The neutral meter for AI agent work: neither side keeps the count.

Buyers and suppliers close invoices on evidence both can verify.

Of 241 merged agent pull requests that claimed passing tests, 30 had a failed check.

Check your invoice

The claim

Two parties who distrust each other compute the same bill from evidence a third party signed. The program releases the money on that signature, with no company and no oracle in the middle.

Limits, in one line: Solana devnet, test USDC, no outside users yet (docs/DISCLOSURE.md).

The meter: two ledgers, one bill

1. Buyer and supplier each keep their own ledger of the evaluations GitHub signed.
2. Each runs knos meter reconcile on its own copy and rebuilds the same statement, line for line.
3. A line is agreed, disputed, duplicate or insufficient evidence, and only an agreed line is billed (docs/METER.md).

The money: released on a signature

1. The buyer fixes the work, the budget and the acceptance terms before it starts; the terms are hashed into the order.
2. GitHub signs the workflow run that judged the work, and a Solana program checks that signature itself before it pays, in test USDC.
3. The smallest example is a bounty: a maintainer comments /knos fund 20 checks: test, and a merge with test passing pays the author.

The number

241 merged pull requests by AI coding agents said tests or CI pass; 30 had a failed check at the head commit (docs/backtest.json, counted by scripts/backtest.py). The second number: 17.8% of first such pull requests, 147 of 826 repositories (docs/BENCH.md). A failed check is GitHub's record, not a judgment of why it failed.

What is real today

Read more

uvx knos shadow invoice.csv: the same check from a terminal, on an invoice of your own. docs/STORY.md: the three-minute demonstration in six beats, each with its evidence. docs/MANIFEST.md: this release on one page: source, the build live at each public program id, every capability's stage, the limits. docs/README.md: every document, under six questions.

Below this line: records that scripts write from their sources, and the licence. Nothing here is typed by hand.

This page names no time for an upgrade, so that it is true before and after one. Whether a proposal is pending, has executed or was withdrawn is read from the chain by knos status, the site's banner and the site's upgrade record (web/upgrades.json is the committed copy). The first deployment (programs) has no upgrade authority and stays as a record; nothing new is funded there.

Reproduced by someone else: none recorded yet. Exercised on devnet: verify_github, fund_from_wallet, refund, work_orders, order_pay, tests_mode, order_auto_accept, top_up, single_use_tokens, meter_batch, meter_seller_claim, passkey_funder, passkey_fund_relay, buyer_page, x402_knos_order, outcome_not_code. Deployed on devnet: verify_gitlab, key_guardian, fund_by_comment, pay_on_merge, hold_and_bind, pause, meter_single, passkey_payee_wallet, upgrade_gate. Everything else is tested locally, implemented or not built: the table with the evidence has one row for each capability, from docs/capabilities.json. Deployed and exercised are counted only at the public program ids; what the 0.3.14 rehearsal ran at staging addresses of its own is in the note of each capability it ran, with its transaction.

MIT, all of it. Built by drexthealpha. Its memory engine is Sibyl.