Private payments for Akash compute

An Akash governance proposal by Terp Network, 2026-10-10. Pool figures are dated 2026-10-21 and AKT amounts use the 2026-10-19 price (pool base snapshot 2026-10-06); the on-chain spend message carries the exact uakt amount.

Summary

This proposal funds research on privacy-preserving payments for Akash's reserve marketplace: whether, and how, tenants could pay for Akash compute and inference privately with shielded ZEC, while providers are paid reliably and accounting stays auditable.

It asks for $12,800 USD worth of AKT in three tranches: 50% up front for the initial work and design, 30% once the gaps the design finds in today's protocol capabilities are closed, and 20% on public deployment of the research prototype. At 2026-10-19 prices (AKT $0.685069) that is about 18,684 AKT, 0.56% of the pool's AKT (3,359,962 AKT on 2026-10-21, after proposals #342–#345 passed).

The work starts from Terp Network's draft AEP Private bid market for providers (re-zerve), which already runs end to end in a local lab. It is separate from the liquidity proposal Pair AKT with TERP & THIOL. Each can pass or fail on its own.

The ask (what Akash governance is voting on)

A YES vote approves:

1. A community-pool spend of $12,800 USD worth of AKT (= $12,800 / P_AKT at the price used for the spend message) to the recipient, which releases it in the three tranches below. The recipient address and who performs the work are announced before the vote.
2. The research scope, milestones, open decisions and guardrails below.

Pool: 3,359,962 AKT (api.akashnet.net, height 29,158,300, 2026-10-21 03:14 UTC); base snapshot 6,119,304 AKT at height 28,949,492 (2026-10-06 21:16 UTC).

In context. Other community-pool requests share the same pool. Proposal #342 (Expanding Akash Network Market Liquidity and Depth, 1,605,394 AKT) passed; its voting ended on 2026-10-14. Overclock Labs' three Q3 2026 funding proposals, #343 Core Engineering 299,732 AKT, #344 Client Engineering 323,415 AKT and #345 AEP Support Services 869,099 AKT (1,492,246 AKT together), passed on 2026-10-21 (voting ended 01:12, 01:14 and 01:18 UTC; final tallies, yes / abstain / no / no-with-veto in AKT: #343 49,925,191 / 573,847 / 9 / 411; #344 49,926,269 / 571,582 / 5 / 0; #345 49,926,779 / 571,816 / 5 / 0), and their spends have executed. The pool now holds 3,359,962 AKT (api.akashnet.net, height 29,158,300, 2026-10-21 03:14 UTC). This ask is 0.56% of that. Together with the separate liquidity proposal (about 205,175 AKT, with its pairing AKT set at 70% of the TERP/THIOL floor value) the two come to about 223,859 AKT, 6.66% of the current pool. The AKT amounts are recalculated from the live AKT price at submission. At the 2026-10-21 price (AKT $0.724893, Osmosis SQS, 03:14 UTC) they would be lower: about 193,903 AKT for this proposal (5.77% of the current pool), 17,658 AKT for the research proposal (0.53%), 211,561 AKT together (6.30%).

Objective

Find out whether, and how, tenants could pay for Akash's reserve marketplace privately with shielded ZEC, while providers are paid reliably and accounting stays auditable. Success is a clear answer and a working, public prototype if the answer is yes. It is not a product launch.

Starting point: the re-zerve draft AEP

The draft AEP keeps the public Akash market as it is and adds a private path in the same provider-services binary. A lease on that path runs in six steps:

1. Public order. The deployer posts SDL; the deployment opens an order, as today.
2. Sealed bid. The provider sends its bid to the deployer as an encrypted envelope (ChaCha20-Poly1305), off the order book.
3. Commitments. Only hashes of the ask and the bid go on the public record (a CosmWasm contract).
4. Match, then allocate. The provider allocates only if the commitments match. On the private path, anything missing or wrong denies; there is no public fallback.
5. Compute and inference. The winner receives a derived access bearer that opens the leased endpoint on the provider's cluster.
6. Shielded pay and close. The deployer's payment sits in a shielded Zcash note held under a FROST threshold key. Close is two shielded spends: what the provider earned, and the remainder back to the deployer.

