Roadmap: vault & token

On-chain staking and slashing, payment channels, the compute vault, and additional settlement rails.

Everything on this page is forward-looking. None of it is live, none of it is promised on a date, and nothing here should be read as an offer of anything. The V1 core was deliberately minimal; the design doc's graduation ladder (section 12.9) lists what comes next, and every rung is additive: nothing in the shipped core has to be rewritten to add them. For what is actually running today, see the implementation status matrix.

The ladder is real: one rung already landed

Session continuity for agentic loops was a V2 item, the largest by economic impact, and it shipped: cache-aware quotes, effective-cost auctions every turn, locked session tariffs, one payment per loop. It is the existence proof that the rest of the list is a build plan, not a wish list.

What comes next

• Payment channels. A channel tier kills the refund problem (charged equals settled by construction) and is the big per-request privacy upgrade, since individual requests stop touching the chain at all.
• On-chain USDC staking and slashing. Today a provider's bond on a model claim is a declared figure and the challenge lane stakes points. In V2 the bond becomes escrowed USDC in an on-chain provider vault, with a withdrawal delay at least as long as the dispute window. Slashed funds split 50/50: half to affected customers as pro-rata refunds, half to an insurance pool backing dispute payouts.
• Sampled audits at scale. Re-run a random 0.5 to 2% of completed requests against a reference deployment and compare by logprob divergence, with adaptive sampling that audits new or flagged providers harder. Canaries are live today; the audit pipeline is the planned complement.
• Live RFQ mode. A real-time quote solicitation lane for long-tail requests that standing quotes do not cover well, at the cost of a 150 to 500 ms bidding window.
• Additional settlement chains. All chain work sits behind a five-operation adapter, so a new rail is one file. The Solana adapter is documented and gets enabled on demand signal, not on a schedule. HyperEVM remains the live rail.
• Permissionless provider onboarding. The V1 allowlist is a stand-in for staking. Once bonds are on-chain and audits are sampled, any provider that posts collateral can quote.
• The Hyper Compute Vault and $HYPER. Both pre-launch. The vault is a team-run maker on the book, quoting open-source models and funded by USDC deposits instead of a balance sheet; yield from its cleared fills flows to depositors, and the book always has an ask. Around it sits the fee circuit: every fill pays a fee, split 50/50 between the protocol treasury and buyback and burn of $HYPER. The vault page in the app shows today's real market volume and renders every vault-specific figure as a blank until the contract deploys.

Related reading: the points program (whose ledger re-denominates at TGE), challenges (points-staked today, vault-era bonds later), and hyper, the first-party provider.