AntSeed

What it does

AntSeed is a peer-to-peer marketplace for AI inferenceInferenceRunning a trained AI model to produce an answer. Inference is what happens when you type a prompt into ChatGPT and get a response. The model takes your input, computes a best guess, and returns it.Like asking an expert for their opinion. The training was the decades they spent becoming an expert. The inference is the 30 seconds it takes them to answer your specific question.Read more →. Buyers discover providers via BitTorrent DHT, route requests through encrypted WebRTC peer connections, and settle payments in USDC through on-chain payment channels on Base or Arbitrum. There is no central APIAPIApplication Programming Interface. A structured way for one piece of software to talk to another. In DeAI, APIs let applications request inference from a model without running the model themselves.Like a waiter in a restaurant. You don't walk into the kitchen and cook your own meal. You tell the waiter what you want, they tell the kitchen, the kitchen cooks it, and the waiter brings it back. The API is the waiter.Read more → gateway. There is no aggregator taking a cut. The codebase is GPL-3.0 on GitHub at antseed/antseed, primarily TypeScript, with v0.1.86 shipped on 14 May 2026.

The protocol distinguishes three provider types. Raw inference providers serve a model directly (Anthropic API, Claude Code, Claude OAuth, OpenAI-compatible APIs from OpenAI, Together or OpenRouter, or local LLMs via Ollama or llama.cpp), setting their own per-token price. Routing service providers build specialised routing logic and earn fees per request routed without running a model themselves. AI agent providers wrap expertise behind a private system prompt, RAG sources, and toolchain.

The team is two people and seven agents per the 12 May 2026 protocol-update post. The publicly named co-founder is Shahaf Antwarg, who previously co-founded and ran CTO at NFTrade, a top-five NFT marketplace from the 2021 cycle. Personal site at antwarg.com. The second human co-founder’s identity has not been published in launch materials. The contract was deployed and started emitting in approximately April 2026; the public commercial launch with 20 providers landed on 15 May 2026 per the Benzinga press release. Strategic partnerships are confirmed reciprocally with Secret Network (for confidential-compute routing through TEEs) and Venice.ai (Erik Voorhees, providing the DIEM inference pool at diem.antseed.com).

Value proposition

The architecture is the part of AntSeed that holds up under examination. Every other “decentralised AI inference” project on the shelf has a gateway: a foundation-operated routing service that consumers query, with permissionless miners attached as backend. Heurist is gateway-routed. So is Akash’s AI service layer. AntSeed isn’t. The buyer client (AntStation desktop or CLI) finds providers directly via DHT, opens an encrypted WebRTC channel, and pays via USDC payment channel. The protocol’s role at the inference layer is the discovery and settlement contract, not the routing service. That’s a structural difference that shows up in the Freedom Score.

The fair launch is the other distinctive feature. AntSeed has no team allocation, no VC round, no airdrop list, no pre-mine. Every ANTS token is emitted on a weekly epoch schedule to network users (50% provider pool, 20% buyer pool), contributors (15%, vested), and a reserve fund (15%). The 12 May 2026 protocol-update post states the position directly: “No pre-mine. No unfair head start. No company sitting behind the protocol extracting value.” That’s a much harder claim than most fair-launch DeAI projects can defend, and the BaseScan contract verification supports it: the contract has no admin mint controls beyond the designated emissions contract, no privileged allocation, no unlock cliff.

The counter-narrative is the token-side gating. ANTS transfers are disabled at the contract level. The verified BaseScan source shows a TransfersNotEnabled error path and admin-gated enableTransfers() and setTransferWhitelist() functions. The token is fully on-chain and emitting on schedule but it cannot move. It isn’t listed on CoinGecko, CoinMarketCap, or any DEX. Fifty-six holders on Base, all reward earners. That’s deliberate, consistent with the fair-launch protection, and tough on the Returns Score.