What stays on Akash: SDL, deployments and orders, the public bid engine, the hostname and inventory operators, the ES256K JWT and AKT/USDC escrow on public leases, and the stock provider image. What the private path adds: the sealed bid envelope, commitment match before allocation, the derived bearer, FROST custody, the payment proof, the CosmWasm record and the shielded two-spend close. The AEP runs in a local lab; it is not the public Akash default.

FROST custody of ZEC payment escrows

The payment for a private lease is held in escrow as a shielded Zcash note (the AEP's Ironwood note format) whose spend key belongs to a FROST threshold group (RFC 9591), not to any one party.

• Key creation. The group key is made by distributed key generation. No trusted dealer: the group must not keep dealer shares, and a dealer-generated key is a protocol error on this path.
• Threshold. t of n is configurable. The first demonstration uses 2 of 3; fewer than t shares cannot spend.
• Seats. In the lease profile the seats are the deployer, the provider and a resolver. The normal close is signed by the deployer and the provider. The resolver signs only after a close grant is recorded, and holds no viewing key on the escrow.
• Release and refund. Close is two shielded spends of the escrow note: the amount earned goes to the provider, and the remainder returns to the deployer. Earned is computed from Zcash block heights, (close height − open height) × rate; whoever closes cannot set it. The public record carries commitments and proof instances only, not hours, rate or the amount earned.
• The record is not the vault. The CosmWasm escrow contract records the grant and its events. It never moves the funds.
• Who holds the shares. Set later. The first demo places one share on an Akash lease and two in isolated Wasmer containers; the AEP invites Penumbra-style validator sidecars to hold shares.
• Open items. Key rotation and resharing when a holder changes; a challenge path if the provider or resolver will not sign; which (n, t) bounds are fixed and which are left to operators; and the signing scheme for the spend, which today uses a named ZIP-32 stand-in because ZIP-312 RedPallas FROST is not yet available.

Research questions

1. Funding path from shielded ZEC into a deployment.
2. Selective disclosure.
3. Refunds to shielded addresses.
4. Pricing between ZEC and Akash compute.
5. A design that works on whichever migration target Akash chooses.
6. Compliance posture before anything goes beyond research.

Open decisions

The research has to settle these. D2 (the design spec) records each decision and the reasons for it.

• Circuit choice. The payment proof today is a Halo2 circuit on the Pasta curves (Vesta IPA, k=17), which needs no trusted setup; its prove time is long. The options are to optimise it, or to change proof system, for example to Groth16 on BN254 as zkAPI uses, which brings a trusted setup and a second curve. A circuit for the proposed minimum-AKT-stake proof does not exist yet.
• Settlement layer. Where the value and the record live. Value: shielded Zcash notes, or another shielded pool. Record: CosmWasm on Terp in the first demo, on Akash itself, or on Akash's migration target. Payment proofs reach the record through a lab IBC v2 attestation; a live light client is still to be chosen.
• Market structure. Whether sealed bids extend the public market or split it; whether a minimum AKT stake gates a sealed bid; which private information retrieval (PIR) scheme serves sealed bids, and who may run the proxy; whether the private gate lands in provider-services upstream or stays a configured fork.
• Prior art. The design reviews what already exists for private payments and takes what fits Zcash and Akash:
  • zkAPI (Ethereum Foundation with the Open Anonymity project, live on Ethereum mainnet since October 2026): deposit once into a vault, then authorise short-lived, capped API keys with zero-knowledge proofs, so "the provider sees the requests, and the payment layer sees the spend" and neither learns the link. Worth taking: unlinking payment from use, note-bound balances, per-session credentials. Not a fit: an ETH/USDC vault and a second proof curve. Code: github.com/ethereum/zkapi.
  • Zcash shielded payments, the settlement asset this research assumes.
  • Penumbra, a shielded Cosmos chain, and a natural home for FROST shares.
  • Privacy Pools on Ethereum, for selective disclosure (research question 2).
  • ZF FROST, the Zcash Foundation's FROST implementation.

Milestones and tranches

At AKT $0.685069 (2026-10-19) the tranches are 9,342,124,661, 5,605,274,796 and 3,736,849,866 uakt (together 18,684,249,323 uakt). The AKT amounts are fixed when the spend executes; the split is by share of that amount.

