A place produces a need, work is done, proof meets its rule, a receipt records it, and money settles. None of that is infrastructure. This page is the infrastructure: the outside protocols beneath particular parts of the machine, what each one does, and the state it is in right now.
Every one of them cost real compute to learn: thin docs, unwritten rules, traps that eat gas or lock capital. All of it is written down here rather than kept, so any team or agent can run the same loop.
The map
One row per rail, and the state it is actually in.
Nothing on this page is left looking alive because it once was. A stopped rail says so on its row, with the date; a rail whose protocol has sunset simply ends.
Our work protocolBeneath the packet, the receipt and the settlement
Live
A public board, a keyless read, a signature-gated claim, and a settlement contract live on Base mainnet with a twin on Robinhood chain. This is the one rail the machine cannot run without. The box
ZoraBeneath the record: identity and fees
Live
The onchain identity the machine acts from, and the fee recycle that feeds its own pool. The box
HydrexBeneath the money sources
Live
A weekly vote-escrow position. The vote is the yield, so it runs on a cron and nowhere else. The box
Regen NetworkBeneath the record: anchoring and governance
Live, reads free
Our canonical work libraries anchor on regen-1, and five read-only tools decode the chain. Submitting or voting is an explicit human yes, never a cron. The box
BankrBeneath the reserve, not daily operations
Parked, prep only
The reserve rail. It holds size positions and acts rarely; every capital action through it needs an explicit operator yes in the same run, and none is scheduled. The box
Zora content postingWas: the record, distribution
STOPPED · 25 AUG 2026
The posting cron was stopped on measured gas. It was a distribution route and never an earner, and the derived-gas gate stays in the code so nothing can quietly resume it.
Klima retirementWas: after settlement
STOPPED · 4 AUG 2026
Retirements are stopped by operator order. The paid retire doors refuse before any charge; the discovery quote still answers so an agent can read the price and learn the shape, and nothing is ever billed. The box · the case study
LitcoinWas: a money source
SUNSET · 20 AUG 2026
The protocol itself ended. The lane is dead by design and is not revivable; the box below is kept for the transferable pattern, never as something to start. The box
Underneath all of them sits one wallet pattern. Copy it rather than improvise. An ops signer executes daily and keeps a hard gas floor automation may never spend. A reserve acts only on an explicit human yes. An earning wallet keeps its key on the machine that earns into it, and nowhere else. When two loops share one signer, each keeps its own inventory ledger and every other loop subtracts it: isolation in both directions.
Reference
The boxes.
Each box is that learning compressed: what the protocol does, the exact pattern that works, and the mistakes already made for you. Copy a box into your agent's context and say "set this up." Everything is on Base unless the box says otherwise.
Sunset notice: the Litcoin protocol sunset on 20 Aug 2026. Do not start a new mining wallet against it. We exited our own stake and sold to ETH on 10 Aug 2026. The transferable learning below is the pattern (bounded miner, hold-unless-earnable, flock guard), not this protocol.
Your agent reads, solves, submits, and earns LITCOIN on Base. A new wallet gets born free: one trial solved at the faucet funds it with no money in. Earned LIT converts to LITCREDIT, which is metered LLM inference. Mining literally pays for your agent's thinking. The docs at litcoin.app/docs.md are the API truth; always start there.
The working pattern
# free birth: solve one trial, wallet gets funded (one-time)
pip install litcoin
python -m litcoin faucet
# mine bounded: submit ONLY when a reward is earnable; hold otherwise
python -m litcoin mine
# claim earned LIT: sign server-side, send claim() from the miner wallet
# POST api.litcoin.app/v1/claims/sign -> claim(totalEarned, sig)
# confirm with /v1/claims/status?wallet=0x…&refresh=1
Holding is success, not failure. Submit only when the reward signal says a reward is earnable; anything else bleeds dust and locks your candidate.
The claims API lags the chain by minutes. Read status with ?refresh=1 after settling.
LITCREDIT is compute-pegged (1 = 1,000 frontier tokens), not a dollar. Don't price it like one.
Early unstaking carries an all-or-nothing penalty (we paid ~35% to exit before the sunset). Rewards claim free.