The architecture

The protocol has two layers. Layer 1 is the discovery and transport layer, off-chain and peer-to-peer. Providers register themselves on a BitTorrent DHT, advertising their service catalogue (model, price, capabilities). Buyers query the DHT directly, pick a provider, and open a WebRTC connection that’s encrypted end-to-end between the two peers. The Secret Network integration adds the option to route through TEE-backed confidential-compute nodes for sensitive queries; the default mode is provider-visible plaintext, which is the inherent constraint of inference without homomorphic encryption.

Layer 2 is the settlement layer, on Base for ANTS and USDC payments and on Arbitrum as a secondary USDC settlement venue. Buyers deposit USDC into the AntseedDeposits contract on Base. For new sessions a hard $1 USDC authorisation cap applies. During a session, the buyer signs cumulative SpendingAuth vouchers off-chain and providers can submit them to settle deltas on-chain. On channel close, unspent funds return to the buyer. New buyer credit limits start at $10 and grow by $5 per unique seller and $0.50 per day, capping at $50. On-chain reputation stats track channel count, ghost count (channels closed without settlement), total USDC volume, and last settlement timestamp per provider.

The live network state is public at network.antseed.com/stats and worth checking. At research date, 17 distinct provider peers show real activity. The largest single provider by request volume, “Dark Signal”, has 289,987 requests, 9.23 billion cumulative input tokens, 5,729 on-chain settlements and 1,463 channels. Another provider, surplusintelligence.ai, has 22,172 requests across 35 unique buyers. A medical-specialist provider, Auralis AI, has 4,990 requests. The settlement counts are independently verifiable on Base block explorer; settlement blocks span the 46023411 to 46062969 range at research date. That’s real on-chain activity. The 12 May 2026 self-report of “close to 1.5B tokens per day” is consistent with these snapshot numbers in aggregate; whether the daily rate is sustained requires window verification.

Tokenomics

ANTS is an ERC-20ERCEthereum Request for Comment. A numbered standard that defines how a specific type of smart contract should behave. Common examples are ERC-20 (fungible tokens), ERC-721 (NFTs), and ERC-4626 (tokenised vaults).Like the standard specification for how a USB plug fits into a USB socket. Any manufacturer can build a USB device, but if they follow the spec, their device works with every other USB-compatible product on the market.Read more → on Base at 0xa87EE81b2C0Bc659307ca2D9ffdC38514DD85263, verified on BaseScan with exact match. Max supply 1,040,000,000. As of 16 May 2026, total minted is approximately 21.72 million ANTS, which is around 2.09% of the cap. The contract is fully on-chain and emitting on the published schedule; the TransfersNotEnabled flag means none of those 21.72M can move between wallets yet.

Each weekly epoch distributes 5 million ANTS across four buckets:

• Provider Pool (seller emissions): 50.00%
• Buyers: 20.00%
• Ecosystem Reserve: 15.00%
• Contributors: 15.00% (vested)

Every 104 epochs (approximately two years) the epoch budget halves. Six halvings total reduce emissions toward near-zero over a 12-year curve. Beyond the epoch emissions there is no inflation. Beyond the four buckets above there is no allocation. The 12 May 2026 protocol-update post states this directly: no pre-mine, no team allocation, no VC, no company. The closest analogue to an insider allocation is the Contributors bucket at 15% per epoch, which is vested and tied to ongoing contribution rather than pre-allocated to a fixed roster.

Two protocol-level fees flow to the Protocol Reserve. A 4% network fee on every settlement, and a 10% fee on the Venice DIEM pool integration. Per the 12 May 2026 post: “If AntSeed succeeds, the value created by the network should strengthen the network itself. Fees should support long-term sustainability, trust, utility, and alignment, not centralised extraction.” The post commits to a “broader token economic architecture” being designed over coming months, “with direct impact on ANTS”. The route from Protocol Reserve to ANTS holder value is not yet defined. There is no burn, no buyback, no fee-share to holders yet operating.

