Apps for humans.
Market data for AI agents.

The same liquidation tape, sold per call in USDC on Algorand mainnet. Five endpoints: volume for one symbol split by exchange and by whether longs or shorts were flushed, a cascade score for whether that volume landed in one burst or spread evenly across the window, the full symbol universe so a caller can tell coverage from silence, per-exchange reporting integrity, and a symbol's complete recorded history for five dollars. Two of the three venues throttle to one update per second and undercount exactly when it matters, and the integrity endpoint says so. Every response says what it is: measured, absent when the market genuinely had none, or unmeasured when our own lookup broke. Unmeasured returns 503 and is never billed, because charging for a zero we invented would make every other number worthless. Discoverable in the x402 Bazaar and the MCP registry. No signup, no API key.

Anyone can put a dataset on chain. Nobody checks the bytes. corpus makes each dataset epoch an ERC-721 vault and puts a machine appraiser between the data and the buyer. The appraiser reads the actual file, runs eight checks covering schema, row count, null rate, types, time ordering, label ordering and outliers, scores it, signs the report and posts the verdict on chain. A refusal is not a low score, it is a hard gate: the purchase reverts and the vault cannot be sold even by its owner. Two vaults are live on X Layer mainnet holding production data. The launch outcomes vault carries 74,802 labels checked against their own observation window close, zero violations, so its no-lookahead property is machine verified rather than asserted. Delivery is hash gated end to end: the gateway finds a file by hashing candidates against the content hash the contract stores, so it cannot serve bytes the vault never committed to. 21 contract tests, 13 gateway tests, MIT.

FTSO publishes the price. haircut publishes whether that price is reachable. FAssets mints FXRP against collateral on Flare, and every collateral ratio you can read on chain is computed at the oracle price. Price the FLR leg at what SparkDEX would actually pay and it recovers 0.09% of its marked value — agents whose pool ratio reads 209–650% land at 0.20–0.61% once repriced, against a 150% minimum. This is not a solvency claim: the USDT0 leg covers FXRP three times over on its own, the Core Vault holds 7.4M XRP against 1.77M FXRP outstanding, and every agent is fully backed. The exposure sits with collateral pool providers. Readings land in an append-only registry on Flare mainnet that stores refusals as first-class entries, because a gap nobody recorded is indistinguishable from a period nobody looked.

A lending market marks collateral with an oracle. A liquidator only acts if the liquidation bonus beats the cost of getting out. overhang measures the difference across every reserve on Solana's tokenized-equity lending markets, quoting real sell-side exits at size and bisecting for the largest clip that still clears. Running every 15 minutes since 11 August: 2,158 sweeps and 25,896 measurements with zero gaps, each one carrying a size-matched control quote so a failed pipe is never published as an absent market. On the xStocks board 82.6 percent of marked collateral cannot be realised, dollar-weighted. The sharper finding is on the other market: Superstate's Opening Bell tokens are transfer-restricted to allowlisted wallets while Kamino's liquidation instruction is permissionless, and the router refuses to quote them at all — TOKEN_NOT_TRADABLE on 2,167 of 2,167 measurements, $24.0M of collateral whose liquidator market cannot exist. A terminal verdict requires six consecutive agreeing observations, so one bad quote never becomes a finding. Curators and oracle vendors are paid by the protocols whose parameters they set. This is the independent read, and it is sold per query in USDC. 115 tests, MIT.

Agent memory that can be cross-examined, built on HydraDB. Every answer carries the claims it rests on and the traversal that found them, and every refusal carries what was searched and why nothing qualified. Scored on all 500 LongMemEval questions: 0.3511 accuracy against a no-memory baseline of 0.000, graded by a disclosed judge stronger than the answerer, with zero questions left ungraded. When it answers it is right 70.5 percent of the time, and 28 of the 30 unanswerable questions were correctly refused. The finding is that the failure mode is over-refusal rather than hallucination: 133 questions were refused with the gold session sitting in the evidence, and every one of them is clickable in the published register. Refusing with the answer present and refusing with nothing retrieved are counted separately, because they need opposite fixes. It won Best Use of HydraDB at Hack Hydra 2026. 334 tests, zero dependencies.

Evidentiary memory for agents, on CockroachDB. An agent that remembers everything it was told remembers the lies too. alibi keeps provenance attached to every stored fact so a claim can be traced back to who said it and when, and quarantines what cannot be vouched for rather than letting it into the answer path. Built on CockroachDB's managed MCP server and distributed vector indexing, with the cluster and object storage on AWS. 127 tests.

A solvency attestation that proves a lending book covers its debt at prices it could actually exit at, not at the oracle mark, without publishing a single position. Selling into a constant product pool moves the price against you, so a book that reads as comfortably collateralised at the mark can be under water the moment anyone tries to realise it. datum puts that depth adjusted comparison inside a zero knowledge circuit: venue reserves, block heights and the required ratio land on the ledger where anyone can recompute them, while position sizes, debt and claimed proceeds stay private witness and are never disclosed. Written in Compact for Midnight and deployed on Preprod, a public test network, where the attested verdict reads NOT COVERED at block 2,295,258. The page reads that attestation straight from the public indexer with no wallet and no install, and a Verify button recomputes the venue hash in the browser.

