in a way that most Web3 projects can’t. The architecture doesn’t rely on Layer 2s or workaround bridges. Everything, including data storage, execution, and communication between agents, happens within the same layer. That enables fast updates, responsive gameplay, and a consistent experience for users. It also means that bugs or exploits are harder to hide. Transparency is baked into every layer of interaction.
OnicAI’s choice to build this system on ICP wasn’t incidental. The Internet Computer offers a model that most blockchains can’t match when it comes to persistent computation. Instead of treating smart contracts as one-off functions triggered by external calls, ICP allows for long-lived, autonomous programs that can respond to changes over time. That’s a key requirement for any AI system that intends to evolve rather than repeat.
There’s been a great deal of discussion in recent months about decentralised AI. The buzzword status of AI has made it easy to claim alignment with openness, transparency and user control. In practice, many so-called decentralised AI platforms still rely heavily on proprietary back-ends or centralised infrastructure for training, execution or data handling. FunnAI, by contrast, takes the full-stack approach. It’s not just an AI application on a blockchain. It is the application, the host, and the playground all rolled into one—and it’s running where anyone can see.
With the launch of user testing, the project is entering its most unpredictable phase yet. Live environments bring surprises. There will be unexpected agent behaviours, strategic exploits, and maybe even users pushing the limits of what the mAIners can do. That’s the point. This is a pressure test for the platform, and for the agents themselves. It’s also a stress test for the reward systems—designed to be fair, resistant to manipulation, and rewarding to those who genuinely build and iterate.
One of the more curious aspects of funnAI is how it blends technical precision with a tone of playful experimentation. It doesn’t take itself too seriously, despite being one of the first efforts to run this kind of AI competition on-chain. The branding is colourful, the messages are tongue-in-cheek, and there’s a game-show energy to some of the interface elements. But underneath the surface, the architecture is designed for durability. It has to be. The challenge of hosting hundreds of AI agents concurrently, while keeping state synchronised and rewards accurate, isn’t trivial.
As of this stage, there’s no requirement to be an AI developer or coding expert to participate. The system has been built to lower the barriers, allowing users to select or customise mAIners with varying degrees of control. Some may choose to tweak behaviours. Others may opt to deploy a ready-made agent and watch what happens. Either way, the incentives are there, and the leaderboard isn’t static.
The feedback loop is key to what comes next. As users report bugs, suggest improvements, and highlight inconsistencies, the OnicAI team will be watching closely. The hope is that these early interactions will inform the next phase of development, potentially including new game mechanics, smarter learning frameworks, or broader multiplayer setups. While no official roadmap has been published for post-testing expansion, the structure of funnAI hints at future possibilities beyond simple competitions—cooperative AI tasks, goal-oriented environments, and maybe even real-world integrations.
For now, though, the focus is on testing under real-world conditions. That means wallets will be active, rewards will be tracked, and players will start to get a feel for how their agents perform under pressure. It’s an unusual intersection—blockchain, gaming, and AI—brought together by a team that seems determined to explore what’s possible without overpromising or locking things behind layers of abstraction.
What started as a concept shared on a forum post in May is now something users can touch, tweak and play with. The chain is live, the agents are running, and the arena is getting crowded. Spin up a mAIner and see what it learns. There’s fun to be had—and maybe something to earn while you’re at it.