Guardrails

• Scope. Research and a public prototype for private payments on Akash compute and inference. Public deployment means the prototype runs where anyone can inspect and try it. It is opt-in, changes nothing in Akash, and puts no user funds at risk beyond the bounded amounts the published design specifies. Terp Network coordinates with open providers that choose to run the provider build supporting this side-market service; no provider has to.
• Tranche gating. Tranche 1 is released up front so the work can start; D1 and D2 are its public account. Tranches 2 and 3 are released only after their milestones are accepted and each acceptance is published.
• Open output. Code and reports are public, under open-source licences.
• No protocol change. Nothing under this proposal changes Akash. Any change to the protocol or the provider needs its own AEP and Akash governance vote.
• Stop rule. Akash governance can stop further releases by vote; unreleased AKT then returns to Akash's community pool, or its successor treasury after the migration. Tranche 1, once released, is accounted for by D1 and D2.

Decision rights

Reporting

A public update at each milestone, and a final report with D5. We will submit each report to Akash's community-pool spend reporting table and post it in the discussion thread.

Risks

• The answer may be no. The research can conclude that private payments are not viable; D5 says so.
• Up-front tranche. Half the amount is released before any milestone is accepted. D1 and D2 are published so the community can judge that work before tranches 2 and 3.
• Migration. The target chain is not chosen; research question 5 and the settlement-layer decision keep the design independent of it.
• Value. The AKT amount is fixed when the spend executes; its USD value can change.

Technical details

• On-chain title: Private Payments for Akash's Reserve Marketplace (48 characters, 48 bytes; limit 10,200).
• On-chain description (summary): Appendix A below, ending with the IPFS link to this full text. Metadata: the same IPFS link.
• Message: one /cosmos.distribution.v1beta1.MsgCommunityPoolSpend. Authority akash10d07y265gmmuvt4z0w9aw880jnsr700jhe7z0f (gov module account). Recipient: announced before the vote. Amount: floor($12,800 ÷ P_AKT × 10^6) uakt at the price used for the message. At P_AKT $0.685069 (2026-10-19): 18,684,249,323 uakt.
• Deposit: 1,000 AKT (1000000000uakt). Voting period: 7 days.

Links

• Starting point: the draft re-zerve AEP, Private bid market for providers, by Terp Network (github.com/permissionlessweb/rezerve)
• Akash migration: AEP-79
• Prior art: zkAPI · Zcash protocol · Penumbra · Privacy Pools · FROST, RFC 9591
• Pool context: proposal #342 · Q3 2026 funding discussions #1541, #1542, #1543
• Live data: Akash community pool · Osmosis SQS AKT price

Appendix A: on-chain description

This appendix is the on-chain description (summary). On-chain it ends with the IPFS link to this document.

The ask. Spend $12,800 USD worth of AKT from the community pool for research on privacy-preserving payments for Akash's reserve marketplace. At 2026-10-19 prices (AKT $0.685069) that is about 18,684 AKT, 0.56% of the pool's AKT (3,359,962 AKT on 2026-10-21, after proposals #342–#345 passed). At the 2026-10-21 price ($0.724893) it would be about 17,658 AKT. The exact uakt amount is set from the AKT price used for the spend message. The recipient and the performer are announced before the vote.

Objective. Can tenants pay for Akash compute privately with shielded ZEC while providers are paid reliably and accounting stays auditable? The work starts from Terp Network's draft AEP re-zerve: sealed bids, payment escrows in shielded Zcash notes under a FROST threshold key, and a shielded close.

Tranches. 50% up front for D1 threat model and D2 design spec (circuit choice, settlement layer, prior art including zkAPI); 30% when the D3 prototype closes the capability gaps found and D4 open code and report are accepted; 20% when the public deployment of the prototype and the D5 build/iterate/stop recommendation are accepted.

Guardrails. Opt-in public prototype with no change to Akash; open output; any protocol change needs its own AEP; Akash governance can stop further releases, and unreleased AKT returns to the community pool or its successor treasury.

This is separate from the liquidity proposal; each can pass or fail on its own.

Full proposal: this document.