ZetaChain

What it does

ZetaChain launched in February 2024 as a Cosmos-SDK L1L1Layer 1. A base blockchain that runs its own consensus mechanism, executes transactions, and settles its own state. Bitcoin, Ethereum, NEAR, and Solana are all L1s. Anything built on top of an L1 is technically a Layer 2 or higher.Like the foundation of a building. Nothing else can exist on top until the foundation is solid. Different L1s make different tradeoffs for what kind of building they can support.L1 in the glossary → whose job was cross-chain plumbing: a threshold-signature wallet held native assets on Ethereum, Bitcoin, Solana and seven other chains, and smart contracts on ZetaChain could move them. On 1 June 2026 the team announced it was sunsetting that entire business. On 17 September it put Proposal 68 to token holders: halt the chain, migrate ZETA one-for-one to a Solana SPL token, and make the company’s consumer AI app the reason the token exists. It passed on 20 September with 99.44% of votes in favour.

That app is Anuma. It’s a multi-model chat product with 1,008 model ids behind one self-hosted gateway, where you can switch between Claude, GPT, Gemini, Grok and a set of open-weight models without losing context. The context is the product. Anuma keeps a per-wallet memory store on your device, encrypted with a key derived from a wallet signature, and two client-side tools let whichever model you’re using search it: a Memory Engine that embeds every message, and a Memory Vault of curated facts.

The company is Meta Protocol, Inc., a Delaware corporation registered in July 2021 with its head office in San Francisco. Its MiCAR white paper names Ankur Nandwani (ex-Coinbase, Brave and 0x), Panruo Wu (an early THORChain contributor) and Charlie McCowan as the management body. It raised $27M of equity in August 2023 from Blockchain.com, Human Capital, Vy Capital, Jane Street and others. Blockchain.com also runs the second-largest validator.

When the chain is gone, what does ZETA do inside an app that routes inference to OpenAI, and is “private memory” sovereign or a branded database? To answer that I read the open SDK, pulled the 101 JavaScript chunks the production app serves, read the contracts the app talks to, and scanned every event they ever emitted.

Value proposition

Fact: the encryption at rest matches the marketing. The @anuma/sdk package is MIT-licensed and public. Messages, conversation titles, vault entries and uploaded files are encrypted with AES-GCM under a 256-bit key derived through HKDF from a wallet signature, the key lives in a JavaScript map and dies on page reload, and backups go to your own Google Drive, Dropbox or iCloud as encrypted blobs. The production bundle at chat.anuma.ai carries the same HKDF info string and ciphertext prefix, so the design and the shipped code agree.

Take: the sovereignty stops at the disk. To make memory searchable the SDK posts each message’s text to Anuma’s /api/v1/embeddings endpoint, where a Qwen3-Embedding-8B model that Anuma hosts turns it into a vector. Every prompt, and every memory chunk recalled into it, then transits Anuma’s Bifrost gateway on the way to the model provider. Anuma’s own gateway post says the backend keeps only the model, token count and cost, and I have no reason to doubt it; I also have no way to check it, because the client and the backend are closed source.

So “Anuma servers cannot read your memory, even if compelled to”, the how-memory-works page’s words, is true of the store and false of the pipe. The store is ciphertext; the pipe carries every word that goes into the store, in the clear, past Anuma’s servers, twice. That’s the same shape we found in Walrus Memory: an ownership layer that’s yours, wrapped around a live path that isn’t. Anuma’s version is better engineered and worse disclosed.

“A key only you hold” is derived by a Privy embedded wallet signing a fixed message silently, with no prompt, whenever the app runs in your session. Privy’s split-key model means neither Privy nor Anuma can sign alone, which is a fair definition of self-custody for a consumer product. It’s still a custodian’s definition, and the help centre confirms the same embedded wallet can’t be used for staking, so the “AI identity” and the “ZETA holder” are two different wallets.