The 12 May 2026 post also introduced two material governance changes. Seller (provider) ANTS emissions are no longer freely claimable: they are routed into a dedicated Provider Pool and locked, pending validation systems “for service validation, audits, attestations, and proofs that providers are serving real inference to real buyers”. Estimated timeline two to three months. Locked emissions may also be subject to slashing for providers found to fake usage. The architecture intent is that real providers benefit from emissions and incentive farmers do not. Buyer emissions remain claimable after epoch finalisation, subject to caps and anti-abuse filters.

No public smart-contract audit has been located. The BaseScan source verification is compilation correctness, not security review. The AntseedDeposits payment-channel logic is custom code on-chain. The protocol carries audit-gap risk that’s standard for an early-stage fair-launch network.

How to participate

Buy inference. Install AntStation (macOS) or use the AntSeed CLI client. Deposit USDC into the AntseedDeposits contract on Base or Arbitrum. Discover providers via DHT. Open a channel, pay per-token in USDC up to the session cap, and earn buyer ANTS emissions at epoch finalisation. New buyers start at a $10 channel credit limit and grow toward $50 through reputation. The buyer ANTS emissions accumulate but cannot be transferred yet.

Run a provider. Three modes. Serve a model you already host (Anthropic API key, Claude Code keychain, OpenAI-compatible endpoint, or a local Ollama or llama.cpp instance) via antseed seller setup. Run a routing service that earns fees on requests routed without operating a model. Or wrap your own expertise as an AI Agent provider with private system prompt and RAG. USDC settles per channel close; provider ANTS emissions are currently locked in the Provider Pool pending validation. Anti-abuse rule: reselling your personal Claude Pro or OpenAI subscription is explicitly prohibited; providers must add real value through skills, workflows, TEEs, or fine-tunes.

Build. GPL-3.0 codebase at github.com/antseed. v0.1.86 released 14 May 2026, TypeScript-heavy. Contributor allocation (15% of each epoch emission, vested) is the de facto grants pathway; no formal application portal documented.

Governance. Off-chain only at present. The 12 May 2026 protocol changes were made unilaterally by the two-person team. The forthcoming economic architecture is signalled to include token-aligned coordination but its design is not public yet.

Honest assessment

What works

The architecture matches the marketing. BitTorrent DHT for discovery means there is no central registry to subpoena, lean on, or rate-limit. WebRTC P2P transport means there is no central routing service to operate. USDC settlement on Base and Arbitrum is operational and verifiable on block explorers. The GPL-3.0 licence is liberal enough that any party can fork the protocol and run a parallel network. The live stats endpoint at network.antseed.com/stats is public, granular, and matches the on-chain settlement record. That’s a different posture from the “trust us, we’re decentralised” pitch most projects ship with.

The fair launch is real and defensible. No pre-mine, no team allocation, no VC allocation, no airdrop list. The 12 May 2026 protocol-update post commits this in writing. The on-chain contract verification supports it. Compared to projects whose tokenomics tilt 20-30% toward insiders before any community emission, AntSeed’s 70% community emission allocation is structurally different.

The founder track record is verifiable. Shahaf Antwarg co-founded and ran CTO at NFTrade, which was a top-five NFT marketplace in 2021. He has a public personal site, gave a conference talk in Tel Aviv on Web3 for developers in 2022, and is on record in the Benzinga launch press release. The second co-founder is not publicly identified, which is a real disclosure gap but doesn’t rise to the anonymous-team disqualifier.

Two strategic partners are reciprocally confirmed. Secret Network announced the AntSeed partnership on their X account and confirmed integration of confidential-compute routing. Erik Voorhees of Venice.ai is quoted in the Benzinga release saying “DIEM was designed to make AI access something users can truly own, not rent”, supporting the diem.antseed.com inference pool. Both are first-party confirmations from the partner side.

What doesn’t work

Token transfers are disabled. That’s the headline weakness on the Returns side. The contract has admin-gated enableTransfers() and setTransferWhitelist() functions and a TransfersNotEnabled error path. Buyer emissions are claimable in the sense that they accumulate to a wallet, but the wallet cannot then transfer them out. There is no public schedule for transfer enablement. The 12 May post implies enablement is tied to the broader economic architecture finalisation, which has no committed date.

