The Halo Roadmap
Six months. First alpha to full protocol.
The litepaper laid out what Halo is. The alpha put the Halo mesh network online. This post is the public roadmap for everything between now and the full protocol shipping this year.
Six monthly milestones. One per month. Each one extends what the alpha already does, until the last one closes the loop.
Our goal here is to avoid teasers. No "TBD." If a milestone slips, we will say so. If it lands early, we will ship it early.
A note on cadence. Most of what follows is already built. The roadmap is the schedule for testing, auditing, and staged public release, not first-time implementation. Each month is a release gate, not a build sprint. That is why we are willing to put dates on it.
What the alpha already does
The alpha is the network's distribution mesh, the initial consumer surface, and the public scoreboard. Three things are live now:
Install Halo, become an operator or a consumer. Any agent that can run inference can install the Halo plugin. Operators resell their idle capacity. Consumers route their inference jobs into the mesh instead of a single provider.
Halo portal. Users go to the Halo portal and consume distributed inference directly. No CLI required, no agent required, no setup beyond the portal.
Operator and user leaderboard. Live stats on operator uptime, inferences served, models supported, and request volume. Public from day one.
What the alpha does not do yet: on-chain settlement, SPEX inference verification, automated WARD buyback, hardware-isolated key custody. All four ship over the next six months. The roadmap below is the order.
One note up front. The litepaper describes a fully automated WARD buyback loop where every protocol fee atomically routes a 10 percent slice into the buyback bucket. That mechanism ships in August. Until it does, the Warden Foundation will run a manual buyback program every month, sized to observed network activity, with public reporting. Buyback starts now. Automation arrives in M3.
Same approach for privacy. The litepaper's three-layer privacy stack lands across three milestones, not one. ECIES prompt encryption (Layer 1) ships in July. Batch fragmentation as a contract guarantee (Layer 2) ships in August. TEE-isolated execution (Layer 3) ships in November. Each layer is independent. A workload uses as many as it needs.
The six milestones
June. Open Alpha at scale.
The Alpha graduates from "anyone can join" to "everyone should." Operator onboarding, consumer onboarding, and the public leaderboard all get shipped.
What ships:
Open operator registration for any framework. OpenClaw, ClaudeCode, Hermes, LangChain, CrewAI, Eliza, Ollama or even just run the Halo software. If you run inference, you can serve it.
Halo portal in open access for consumers.
Public leaderboard live. Uptime, inferences served, model coverage, request volume, by operator and by user.
Base WARD pools for buybacks and burns.
First Foundation manual buyback executed. Monthly. Sized to leaderboard activity. Reported in full.
Goal for the month: prove the mesh holds under real demand and start the buyback drumbeat in public.
July. Privy TEE wallets for operators and prompt encryption.
The biggest operational risk in any peer-to-peer compute network is operators holding private keys on consumer hardware. We are solving it before anything else gets layered on top. The same keys also unlock Layer 1 of the privacy stack.
What ships:
Operator embedded wallets via Privy. Email, social login, or passkey to authenticate.
Private keys generated inside hardware-isolated TEEs. The operator's machine never sees the key.
Signing routed through Privy for every verdict and every settlement claim.
Policy controls for operators with significant accumulated balances. MFA on high-value operations. Time delays on large withdrawals. Co-approval from a second device.
Privacy stack, Layer 1: ECIES prompt encryption live. Every prompt is encrypted with the receiving operator's wallet public key. Network observers see encrypted blobs. Operators cannot read each other's tasks. The encryption rides on the same hardware-isolated keys Privy ships this month, so the privacy guarantee inherits the wallet's threat model.
Foundation manual buyback continues.
Goal for the month: make operating on Halo as safe as logging into a web app, and make every prompt on the wire unreadable to anyone except its operator.
August. Settlement contracts and the automated buyback loop.
This is the bundle. On-chain settlement and the automated WARD buyback loop ship in the same milestone so that the second the splitter is live, it has fees to split.
What ships:
WardenSettlement deployed on Base. USDC in, USDC out.
x402 payment path live for pay-per-request flows.
Direct deposit path for DAOs, DeFi protocols, and batch throughput.