Portable credit history, proven rather than reported. Every borrow, repayment and liquidation on Aave v3, Morpho Blue and Spark is turned into a record on Creditcoin, and a record exists only because the Attestcoin native query verifier proved the Ethereum mainnet transaction behind it sat in an attested block. No oracle, no indexer taken at its word, no self report. From those records a deterministic score is computed on chain, readable by any lender. The recorder has been tailing the three pools unattended and has put over 36,000 proven credit events on chain across more than 6,200 scored borrowers, with running totals published at /totals so the count can be checked rather than believed. The contract has no owner, no admin, no pause and no way to edit or delete a record. Runs on Creditcoin CC3 testnet.

An agent buying data per call has no memory of what it bought. It pays, gets an empty payload back, and next run it pays the same provider again. Every purchase is a cold start. cassum keeps one record per provider — calls, empties, USDC spent, and a journal entry per paid call holding the settlement transaction hash — then ranks providers on cost per delivered payload instead of sticker price. The proof is a deletion test the page runs server-side on click: the same workload takes 24 paid calls with memory and 80 without, for the same 20 delivered payloads. That fleet is simulated on purpose — a live run pays a treasury this project controls, so any saving from it would be circular. Settlement is real, on Base mainnet or Solana.

A lender that prices collateral at an oracle mark is solvent only on paper: the mark says $100,000 and liquidation discovers the pools return less. statera measures what a tokenized stock would actually fetch if that size were sold into live X Layer liquidity, and publishes it on chain — 18 rows covering TSLAx, NVDAx and SPYx, raw and wrapped, at $1,000, $10,000 and $100,000. A keeper checks every ten minutes and posts on a 100-minute heartbeat or a move above 100 bps. The worked lender beside it reads the realisable value and never the mark, and refuses outright when a row is stale, unmeasured, or says the pools cannot fill that size at all. The publisher is immutable: no owner, no upgrade path, no pause.

An approval gate for agents, sold by the call. The agent says what it is about to do, speculum decodes the transaction itself, and the two are compared before anything is signed. PASS means the bytes do what was declared. BLOCK means they disagree or the action cannot be undone, so a human is needed. REFUSE means the effect could not be determined, and it outranks BLOCK because the service does not guess. A check costs 0.001 HBAR, or 0.005 with a live balance simulation, over x402 v2 on Hedera testnet. Every paid verdict is written to a public Hedera Consensus Service topic that anyone can read. Built at ETHOnline 2026.

Talk to any MCP server. Type a website or paste a remote server's URL, press talk, and its tools become voice tools for AssemblyAI's Voice Agent API, which has no MCP support of its own. A sweep of the official MCP registry found 12,012 servers that list their tools with no login, and all 202,191 of those tools converted. interpres connects only to servers like that and never sends them a credential. Sessions end after five minutes, or after one minute in which nobody speaks, and it logs tool names and timings, never IP addresses. Built for lablab.ai's AssemblyAI Voice Agent hackathon.

The same per-call model on a third chain family. Three priced endpoints — ticker, spread, volatility — plus a free discovery route, metered in USDC over x402 on Aptos. Payment challenges arrive base64 encoded in a response header rather than the body, and the facilitator sponsors gas, so a caller spends USDC and nothing else. No signup, no API key. Testnet, with mainnet held deliberately as future work rather than shipped early.

Five collectors publish a dated finding every day under CC BY 4.0, with permalinks, JSON-LD and a machine feed. One measures what lending collateral can actually be sold for: two Superstate reserves on Kamino have never routed at any size across more than 1,100 consecutive sweeps, while Kamino's liquidation instruction stays permissionless.

31,228 paired observations of tokenized US equities against the stock they track — 11 symbols on Raydium, 19–29 July 2026, on a 300-second cadence at 99.1% coverage. Median deviation is 24 bps while the underlying is trading and 52–58 bps when it is not, because the reference price freezes at the closing bell and the token does not. On MSTRx and CRCLx the gap runs 7.5× wider — the two most crypto-correlated names in the basket, repricing news no closing print can hold. Raw ticks and a daily aggregate on Hugging Face, chart on Dune, CC BY 4.0.

Drop AgentFeed into any elizaOS agent. One npm install gives it an action for each of the 83 paid data routes and a free one: the liquidation tape, squeeze scores, the funding radar, orderbooks, whale positioning, Solana token checks, sanctions screening. It pays per call, only for what it uses.

Trust ratings for tokenized real-world assets. Five transparent pillars — peg integrity, backing, issuer, liquidity, redemption — scored from live on-chain data, with every score showing its reasons.

