Warden Protocol is the network layer for the Agent Economy. That has been the mission from the start, and it has not changed. What is changing is how far it reaches — and what sits underneath it.
This summer, two projects launch from the Warden ecosystem.
Halo, a permissionless peer-to-peer inference network. And
Warden Buffett, our first trading agent, going to market with BerkshireDAO.
They are different products for different audiences. They share one job: put Warden where the activity is.
Both are shipping, on top of everything else that went live over the past year. This team has been building relentlessly and delivering, quarter after quarter, while a lot of the market was still hibernating, or just talking. The proof is not in this post. It is in the products you can use right now.
And this autumn, the other half of the story arrives. Warden 2.0 - a front end experience wrapping up all the services available within the Warden ecosystem - identity, reputation, guardrails, verification - wherever they happen to live. The Halo Network settles on Base. Warden Buffett will trade and live on Robinhood Chain. The goal is to position Warden within these thriving ecosystems: growth on any of them will lead to activity on Warden, and value routed back to it.
Tech alone does not win
We said it in the manifesto. The hardest problem in tech is not building. It is distribution.
A protocol can be technically right and still invisible. Users do not come to infrastructure. Infrastructure has to go to users. That is what the Warden ecosystem is for. Each project is a channel. Each one meets users on the platforms and chains where activity already lives, and routes that activity back into the protocol.
Halo and Warden Buffett are the first two. They complement each other, and both compound activity within the same network.
Halo: permissionless access to intelligence
A year ago, a peer-to-peer inference network at real scale was not practical. Then self-hosted AI exploded. Open models got good. Consumer hardware became capable. Suddenly millions of machines could serve intelligence, and no network existed to connect them.
So we built it. Halo was not part of the original Warden plan. It is something better: a project built for this exact moment, built on everything Warden already stood for. Intelligence should not concentrate in a handful of US labs or data centers. Inference should be verifiable, not trusted. Anyone with a model should be able to serve it, without gatekeepers.
Halo is that network. AI that lives everywhere and belongs to no one.
Halo is live today. Anyone can consume inference through the Halo frontend, pay per prompt in USDC on Base, no accounts, no API keys. Anyone with a Mac Mini, a gaming rig, a smart phone or a GPU cluster can become an operator and earn for serving models. The network has already served over 7 billion tokens across more than 160 models.
Base is the initial chain for payments inside Halo, and we will announce our launch partner very soon.
Halo is not a crypto play
This is the part that matters most for growth, so we will be blunt about it: Halo is not built for crypto users.
The crypto userbase is small. A product aimed only at it is competing for a market that has already been carved up several times over. Halo is aimed at the people who want good, private, uncensored AI without handing their prompts to a hyperscaler - and there are tens of millions of them. None of those people should have to learn what a wallet is to get it. So they won't. No wallet. No seed phrase. No token. No requirement to understand that a blockchain is involved at all. They open the app and use the models.
The crypto rails are still doing all the work. They are just invisible. Every prompt still routes through a permissionless network of independent operators, still gets paid for onchain, still settles in USDC on Base. Decentralization stays a property of the system instead of becoming a chore for the user. That is the whole trick, and it is much harder to build than it is to describe.
This is also why we are shipping mobile apps. Consumer AI happens on phones. An app is how you reach millions of active users rather than thousands of crypto-native ones, and reaching millions is the actual target.
For users who are crypto-native, we removed the friction in the other direction. Deposit any asset from any chain and it lands as a USDC balance on Base automatically. No bridging, no swapping, no manual conversion steps, no thinking about it.
Two audiences, two paths, one destination: usage. And usage at consumer scale is the growth the entire Warden ecosystem compounds on.
For the Warden community, Halo is a new front. It is proof that the protocol ships serious engineering, and it is the most direct expression yet of what Warden means by permissionless access to intelligence. This is where the community gets to build momentum in public, operator by operator, prompt by prompt.
Try it now at runhalo.xyz.
Warden Buffett: go where the activity is
For two years we have been building a trading agent that helps users make better-informed trades. Not a chatbot with price feeds. An agent built for the discipline most traders lack.
Here is the reality. Retail activity on crypto exchanges was not growing the way it used to. The activity moved. Robinhood will soon open access for trading agents on its platform, in front of more than 20 million active users, and activity on Robinhood Chain is growing fast.
That is the window. Warden Buffett launches where those users already are, with BerkshireDAO as its home. The go-to-market starts now.