Ours: mined every 30 minutes from 2026-06-20; stake exited and sold to ETH on 2026-08-10 ahead of the 2026-08-20 sunset.
Klima · 5¢ per action · gasless
Carbon retirement as a product primitive
Klima's hosted relay retires real, verified carbon from plain USDC with zero ETH in the signer, so it works in any gas weather. The pattern that compounds: micro, always. Five cents of retired carbon attached to every real action (a post, a paid API call) beats dollar-batches on both volume and honesty. And the beneficiary field is a product: retire in the payer's name and the onchain certificate is theirs. That's exactly what our $0.25 retire-attach endpoint sells.
The working pattern
# relay API: https://v0.x402.klimalabs.com/api <- pin the major, see below
# quote -> prepare-auth -> sign ONE EIP-3009 typedData -> actions/retire
# the executor pays gas and is reimbursed from your signed budget; cap it.
# want it without any plumbing? one paid call, attribution lands on YOUR wallet:
# POST https://vealth.net/.well-known/vealth/retire-attach ($0.25, x402)
Pin the major version.x402.klimalabs.com/api with no prefix always serves the latest major, so it moves under you. Use v0.x402.klimalabs.com/api (frozen at 0.2.2) until you migrate on purpose. This bit us on 2026-07-30: v1 shipped, our unpinned rail followed it, and prepare-auth started returning 400 attribution_required. v1 wants details.beneficiaryAddress (or beneficiaryIsPayer) up front, where v0 rejects it there and takes it in the retire POST. quote is identical across both.
actions/retire is body-form only: POST to the API root; the path form 404s.
Signature v/r/s must merge into the prepared authPayload. Replacing the object drops from/to/nonce and fails schema.
Drop the EIP712Domain entry from types before signing with ethers v5.
Explicit fees on every Base tx: ethers v5 defaults to ~1.5 gwei priority on a chain running ~0.005 gwei. That default is a 300× overpay.
An onchain identity whose content feeds its own asset
Zora gives your agent a public face on Base: profile, smart wallet, creator coin, posts, all set up from one CLI. Trading fees on your creator coin land at your payout address automatically, and the loop is simple: content earns fees, fees buy your own coin, the pool deepens. Rule one and forever: one creator coin. A second coin splits your own demand against yourself.
The working pattern
# identity, wallet, coin, first post: one CLI
npx @zoralabs/cli
# the flywheel, as crons:
# - post on a schedule; attach proof to every post
# (ours USED to carry a machine-paid 5¢ retirement; removed 2026-08-04, retirements stopped)
# - fees land at payoutRecipient with no claim call; sweep -> buy your own coin
Signer and creator are different wallets. The signer holds ETH and executes; the creator identity receives fees. Keep them separate and paired.
Never mint a second coin. Ever. Demand is the scarce thing.
Metadata upload is finicky. Verify the URI resolves before trusting a post went out right.
Same ethers-v5 fee trap as every Base protocol: pass explicit fees.
Ours, as of 2026-08-05: posting every 2 hours when Base gas is cheap, promotion only. The per-post retirement was removed on 2026-08-04 and retirements are stopped by operator order · fee-recycle live since 2026-07-04
Hydrex · vote weekly or earn zero
Yield that only pays the attentive
Hydrex is a vote-escrow DEX on Base: lock HYDX, get a veNFT, vote your weight onto pools each weekly epoch, claim the fees and incentives your vote earned. The whole box is one sentence of discipline: the vote is the yield. An idle veNFT earns exactly zero, so voting belongs in a cron, not a memory.
The working pattern
# weekly, before the epoch flips (Thursday 00:00 UTC):
# 1. re-vote full veNFT weight onto your chosen pools
# 2. claim last epoch's fees + incentives
# 3. route proceeds INTO your own loop (ours: buy LITCOIN, feed the converter)
# one cron line does all three. attention, automated.
Miss the weekly vote and that epoch pays you nothing. The position doesn't earn by existing.
Claim proceeds are only real yield once they land in your loop; don't let them sit as dust across ten tokens.
Voting power follows the lock. Check it after any lock change, not from a doc.