Give it a web page or PDF link and the fields you want, and it returns one JSON object. You name up to 20 fields and their types: text, number, true or false, date, list. A field the page does not state comes back null, never invented. It reads one URL at a time over a plain fetch, so it does not crawl or run a browser, and it says how much of a long document it read. A page that states none of your fields is not charged. The same extraction is the extract tool in AgentFeed.

The next FOMC rate decisions and the CPI, nonfarm payrolls and PCE release dates, each with its New York time, the same moment in UTC with daylight saving handled, and the official Fed, BLS or BEA page it was read from. Dates are read from those pages, never typed in. It says plainly what it is not: dates and times only, four US releases, no actual or forecast values.

Cleans the email list a scraper left behind. It strips mailto:, %20 and u003e debris, drops junk that was never an address (logo@2x.png, tracking ids at sentry hosts, name@example.com), removes duplicates, and checks each domain with DNS: does it take mail, is it disposable, is it a role account, is it a typo of gmail.com. It does not check that a mailbox exists: outbound port 25 is blocked on the platform, so there is no SMTP probe and it never says an address is deliverable. A repeat, an unfinished lookup and a line that is not an address are never charged.

For BTC, ETH, SOL or any Hyperliquid perp: the USD at risk in each 5% price band, long and short, read from the real open positions of the top 250 Hyperliquid accounts and the exchange's own liquidation prices, not from an assumed leverage mix. Each coin also lists its largest positions with entry, liquidation price, leverage and unrealized PnL. It says plainly what it is not: Hyperliquid only, those accounts only, a snapshot refreshed every 5 minutes at most, and notional, not margin. A coin with no bands is not charged.

PDF links in, markdown out, page by page: every page under its own heading, with the page count, word count, file size and the PDF's title when it has one. Up to 50 public links a run, each fetched once, not stored, robots.txt honoured. It reads the text layer only: a scanned PDF with no text is not OCRed and is not charged, tables and multi-column pages may not read in order, and text is cut at 300,000 characters and flagged truncated.

Search papers by topic or look them up by DOI and get one JSON item per paper: title, year, DOI, authors, venue, citation count and a link to a free copy when one exists. The data is OpenAlex's open CC0 dataset, not a scrape of Google Scholar, so coverage and citation counts will differ from Google's. Up to 25 papers per search, abstracts on DOI lookups only. Searches share a daily OpenAlex allowance, and a search skipped when it runs out is not charged.

A US-listed ticker in, its latest EDGAR filings out, newest first, one JSON item per filing: form type, filing date, report date, accession number, description, 8-K item codes and links to the document and the filing index. No CIK lookup. 10-K, 10-Q, 8-K, Form 4 and the rest, up to 50 filings per ticker. It returns metadata and links, not the filing text, covers US-listed tickers only, and is not affiliated with or endorsed by the SEC.

Sharp-money detection on pre-match odds. It watches the line move against the money and flags the moves that come from informed volume rather than public sentiment — the ones books adjust for quietly. Built for the TxODDS Trading Tools track, where it took 2nd place and 4,000 USDT. The interesting engineering problem was proving a detector works when it has correctly fired zero times: no alerts turned out to mean thin fixture coverage, not broken code, and the difference was only provable by driving a synthetic tick through the live path end to end.

Live World Cup alerts for people who cannot watch the match. Goals, cards and state changes pushed to Telegram the moment they happen, built for the Consumer & Fan Experiences track. The detail that mattered was starting the watcher before kickoff rather than at it — a bot that wakes up when the match begins has already missed the first goal.

The starter kit I wish had existed before shipping eight apps to the Solana dApp Store. Seed Vault wiring, wallet session handling, the publishing pipeline and the parts of the store submission that are only obvious after you have failed them once. Extracted from production apps rather than written as a demo, so the defaults are the ones that survived review.

Who pays when an AI agent pays wrong. The person signs one Solana transaction that caps what the agent can spend, the seller checks it before it takes the money, and a resolver assigns liability from signed receipts and the chain. 300+ tests, and devnet runs that settled through the live payai and x402.org facilitators. Nothing on mainnet yet.

Rehearse the work conversation you are dreading. An AI counterpart argues back like a real person, and you choose how it plays: cooperative, defensive, passive-aggressive, stonewalling or emotional. Every rehearsal ends in a debrief that scores you 0 to 10 on clarity, specificity, empathy, firmness and outcome, quotes your strongest line back and rewrites your weakest. Twelve scenarios, from asking for the raise to saying no to the project that will bury your quarter. No account, and your history stays on the phone. Three rehearsals are free.

Never worse than the best pool. Sent to the best single pool, a $100,000 stock-token order on Robinhood Chain left a median $199 on the table in 6,329 of 7,493 executable quotes. partitio splits an order across pools only when the split pays and floors every fill against Chainlink inside the contract, so a price outside the band reverts instead of filling. The relayer pays the gas: real round trips have filled on mainnet from a wallet that holds no ETH and has never sent a transaction. 34 stock tokens, an ownerless and immutable router, and a beta that caps trades at $50.