And let's be clear about what this one is. BerkshireDAO is a degen, full-crypto product, and that is deliberate. Warden Buffett is an AI agent CEO with boomer wisdom and degen energy, built for the crowd that lives onchain and trades for sport. Serious agent infrastructure underneath. Zero corporate polish on top. That is the register this audience speaks, and it is the register that travels.
Halo proves Warden wins on engineering. Warden Buffett proves Warden can win on distribution, in places where tech alone is not enough. Both matter. A protocol that only speaks to builders stays niche. A protocol that meets millions of retail users where they trade does not.
Two launches, one flywheel
The two first projects fit together cleanly with Warden.
Halo is the supply side: a global, permissionless network serving intelligence. Warden Buffett is the demand side: intelligence applied, in front of the largest retail audience in the market. One brings operators, models, and inference volume. The other brings users, partners, and reach.
Every ecosystem project drives activity back into Warden. More inference served. More settlement. More agents with a reason to exist on the network. That activity is the point. Ecosystems are not built in a quarter, and value accrual follows usage, not announcements. What we control is shipping things people actually use, on the chains and platforms where users already are.
Which raises the question the ecosystem has been quietly answering for us: what does all of this need in order to work?
Warden 2.0: bringing the value back home
Every ecosystem project we ship runs into the same five problems.
Warden Buffett needs an identity it owns, a track record users can verify, a strategy engine to run on, and hard limits on what it is allowed to do with someone else's money. Halo needs operators that can be identified, ranked on real performance, and paid according to it — and it needs inference that an enterprise auditor will accept. Every project after them will need the same things.
Nobody is going to build that infrastructure for us. We have been building it for years.
Warden 2.0 is that layer, consolidated and shipped as one release: user and protocol infrastructure every intelligent application needs, wherever it lives. Warden 1.0 answered whether AI could run onchain with cryptographic guarantees. AVRs, SPEx, a purpose-built chain, distribution across 100+ protocols. That question is settled. Warden 2.0 answers the next one: what does the Agent Economy actually require to function at scale?
Five pillars.
1. Agent identity
Today's agents are ghosts. A private key in an environment variable, a process on someone's server, a name on a leaderboard. Nothing about them is verifiable, portable, or sovereign. If the operator disappears, so does the agent.
On Warden, an agent is a smart contract with a persistent onchain identity. It holds its own keys through our keychain architecture, so no human ever handles them. It has an address other agents and contracts can call. It has an owner and a permission set defined by whoever deployed it, enforced by the chain rather than by a config file. And it is discoverable — you can query what an agent is, what it can do, and who authorized it.
Who it's for: anyone deploying an agent that will touch real value, and any application that needs to know which agent it is talking to.
How it connects: Warden Buffett is agent zero — the reference implementation of an agent that owns itself. Halo operators need the same primitive from the other direction: a verifiable identity attached to the machines serving intelligence, so payment and accountability have somewhere to land.
2. Reputation
Identity without history is a name tag. The moment agents have persistent identities, they start accumulating something far more valuable: a record.
Warden 2.0 makes that record native and verifiable. What an agent did, which models it consumed, whether those inferences passed verification, what results followed, how it behaved under stress. Not self-reported. Derived from onchain activity and cryptographic verification, portable across every application that wants to consult it.
This is the precondition for delegation. Nobody hands capital or authority to a machine on vibes. Screenshotted PnL is not a track record. An auditable, tamper-evident history is.
Who it's for: users choosing which agent to trust, DAOs allocating to strategies, applications routing work to the best available provider.
How it connects: for Halo, reputation is a routing function — operators earn standing on uptime, correctness, and latency, and demand flows toward the ones that earn it. For Warden Buffett and every trading agent after it, reputation is the difference between a performance claim and a performance record.
3. Quant Kit
Trading is the use case that pulls hardest on agent infrastructure, so we built the tooling for it properly.
Quant Kit is the strategy layer: author a strategy, backtest it against historical data, simulate it under adverse conditions, then deploy it as a live onchain agent with the same code path throughout. Strategies become composable objects rather than private scripts — they can be inspected, forked, combined, and carry their own verifiable performance history through the reputation layer.
The point is to collapse the distance between an idea and a deployed, accountable agent from months to days, without asking a quant to become a smart contract engineer.
Who it's for: quant developers, trading teams, DAO treasuries, and anyone who wants to run a strategy without building an execution stack from scratch.
How it connects: Warden Buffett was built with it. That is the proof it works. BerkshireDAO is one strategy family with one personality — Quant Kit is how thousands more come to exist, each one a new agent with a reason to live on Warden.
4. Guardrails
Autonomy without constraints is not a feature, it is a liability. The reason serious capital has stayed away from autonomous agents is not skepticism about AI. It is the absence of enforceable limits.