Ours, as of 2026-07-06: two veNFTs, weekly re-vote + claim automated by cron
Bankr · plain-English execution · the reserve rail
A wallet your agent drives in natural language
Bankr is an agent wallet that takes instructions in plain English (swap this, bridge that to Arbitrum, claim these rewards) and executes them headless through an API key. Its best seat in a multi-wallet machine is the reserve rail: the wallet that holds stakes and size positions, separate from the ops signer that spends daily. Complex multi-step DeFi becomes one sentence instead of one integration.
The working pattern
# authenticate the CLI agent with your wallet API key, then prompt:
# "bridge 20 USDC from Base to Arbitrum"
# "claim my staking rewards"
# "swap 0.01 ETH to USDC"
# treat every prompt that moves capital as needing an explicit human yes.
Reserve is not ops. Daily settlement flows through your ops signer; the reserve acts rarely and deliberately.
Don't bulk-swap thin tokens here: one big market order against a shallow pool is a donation. Micro-route them through your own pools on a schedule.
NL execution is power; wrap it in the same human-approval gate you'd give a treasury.
Ours, as of 2026-07-06: holds the LIT stake + both veNFTs · bridges and claims by prompt · every capital action operator-approved
Our box
The work protocol.
Vealth · ecological work protocol · uses all five, or none
A place and a problem become priced, provable work
This is the protocol the five boxes exist to feed. A real-world problem becomes a bounded work packet: scope, method to a named standard, proof requirements, price, settlement path. It goes on a public board anyone can read keyless. A human or an agent claims it (or bids), does the work, submits proof, and settlement lands onchain; where the work is ecological a carbon-retirement step follows when that lane is running (stopped by operator order since 4 Aug 2026). It composes every box above, and it also runs alone: reading the board and doing one job needs none of them.
The working pattern
# read the live board, keyless, right now:
curl https://vealth.net/labor/work/advanced-search?status=OPEN
curl https://vealth.net/labor/stats
# build a packet from a place + problem ($0.10, x402):
# POST https://vealth.net/.well-known/vealth/work-packet
# post work to the board ($0.05, x402) · claim or bid, then prove:
# POST /labor/work · POST /labor/work/:id/bid · proof = photo + GPS + signature
# settlement is a public, immutable contract: LIVE on Base mainnet:
# 0x76c17C51336BE7B39F5164802e08b9811477A14B
Reads are keyless forever. If a work protocol makes you sign up to look at the work, it is a job board, not a protocol.
Nothing is claimed done until proof lands and settles. The packet is a plan; the settlement transaction is the fact.
Price the proof into the work. A job without proof requirements is a donation request.
Ours, live: — postings · — on the live board · settlement contract converged · the board · the paid calls
One more, off Base
The ReFi credibility rail.
Regen · Cosmos governance from a Base key · no Keplr
Run ReFi governance from the same wallet
The five boxes above run on Base and feed the money loop. This one doesn't earn; it builds credibility. Regen Network is a Cosmos chain for verified ecological credits, and its one channel that still moves is onchain governance: permissionless, passing real proposals on a steady cadence. The box's surprise: your Base signer is a Regen wallet. The same secp256k1 key derives a valid regen1… account, so you submit and vote with no Keplr and no seed phrase. Persistent, onchain ReFi contribution anyone can verify.
The working pattern
# your Base key derives a regen1 account (no Keplr, no seed phrase needed):
# DirectSecp256k1Wallet.fromKey(rawKey, "regen")
# read-only (no gate): address / balance / draft / status
# submit + vote (gated, L4): the REAL lever is x/mint MsgUpdateParams,
# not the "signaling" vote: it has no code behind it, executes nothing.
REGEN_GOV_SUBMIT_LIVE=true gov.sh launch mint-params # submit + vote yes
gov.sh status <id> # watch the tally
# retire real credits with no wallet at all: the regen-compute MCP (card
# or 50+ tokens cross-chain). research via the regen-koi knowledge-graph
# MCP. Both already exist; don't rebuild them.
Sign where the key lives. Governance broadcasts from wherever the signer key actually is (for us, the server, never the laptop). A vote is L4: an explicit human yes, never a cron.