Tokenomics

ZETA has a fixed 2.1 billion cap. Allocation at launch, from the project’s own distribution page:

• Protocol treasury: 24% (2% at launch, the rest over 36 months from month 12)
• Core contributors: 22.5% (six-month cliffCliffA waiting period at the start of a token vesting schedule during which no tokens unlock at all. After the cliff ends, tokens begin releasing according to the vesting schedule.Like a probationary period at a new job. You don't get your stock options on day one. You wait 12 months to prove you'll stick around, then everything starts unlocking normally.Cliff in the glossary →, then monthly to about February 2028)
• Purchasers and advisors: 16% (same shape)
• Ecosystem growth fund: 12% (1.5% at launch, then over 42 months from month 6)
• User growth pool: 10% (airdropsAirdropDistributing tokens for free to eligible wallets, usually to reward early users, bootstrap a community, or decentralise token ownership away from a small group of insiders at launch.Like a supermarket handing out free samples to people who already shop there. The samples cost the supermarket nothing to print. The goal is to convert casual shoppers into loyal customers by giving them something tangible to talk about.Airdrop in the glossary →, XP campaigns)
• Validator incentives: 10% (block emissions over four years)
• Liquidity incentives: 5.5%

Insiders and purchasers take 38.5%; company-controlled pools take another 36%. The community share is the 25.5% paid out through stakingStakingLocking up a cryptocurrency to help secure a blockchain network, usually in exchange for rewards. The locked tokens act as a security deposit that can be taken away if the staker misbehaves.Like putting down a large rental deposit for an apartment. You get the money back if you behave, you earn interest while it's locked, and the landlord takes it if you trash the place.Staking in the glossary →, airdrops and liquidity programmes. CoinGecko and CoinMarketCap both carry 76.5% circulating, which is one figure, not two: CMC restates CoinGecko. The chain itself attests 1.709 billion ZETA on the L1, and walking the published vesting table to month 31 lands nearer 68%, so I’ve treated the aggregator number as unverified.

Emissions are 6.75 ZETA a block, which at the 3.75-second block time I measured is about 56.8 million a year, 2.71% of the cap, three-quarters of it to validators. On the 455 million bonded that’s the 9.37% APR the Anuma dashboard advertises. Proposal 68 says those rewards run until the halt height that Proposal 2 will set. What staking becomes on Solana is, in the team’s words, “under active exploration”.

Fact: the token’s link to Anuma has changed form three times in four months. On 1 June a ZetaStaking contract went live on the ZetaChain EVM (lock ZETA, earn a non-transferable ANUMA token at 12% a year, burn ANUMA for credits one to one), and on 5 June the admin cut its rate by 71%. On 24 June it paused new locks, on 25 June the programme was relaunched as “stake to any validator and we read your balance”, and on 3 July it paused claims and added a withdrawImmediate function the verified source calls the migration path. Lifetime: 79 stakers, 347,842 ZETA in, 159 ANUMA minted.

Take: Proposal 68’s line that “lock ZETA, receive credits … already works” describes the paused contract, and the live programme is something else. Today you delegate ZETA through ZetaHub, sign a message to bind that wallet to your Anuma account, and Anuma’s server grants USD-denominated credits each month, expiring 30 days later, at a rate the help centre says “can change over time”. On the day I read the dashboard, 10,000 ZETA earned 2,377 credits a year, roughly 4% of the stake’s value at Starter pricing, on top of the 9.37% staking yield, and Pro switches on at 100,000 ZETA. Nothing is spent, burned or settled on-chain; “locked ZETA comes out of circulating supply” means a 21-day unbond.

The product’s own money runs the other way. Anuma sells Free, Starter and Pro plans through Stripe, and the credit programme is a cost the company carries to reward holders. Chain fees, the one revenue line DeFiLlama can measure independently, were $13.57 in the 30 days to 21 September and $32,717 over the year. Anuma’s revenue isn’t disclosed anywhere, including in the MiCAR white paper, which describes the company’s 2022 to 2024 income as “modest” and shifting to “sponsorships”.