The team is two human co-founders. The 12 May post acknowledges this directly. Lean is one framing. Bus factor is the other. A two-person team operating a 5-week-old permissionless inference market with locked emissions, planned slashing, and a 4% protocol fee accumulator has limited operational redundancy. Security response, incident communication, and treasury stewardship all live on a thin bench.

No public audit. The custom AntseedDeposits payment-channel logic is the part of the on-chain surface that handles real USDC. The contract is verified-compiled. It is not third-party security-reviewed. An exploit at this stage would be material and recovery resources are limited.

Validation-system execution risk. The locked Provider Pool emissions are the core of the seller-side incentive. The 12 May post commits the team to “audits, attestations, and proofs that providers are serving real inference” within “two to three months”. That’s a meaningful engineering target with knock-on implications: slashing parameters need to be specified, an appeals process needs to exist, and the validator code needs independent review before locked emissions can become claimable. Slippage on that timeline weakens the provider incentive and the network throughput together.

The Subscription Pool design is undefined. The 12 May post signals a “broader token economic architecture” with “direct impact on ANTS” but does not specify the mechanism. The protocol could choose to route Protocol Reserve fees to a buy-and-burn, to staked ANTS holders, to a treasury that funds development, or to some combination. The framing is open; that’s honest but it’s also a real Returns Score uncertainty.

The risk

Two specific risks compound. First, the locked-emissions / disabled-transfers combination creates a window where both providers and buyers earn ANTS but neither can liquidate or even transfer them. If the validation system slips and transfer enablement slips with it, the work-incentive coherence weakens. Second, USDC settlement (not ANTS settlement) is the real revenue layer. The protocol could be commercially successful at the USDC fee accumulation level without that success ever translating to ANTS holder value, depending on what the Protocol Reserve mechanism does. That’s the same structural risk Heurist faces with x402 USDC settlement on its agent layer, and it applies here for the same reason.

There is also a small concentration vector at the L1 settlement chain. ANTS lives on Base, whose sequencer is operated by Coinbase. Arbitrum is the secondary settlement venue but the canonical ANTS contract is Base-only. Censorship resistance is strong at the protocol’s P2P layer (DHT + WebRTC) but the token contract inherits Base’s sequencer single-point-of-failure.

My position

Fact: AntSeed runs a GPL-3.0 P2P inference marketplace with BitTorrent DHT discovery, WebRTC transport, USDC settlement on Base and Arbitrum, fair-launch ANTS emissions to network users on weekly epochs, 17 active providers with verifiable on-chain settlements, and reciprocal partnerships with Secret Network and Venice.

Take: This is the architecturally honest end of the DeAI inference shelf, and a useful foil to projects that talk decentralisation and ship gateways. The Returns Score is F because the team intentionally turned off the transfer market during the fair-launch protection window, which is consistent with the no-pre-mine framing but produces zero tradable liquidity at research date. The interesting watch is whether the team can hit their validation-framework timeline, define a credible Protocol Reserve to ANTS holder mechanism, and enable transfers under a schedule that holders can plan against. None of those are speculation. All three are commitments the team has already made.

I don’t hold ANTS. The token isn’t transferable yet, so this is more of an interest-disclosure point than a position statement. The protocol itself is worth running as a buyer for the credit-limit accumulation and the buyer emission. If the team ships their committed economic architecture under a credible schedule, the Returns Score has real room to move and the Freedom Score is already in the C band.

Freedom Score: 66/100

AntSeed scores 66/100 (C grade). Full methodology at Freedom Score Methodology.

Infrastructure decentralisation (14/20): BitTorrent DHT for provider discovery, WebRTC for encrypted P2P transport, GPL-3.0 open-source codebase. Permissionless provider registration. 17 distinct active providers visible at the public stats endpoint with verifiable on-chain settlement transactions on Base block explorer. Multi-chain USDC settlement on Base and Arbitrum. Secret Network integration enables TEE-backed confidential-compute routing for sensitive queries. No central API gateway. Held below the top band because the network is early-stage at research date with 17 providers, which is small for “fully permissionless, geographically distributed” classification.