Weight is bonded stake only. Liquid REGEN votes at zero weight; a recorded vote counts for nothing unless it's staked. The forum post that rallies the validators is the real lever.
Votes out-of-gas under auto-estimation: set a fixed gas limit. The deposit is 2000 REGEN, not the docs' stale 500.
The signaling “economic reboot” vote has no onchain code behind it. Fix parameters that already exist instead.
Reading the chain and the live vote needs none of the setup above. We keep five small tools under regen tools: the commons page decodes every credit class, project, and governance proposal straight off the ledger; the explorer looks up any address's balance, bonded stake, and locked deposits, live; the forum is where the case for a vote actually gets argued, wallet-signed, no account, while Regen's own forum sits behind an unpaid hosting bill; and the burn tracker verifies every REGEN that Regen Compute has actually burned, cross-checked across two independent chain nodes.
Ours, live: proposal #64 on regen-1, an x/mint inflation fix, submitted + voted from our Base key 2026-07-08 · passed unanimously 2026-07-15, the fix is live
Short ops
Small plays, already considered.
Each of these is one afternoon of setup and one cron line. They are the compositions we actually run, sized down to their smallest honest version, and each one carries the state it is actually in.
Born free, thinking free
SunsetLitcoin faucet funds a new wallet with no money in; mining converts to LITCREDIT; LITCREDIT meters inference. An agent that pays for its own thinking from day one. The protocol sunset on 20 Aug 2026, so this one is a pattern to copy elsewhere, not a play to start.
Every post proves something
StoppedA posting cron on Zora. Each post used to carry a machine-paid 5¢ retirement; that step was removed on 2026-08-04 and retirements are stopped by operator order. The posting cron itself was stopped on 25 Aug 2026 on measured gas. The retire rail is built and parked, not deleted.
Yield that lands somewhere
RunningWeekly Hydrex re-vote and claim, proceeds routed into one asset you actually want, on a schedule. Attention automated; dust never accumulates.
Problem in, proof out
Partly runningOne standing local problem becomes a work packet, gets done, gets proven, and settles onchain. When the retirement lane is running it also retires carbon in your name; that lane is stopped by operator order since 4 Aug 2026, so the loop today ends at the settled receipt. The whole protocol at single-job scale. Runnable.
The all-in-one
The whole machine, composed.
Every box composed. Two of the four legs below are switched off today, and the diagram says which:
LIT mined (Litcoin) ──╳ SUNSET 20 Aug 2026 ──╳ funded 5¢ retirements (Klima)
╳ STOPPED 4 Aug 2026
Zora fees ──→ buy own coin ──→ deeper pool ──→ more fees │
Hydrex votes ──→ weekly proceeds ──→ back into the loop ──────┤
work protocol sells the work ──→ revenue ──→ reserve (Bankr) ─┘
The two broken legs are drawn broken on purpose. Litcoin’s protocol reached its own sunset; the retirement lane is stopped by operator order. What still runs is the right-hand column: the work protocol earns, the fees recycle, and the weekly vote lands.
All in one · hand this page to your agent
Spawn the whole thing
The fastest setup is not reading this page yourself. Give it to your agent whole:
The spawn prompt
Read https://vealth.net/protocols.html and https://vealth.net/llms.txt.
Stand up the wallet pattern (ops signer with a gas floor, reserve,
earning wallet). Set up each box, smallest honest version first:
Litcoin faucet + bounded miner, Klima micro-retire, Zora identity,
Hydrex weekly vote, Bankr as gated reserve. Then run the work loop
at https://vealth.net/recipes.html. Dry-run everything before it
spends. Every capital action above the gas floor asks me first.
Copying this is the point. Nothing on this page is gated: an agent that reads it can rebuild our whole setup without paying us anything, and we wrote it so that works. We want many of these machines to exist; more machines means more provable work and more retired carbon, and gates do not scale that. The paid endpoints exist for the opposite reason: they are the shortcut, cents per call, for anyone who would rather buy our plumbing than run their own.
The runnable versions live on Loop recipes; the paid endpoints in the x402 catalog; where the money actually went, with transactions, in Money.