How to participate

Use Anuma. Sign up at chat.anuma.ai or in the mobile app; a Privy wallet is created for you. The free plan gives 100 credits a month and every mode, including private mode, which routes to open-weight models on Anuma’s own servers and is enforced server-side. Turn on PII redaction in settings if you want emails, phone numbers and card numbers swapped for placeholders before a prompt leaves the device; it’s off by default and it’s regex, so it won’t catch a name.

Delegate ZETA for credits. Move ZETA onto the ZetaChain L1 (Ethereum ZETA won’t count), delegate to any validator through ZetaHub, then connect an external wallet such as MetaMask on the Token Dashboard and sign once. Credits accrue against the bound balance and expire monthly; the ZETA never leaves your wallet. The 21-day unbond spans a snapshot date that hasn’t been published.

Run a validator. 44 of 75 slots are filled, entry is by delegated stake rather than hardware, and rewards stop at the halt. I wouldn’t start now.

Build. The Anuma Developer Platform gives you a React and Expo SDKSDKSoftware Development Kit. A collection of code libraries, documentation, and tools that lets developers integrate a service into their applications without writing everything from scratch. SDKs are how projects become easy to build with.Like a plug-and-play kit for building furniture. You don't have to mill your own wood, forge your own screws, or design the joinery from scratch. The kit gives you pre-cut parts and instructions so you can assemble the thing in an afternoon.SDK in the glossary →, an OpenAI-compatible Portal API and Privy auth, with inference billed from an app balance. The “earn when others use your agents or memory patterns” use for ZETA is a stated intention with no code behind it yet.

Honest assessment

What works

The memory engineering is the best consumer implementation of encrypted, portable AI memory I’ve reviewed. Field-level AES-GCM, a key that never touches disk, chunked semantic search over your own history, curated vault entries with save confirmation, export and delete that behave, and encrypted backup to storage you already own. It’s open source at the SDK layer and the production bundle matches it. Most “private AI” products can’t show you any of that.

The product is live and busy in the way a consumer app should be: mobile, SMS, image and video generation, a council mode that runs one prompt across several models, and weekly changelogs. Private mode on open-weight models works as described and is enforced server-side.

The corporate side is unusually legible for this sector. A named Delaware entity, a MiCAR white paper with the management body listed, DFSA recognition in Dubai, 16 public audit reports on the chain and contracts, and an on-chain vote that, whatever you think of the outcome, was run by the book.

What doesn’t work

The privacy claim is written for the store and read as if it covered the pipe. Anuma’s embedding server and gateway see every message in plaintext; the closed provider sees whatever context the request needs, which is how the proposal phrases it, and that context is your recalled memory. The client repository is private, so the part of the system that decides what leaves your device is the part you can’t read.

The token story is a subscription discount wearing a whitepaper. The 1 June post listed six uses for ZETA:

• Lock for credits: ran for 23 days before the contract was paused.
• Staking: ends at the halt.
• Earn from published agents, x402x402An open payment protocol from Coinbase that repurposes the long-dormant HTTP 402 status code. A server responds with a price, the client pays in stablecoins on-chain, and the request is fulfilled. No accounts, no API keys, no card details.Like a vending machine for HTTP. The endpoint says "pay 1 cent for this", the agent drops in a coin, and the document comes out. Settles on a blockchain underneath, but the payment layer is invisible to the caller.x402 in the glossary → settlement, on-chain memory permissions, decentralised storage: no code.

What survives migration is: hold ZETA, get credits the company prices.

The growth numbers count wallets. “More than 300,000 people” are Privy wallets created at sign-up, the 18 August post says so, and the company’s own 1.2 million inference requests over seven months work out at about four per account. That’s a funnel, whatever the sign-up counter says.

The risk

Proposal 2 doesn’t exist yet. Snapshot height, halt height, claim process, exchange list, SPL mint authority and audit are all “to follow”, and the proposal says timing depends on exchanges confirming a swap. Until then ZETA is the gas token of a chain that has voted to die, on a 21-day unbond, with 49 exchange pairs that each need re-listing.