Governance decentralisation (4/20): Two-person team retains unilateral admin authority. The 12 May 2026 protocol update was a unilateral decision: lock seller emissions in a Provider Pool, introduce a 4% network fee, route the 10% DIEM pool fee to Protocol Reserve, signal future slashing. The post frames these as protocol stewardship, which is reasonable for a 5-week-old network, but mechanically governance is team-controlled. No DAODAODecentralised Autonomous Organisation. A way to coordinate decisions and manage a treasury using token-weighted voting instead of a traditional company structure. Token holders propose and vote on changes directly.Like a shareholder-run company where every shareholder can vote on every decision, the votes are public, and the company can't do anything the shareholders don't approve. The coordination is messier than a normal company but nobody has unilateral control.Read more →, no token-voting, no public proposal register. Future “broader token economic architecture” signalled but not designed.

Token distribution fairness (13/15): Among the cleanest fair launches on the DeAI shelf. Per the 12 May 2026 protocol-update post: no pre-mine, no team allocation, no VC, no company. Per-epoch allocation: 50% Provider Pool, 20% Buyers, 15% Ecosystem Reserve, 15% Contributors. 70% of each epoch’s 5M ANTS flows directly to network users (buyers and providers); 15% is the closest analogue to an insider bucket but is vested and tied to ongoing contribution. Top of the “fair launch or fully community-distributed” band; not 15 because Contributors is still mechanically a team-discretionary allocation.

Censorship resistance (12/15): Strong by design. WebRTC P2P transport with no central routing. BitTorrent DHT discovery with no central registry. GPL-3.0 open source. No KYC required. Anonymous-by-design at Layer 1. Multi-chain settlement (Base + Arbitrum) limits single-chain sequencer risk. Secret Network integration adds TEE-backed routing. ANTS transfer restriction is a separate concern (it’s about token tradability, not inference access). “Strong censorship resistance with few realistic attack vectors” band.

Data sovereignty (11/15): Better than gateway-routed alternatives. Encrypted P2P connections mean no central server sees queries. Secret Network integration enables TEE-backed confidential queries. The AI Agent provider model is explicit that system prompt, RAG sources, and toolchain stay private at the provider side. Default mode still has the provider seeing the prompt in plaintext, which is inherent to inference unless TEE is used. Top of “users control their data with some platform dependencies” moving into “strong self-custody”.

Open source and transparency (12/15): GPL-3.0 codebase on GitHub at antseed/antseed, v0.1.86 (14 May 2026), TypeScript-dominant. Token contract verified on BaseScan with exact match. Public live stats endpoint at network.antseed.com/stats. Lightpaper published. The 12 May 2026 protocol-update post is detailed and open. Held below 13-15 band because no independent smart-contract audit located and no governance log or treasury transparency mechanism documented.

Path to improvement

Three changes would materially raise AntSeed’s Freedom Score:

1. Ship the validation system on schedule. The locked Provider Pool emissions and planned slashing are designed but not deployed. Delivering audits, attestations, and proofs-of-service within the committed two-to-three-month window converts the seller-side narrative from “intent” to “operating”.
2. Introduce token-weighted governance. The 12 May post signals this as future work. Even a basic Snapshot space with proposals against the Protocol Reserve mechanism would move governance out of the foundation-style “we have decided” register.
3. Commission an independent audit. Halborn, Trail of Bits, OpenZeppelin, or equivalent on the ANTS contract and the AntseedDeposits payment-channel logic. The custom payment-channel code is the highest-impact surface for an exploit.

Returns Score: 32/100

ANTS scores 32/100 (F grade). Full methodology at Returns Score Methodology.