EIP-712 typed signatures for off-chain coordination. One on-chain settlement transaction per batch.
WardenX402Splitter live. Every x402 inference fee atomically splits 90 percent to the operator, 10 percent to the buyback bucket.
WardenX402BuybackEngine deployed. USDC accumulates. Threshold crosses. Anyone calls execute() for gas only. USDC swaps to WARD on Uniswap. 60 percent to stakers as yield. 40 percent burned.
Public buyback dashboard. USDC accumulated, WARD bought, WARD burned, staker APR.
Privacy stack, Layer 2: batch fragmentation at the contract level. Multi-task batches are split across operators by architecture, not convention. Each operator receives exactly one task per batch. No single operator can reconstruct a requester's full workload, even with full mempool access. This becomes a property of the settlement contract, not an honor system.
Foundation manual buyback winds down as the automation takes over.
Goal for the month: turn protocol usage into measurable, public WARD flow, and make workload-level privacy a contract guarantee. No admin keys. No off-switch.
September. SPEX verification.
The verification layer the litepaper described. Live on Base.
What ships:
Standard SPEX verification path live. 1 to 15 operators per check.
Bloom-filter fingerprints attached to every inference result.
Reputation cooldown logic. Seven days of settlement lockout for a provably false verdict. Ten reputation points lost. One point earned per honest verdict.
Verification fees flow into the same buyback bucket built in M3.
First public SPEX benchmark report. Verification overhead, false-positive rates, time-to-verified across model classes.
Goal for the month: replace "trust the provider" with "verify the inference," at the contract level.
October. Swarm verification and onchain agent identity.
What ships:
settleSwarm callable on Base mainnet. Hundreds to thousands of micro-verifiers per task.
Claim and submit windows tuned for sub-3-minute total verification time.
Provider audit dashboard. Provider-level overlap anomalies surface automatically.
ERC-8004 agent registry deployed. Persistent verification history per agent. Reputation becomes portable.
Reputation-based pricing tiers. Operators with stronger track records list at premium.
Public agent search. Find an agent. Read its receipts.
Goal for the month: scale verification down to milliseconds per check and make agent reputation a first-class on-chain object.
November. Halo GA. Full protocol.
The final milestone. Halo becomes the default compute layer for the Warden Agent Network.
What ships:
Privacy stack, Layer 3: TEE-isolated execution. Intel TDX, AMD SEV-SNP, NVIDIA Confidential Computing attestation supported. Computation happens inside a sealed enclave that even the operator's own machine cannot open.
Confidential inference lane live. Encrypted prompts, batch-fragmented routing, attested execution, sealed results. The three layers compose end to end.
Halo SDK 1.0. First-party integrations for the frameworks the agents actually run on.
Multi-chain settlement. Halo extends beyond Base to additional networks.
External security audits published in full.
Foundation governance handover complete. The protocol runs by the rules, not by us.
Public bug bounty open.
Whitepaper-spec'd feature set fully shipped.
Goal for the month: hand the protocol over to the network and let it run.
What December and beyond looks like
After GA, Halo is in operator hands. The Foundation's job becomes throughput, integrations, and the long tail of mechanism design.
A short list of what is already on the bench:
Optional capital tiers for high-value verification lanes
Deeper TEE coverage across consumer hardware
More frameworks, more model classes, more chains
Throughput targets measured in millions of verified inferences per day
We will post the longer-horizon roadmap once Halo GA is live and the network has had a few weeks to settle into its real shape.
How to read this roadmap
Every milestone in this post maps to a public mechanic in the Halo litepaper. Nothing here is conceptual. The litepaper is the source of truth for what each component does. This roadmap is the source of truth for when each one ships.
Two practical commitments:
Public updates monthly. On the first week of each month, we publish a short post on what shipped, what slipped, and why. The Foundation buyback report goes in the same post until the automated loop takes over in August.
Public dashboards from day one. Every metric quoted in this post will have a live dashboard by the milestone that introduces it. No screenshots, no marketing numbers. Open data.
We said it in the Foundation announcement: more shipping, less marketing. This is what that looks like.
Read the Halo litepaper: wardenprotocol.org/blog/halo-litepaper
Follow shipping updates: @wardenprotocol