The chain’s admin surface is wider than the vote suggests. The x/authority module hands a seven-member group with a six-of-eight vote threshold the messages that update contract bytecode, system contracts and TSS fund migration, and a separate policy lets any one of seven act in an emergency. In August the core team shipped a mainnet build of an “emergency TSS drain” that moves every native asset the threshold wallet still holds (90.9 ETH, 1.77 BTC and about $100k of stables on 21 September) to fixed addresses on a trigger from a company-hosted API, with, in the RFC’s words, “zero zetacore changes”. It hadn’t fired when I checked, and it doesn’t need a vote to.

The credit rate is a company decision. It was cut once on the contract path, the help centre reserves the right to change it, and the treasury paying for it holds 24% of supply under the same company’s control.

My position

I don’t hold ZETA. I’d use Anuma for the memory layer and I’d think hard before putting anything sensitive through a closed model on it, which is the same advice I’d give for ChatGPT. I would move on the token if Proposal 2 landed with an audited SPL mint and a snapshot I could plan around, if the credit programme moved on-chain with a published rate, and if Anuma disclosed subscribers or revenue. None of those has a date.

Freedom Score: 39/100

ZetaChain scores 39/100 (F grade). Full methodology at Freedom Score Methodology.

Infrastructure decentralisation (7/20): The L1 is permissionlessPermissionlessA system anyone can use or build on without asking a gatekeeper. No application, no allowlist, no approval step. If you meet the protocol's on-chain rules, you are in. The opposite of permissioned.Like a public road versus a members' club. Anyone with a car can drive on the road; the club checks your name at the door. Permissionless protocols are the road.Permissionless in the glossary → proof-of-stakeProof of StakeA consensus mechanism where validators earn the right to create new blocks by staking tokens as collateral. If they misbehave, the network slashes their stake. Proof of Stake replaced energy-intensive mining on most modern chains.Like being a licensed auctioneer. You post a bond to get the license, you earn fees for every auction you run, and you lose the bond if you rig an auction. The bigger the bond, the more auctions you get to run.Proof of Stake in the glossary → with 44 bonded validators, top-heavy (the four largest hold 37.8%, the eight largest 70.3%) and scheduled for removal. Everything the AI product runs on is one company: the Portal API, the Bifrost gateway “in our own cluster”, the embedding model, the credit ledger and the moderation layer. After migration the decentralised part is Solana, which ZetaChain neither runs nor secures.

Governance decentralisation (6/20): The vote was real and decisive: 265 million ZETA voted, 58% of bonded stake, 99.44% Yes. It was also narrow. Fourteen of 44 validators voted, the five largest (46% of stake) didn’t, and 73 non-validator addresses carried about half the Yes. Three-quarters of supply went to insiders and company pools, the authority module gives a seven-member group six-of-eight admin power and any one of them emergency power, the core team writes Proposal 2, and nothing is defined for Solana.

Token distribution fairness (5/15): 38.5% to contributors, purchasers and advisors, 36% to the treasury and growth fund, 25.5% to community programmes, no public sale. 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.Vesting in the glossary → is meaningful and Proposal 68 leaves it unchanged, which is what keeps this out of the bottom band.

Censorship resistance (4/15): An account-based company product: the bundle carries an account-suspended path, the help centre describes safety screening that quarantines content and a moderation filter on memory saves, and the default routes go to OpenAI, Anthropic, Google and xAI with their policies attached. Private mode on open-weight models is the mitigation and it’s still Anuma’s server. On the chain side, a centrally triggered drain over TSS custody with no governance gate.

Data sovereignty (9/15): The at-rest design is verifiable and good: AES-GCM under a wallet-derived HKDF key held only in memory, in an open SDK the production bundle matches, with export, delete and encrypted backup to your own cloud. The live path is where it breaks: plaintext to Anuma’s embedding endpoint, plaintext through Anuma’s gateway, regex PII redaction that’s off by default, a key derived by silent signing from a Privy wallet, and a closed client. Users control their data with platform dependencies, which is the middle band.