Token utility (7/20): Designed utility is clear (buyer and provider reward emissions, future Subscription Pools, reputation weighting, access tiers, coordination per the 12 May 2026 post) but unfired. ANTS transfers are currently disabled at the contract level. USDC, not ANTS, settles inference payments. Provider Pool emissions are locked pending validation. The token is real but its transactional utility today is zero outside reward accrual.

Value accrual (6/20): Two real fee accumulators flow to Protocol Reserve: 4% network fee on settlements and 10% DIEM pool fee on the Venice integration. The 12 May 2026 post explicitly states “direct impact on ANTS” but the route from Protocol Reserve to holder value is undefined. No burnBurnPermanently removing tokens from circulation by sending them to an address that no one controls. Burns reduce total supply, which (all else equal) makes each remaining token worth more of the network's value.Like a company buying back its own shares and shredding them. The company's total value stays the same, but each remaining share now represents a slightly bigger slice of that value.Read more →, no buyback, no fee-share yet. Locked Provider Pool emissions may be subject to slashing (reducing supply). Better than zero designed accrual; below proven accrual because the route is still being designed.

Supply dynamics (9/20): Fixed 1.04B cap. Disinflationary schedule with 5M ANTS per epoch initially, halving every 104 epochs across six halvings for a ~12-year curve. Approximately 2.09% minted at 5-week mainnet age. No insider, team, or VC vesting overhang because there is no insider, team, or VC allocation. Supply pressure ahead is real (98% still to be emitted) but it flows to network users rather than to early investors. Structurally healthier than projects with insider vestingVestingA schedule that locks up tokens allocated to insiders, investors, and team members, releasing them gradually over months or years. Vesting prevents insiders from dumping on public buyers immediately after launch.Like a new employee's stock options at a startup. You don't get all the shares on day one. They unlock over four years so you stick around and do the work rather than cashing out and leaving.Read more → overhang against thin floats.

Revenue sustainability (9/25): Real on-chain settlements happening at early-stage mainnet age. Live stats endpoint at network.antseed.com/stats publishes per-provider request, settlement, channel, and unique-buyer counts. Aggregated across top providers, cumulative input tokens are in the 10B+ range with thousands of settlements. 4% network fee accumulates to Protocol Reserve on every settlement. Self-reported headline is “close to 1.5B tokens per day”. Held below mid-band because the network is early-stage, the route from Protocol Reserve to ANTS holders is undefined, and no third-party verifier yet tracks AntSeed.

LiquidityLiquidityHow easily a token can be bought or sold without moving the price. High liquidity means you can enter or exit large positions quickly at the quoted price. Low liquidity means even small trades can swing the market.Like the difference between selling a house and selling a share of Apple stock. The house might be worth more on paper, but finding a buyer at that price takes weeks. The Apple share converts to cash in one click.Read more → and access (1/15): Transfers are disabled at the contract level. Verified BaseScan source shows TransfersNotEnabled error and admin-gated enableTransfers() and setTransferWhitelist() functions. Not listed on any CEX or DEX. Not tracked by CoinGecko or CoinMarketCap. 56 total holders on Base. The only acquisition route is participation in the network. This is a deliberate fair-launch protection but it produces functionally zero tradable liquidity at research date.

Path to improvement

Three changes would materially raise ANTS’s Returns Score:

1. Enable transfers under a credible schedule. Even a phased enablement (whitelist by reputation tier, then unrestricted) gives holders a planning horizon. The current TransfersNotEnabled status without a published date is the single largest Returns drag.
2. Define and ship the Protocol Reserve to ANTS route. Buy-and-burn, fee-share to staked ANTS, or treasury operations with on-chain proof. The 12 May 2026 post commits to “direct impact on ANTS”; converting that commitment to a mechanism turns Value Accrual from designed to operating.
3. Publish sustained-window network metrics. The snapshot stats are real and verifiable, but a public dashboard with rolling 7-day and 30-day settlement volume, fee accumulation, and active-provider count would let third parties price the network on revenue rather than on launch press.

Score change log