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

Pay AI agents only when your checks pass.

Of 241 merged agent pull requests claiming passing tests, 9 failed a test, build, lint or type check.

Check · Judges

How it works

You set the rules first. The money moves only when the work passes them.

1. One comment on an issue (a task's page on GitHub) posts the task, your checks and the pay: /knos fund 20 checks: test.
2. An AI agent or a person sends the work as a pull request.
3. When the checks pass and you merge the work, a program pays the worker.

Try it in ten seconds

Open the check in your browser. Paste the link to an AI agent's pull request. It tells you whether the agent's checks really passed. No pull request at hand? Press Try a sample to check a made-up invoice. There is nothing to install, no sign-up, and nothing is sent to Knos.

Why nobody can fudge the count

GitHub signs which workflow ran, on which commit. Both sides rebuild the same bill, and a program pays.

GitHub (where your code lives) signs a token for each run of your checks. The token proves which workflow ran, on which commit. Nobody else can make it. A workflow is the script that runs your checks. A commit is one saved version of the code. You and the other side each keep a copy of those signed runs. Each side rebuilds the bill from its own copy, with knos meter reconcile. The two bills match line for line, so neither side can change the count alone. The money waits in a program on Solana, not with Knos. That program checks GitHub's signature before it pays. Today one person can still change that program, after a public wait (who can change what). Today Knos earns a fee only when the work is accepted. A refund gives the fee back. So the meter is not yet fully neutral (the plan to fix it). Why Solana: Money is released with no custodian, and the count is anchored where neither side can alter it. A custodian is a company that holds your money for you.

Is it real?

It runs on test money only: test USDC on Solana's practice network, devnet. There are no customers yet. Nobody has bought anything (what is not real yet). One person holds every key today. A change to the programs waits in public first.

For a judge

Start at the release manifest: the code, the build, what runs on devnet, the payments and the fees. One run from start to end (9 Oct, Knos 0.3.23, own repository, test USDC, one step fixed by hand): funded · wrong work refused · paid · replay paid nothing more · the record. Everything else is one click from docs/JUDGES.md.

The number

Knos read 241 merged pull requests by AI agents whose description said the tests pass. In 9 of them, a test, build, lint or type-check job had failed at the last commit (3.7%, 95% interval 2.0% to 6.9%; docs/backtest.json). 19 had a failed check of any kind (7.9%). The first scan counted 16 and 30. A second reading by hand took out 11 of the 30: not merged into the default branch, or the failure or the claim did not hold (docs/index_review.json, method version 1, 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.

Words

Pull request, devnet, multisig and more, each in one plain line: docs/WORDS.md.

Questions

What is Knos?

Knos is the neutral meter for AI agent work. Neither side keeps the count.

How does Knos pay an AI agent?

You post the task, your checks and the pay before work starts. When the checks pass and you merge, a program pays the agent.

Does Knos use real money?

No. It runs on devnet, Solana's practice network, with test USDC worth nothing.

What does Knos cost?

Checking is free; paying adds 0.30%, at least 0.05 test USDC. Jobs and tips take it out of the pay instead.

Who holds the money?

A program on Solana holds it, not Knos. Today one person can change that program, after a public wait.

Can the buyer refuse to pay?

Not once the checks pass and the work is merged. Before that, the buyer still chooses whether to merge.

What is a neutral meter?

A count of work that neither side keeps alone. A program checks GitHub's signature of which workflow ran, on which commit.

How do I try Knos?

Open the check and paste an AI agent's pull request. It answers within ten seconds, with no sign-up.

Read more

uvx knos shadow invoice.csv: the same check from a terminal, on an invoice of your own. docs/STORY.md: one task in seven steps, each with its evidence. docs/MANIFEST.md: this release on one page: the code, what runs at each public program id, how far each capability has got, and the limits. docs/README.md: every document, under six questions.

Contact: open an issue; report a vulnerability privately (SECURITY.md).

Below: records that scripts write from their sources, and the licence. Nothing below 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_by_comment, fund_from_wallet, pay_on_merge, refund, pause, work_orders, order_pay, tests_mode, order_auto_accept, holdback_release, top_up, single_use_tokens, meter_single, meter_batch, meter_seller_claim, passkey_payee_wallet, passkey_funder, passkey_fund_relay, buyer_page, x402_knos_order, upgrade_gate, oidc_strict_json, outcome_not_code, es256_tokens, presentation_grace. Deployed on devnet: verify_gitlab, key_guardian, hold_and_bind. 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.