Open source and transparency (8/15): Node, protocol contracts and SDK are MIT, the ZetaStaking and ANUMA contracts are verified, 16 audits are public, and the MiCAR filing names the entity and its officers. Against that, the client and backend are private, the 25 August security patch was applied privately and published nine days later, the credit rate is behind a login and “can change”, and no Anuma usage or revenue figure is independently reportable.

Path to improvement

Three changes would materially raise the score:

1. Take the plaintext off the wire. Client-side embeddings (a small on-device model, or encrypted embeddings the server can’t invert) would make “Anuma cannot read your memory” true end to end. It’s the single change that would move this project out of the F band.
2. Open the client. Publishing ai-memoryless-client would let anyone verify what leaves the device, which is the claim the whole product rests on.
3. Define Solana governance before the halt. Proposal 2 should name who holds the SPL mint authority, how the credit rate is set, and what token holders get to vote on once there is no chain to vote on.

Returns Score: 32/100

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

Token utility (8/20): Today ZETA is gas, stake and vote on an L1, plus the credit programme (delegate, bind, receive credits; Pro at 100,000 ZETA). Gas and voting end at the halt, staking’s future is “under active exploration”, and the one on-chain meter has been paused since June with 159 ANUMA ever minted. What survives is a stake-for-subscription right the company can reprice.

Value accrual (3/20): Nothing Anuma earns reaches the token: no fee share, no buyback, and the only burn is the L1’s base fee on tens of dollars of monthly gas. The credit programme runs the other way, company money to holders. Delegation is a 21-day unbond, not a lock, and the aggregators count bonded ZETA as circulating.

Supply dynamics (9/20): Fixed 2.1B cap and 76.5% circulating per CoinGecko, unverified against the chain. About 490 million ZETA of contributor, purchaser, treasury and growth-fund allocations still vest through February 2028, around 30% of today’s float, with the 24% treasury under company control. The migration adds its own supply event: 18-to-9-decimal rounding, a snapshot with no date, and an SPL mint with no audit published.

Revenue sustainability (4/25): Paying customers exist and nothing is disclosed, so the integrity verdict is unverifiable and the score is discounted to match. The one independent figure is DeFiLlama’s L1 gas, $13.57 in the 30 days to 21 September. Four inference requests per sign-up, on the company’s own numbers, doesn’t read like a subscriber base.

Liquidity and access (8/15): 49 CoinGecko pairs with Coinbase, OKX, Bybit and Gate listed and no Binance spot. Vote-week volume ran at 1.5 times market cap, 54% of it on Upbit and Bithumb won pairs. The token is 97.9% below its February 2024 ATHATHAll-Time High. The highest price a token has ever reached. ATH is usually quoted as a reference point for how far the current price has fallen (or risen) since the peak.Like the record lap time on a racetrack. It tells you what the car has been capable of at its absolute best, not what it will do today. Whether that record gets broken again depends on conditions that may or may not come back.ATH in the glossary →, set its all-time low on 14 August 2026, and every pair has to re-list an SPL token once Proposal 2 names the venues.

Path to improvement

Three changes would materially raise the returns score:

1. Put the meter on-chain and publish the rate. A Solana programme that locks ZETA, mints credits at a published rate and burns them on use would turn a loyalty scheme into a measurable utility.
2. Disclose Anuma’s revenue. Subscriber counts or a monthly revenue figure, ideally with Stripe-attested totals, would let the revenue dimension rest on something other than a discounted self-report.
3. Land Proposal 2 with an audit. A dated snapshot, a named mint authority and an audit of the claim contract would remove the migration discount from supply dynamics and liquidity in one move.

Score change log

Date	Change	Freedom	Returns
2026-09-23	Data. Integrity records re-cut: 7 claims now quote ZetaChain’s own pages verbatim and 3 are our own measurements. Scores unchanged.	39	32
2026-09-22	Initial full review following the 21 September triage. Freedom 39/F, Returns 32/F.	39	32