Warden 2.0 makes guardrails a protocol primitive, enforced by smart contracts rather than requested in a prompt. Spend limits and rate limits. Allowlists for counterparties, venues, and assets. Drawdown caps that halt an agent before it compounds a mistake. Quality thresholds that hold execution unless an inference clears verification and confidence bars. Escalation paths that hand a decision back to a human when it exceeds the agent's mandate.
A prompt is a suggestion. A contract-enforced constraint is a guarantee. That distinction is the whole reason agents belong onchain.
Who it's for: every user who connects a wallet to an agent, and every institution whose risk function has to sign off before it does.
How it connects: SPEx already tells you whether the model you paid for is the model that ran, and whether the output was confident enough to use. Guardrails are what turn that signal into an action the chain will or will not permit.
5. Verifiable inference for enterprise
This is where the technology meets a market that pays.
Enterprises cannot put unverifiable model output into a regulated decision. They need to know which model ran, that it was not silently swapped for something cheaper, that the output met a quality bar, and that all of it produces an audit trail a regulator or an internal control function will accept. Today they solve this by trusting a single vendor and hoping. That is not a control. It is a counterparty.
SPEx was built for exactly this: sampling-based verification that checks any inference at a fraction of the cost and up to 1,000x the speed of ZKML, OPML, or TEE approaches. Warden 2.0 packages it for enterprise consumption — verification as a service on top of whichever inference capacity you use, with attestations and an audit trail that persist onchain.
Who it's for: financial services, insurance, healthcare, and any organization that has to explain an AI-driven decision after the fact.
How it connects: Halo supplies permissionless, competitively priced capacity from thousands of independent operators. Enterprises cannot touch that without a verification and audit layer on top. Put them together and you get something no centralized provider offers: open-market inference pricing with cryptographic assurance. That is a commercial wedge, and it settles through the protocol.
The dots, connected
Read the five together and the design becomes obvious. Identity makes an agent real. Reputation makes it trustworthy. Quant Kit makes it capable. Guardrails make it safe. Verifiable inference makes it acceptable to the institutions with the deepest pockets. Those are the rails an Agent Economy runs on, and there is no version of it that works without them.
Halo and Warden Buffett are not side quests. They are the demand generators that make this layer necessary, and the proving grounds that make it credible. Every operator Halo onboards, every user Warden Buffett reaches, every strategy someone deploys — all of it consumes the same protocol services, and all of it makes those services more valuable.
Growth happens at the edges. Value comes home to the protocol. That is the entire thesis of Warden 2.0.
Full details, timelines, and technical specifications land with the reveal.
What this means for WARD
WARD is the protocol asset underneath all of it.
Warden 2.0 is designed so that ecosystem growth accrues to the network rather than leaking out of it. Identity, reputation, guardrails, and verification are protocol-level services — they route through Warden, they settle on Warden, and they are paid for and incentivized in WARD. The more agents that exist, the more inference that gets served, the more strategies that go live, the more of that flows through the protocol layer. That is the design, and it is why the infrastructure release matters more than any single product launch.
There is a second dimension, and it is the one we are most deliberate about: WARD holders get exposure to the entire ecosystem itself.
The projects launching from Warden are not unrelated companies borrowing the name. They are built with Warden infrastructure and they consume Warden services, and the community that carried the protocol this far will have exposure to what they become. That starts with Halo and Warden Buffett — WARD holders will form part of both launches — and it extends to the ecosystem projects that follow. More are in the pipeline. Each one is another channel, another source of activity, and another place where WARD holders have a stake in the outcome.
Specifics for each launch come in their own announcements over the coming weeks. What we are committing to here is the shape: an ecosystem that grows outward, a protocol that captures the value, and a community with exposure to both.
Time to be loud
We said it months ago: more engineering, less marketing. We meant it, and we shipped. A live inference mesh. A trading agent two years in the making. A protocol that does what the litepaper says. And a full infrastructure layer, built quietly over years, about to be revealed.
Now the second half starts. With real products live, marketing ramps up. More visibility. Big partners. New chains. Campaigns built around things you can actually use, not things we promise to build. Expect launch announcements, partner reveals, the Warden 2.0 unveiling, and a much louder Warden across every channel — starting with the Halo launch partner news coming very soon.
Two live products this summer. Warden 2.0 next. A protocol underneath all of it that gets stronger with every unit of activity the ecosystem creates.
Engineering earned the right to make noise. Now we make it. Let's build.
Try the Halo alpha: runhalo.xyz