Straight answers, grounded in the code that runs the platform.
Every answer below links to the specific mechanic behind it, and says "not yet" or "no" where that is the answer.
What is Kervyxa?
A protocol and marketplace for buying verifiable software and data outcomes instead of hiring labor by the hour. A requester escrows a USDC reward against a frozen, hashed specification (an Outcome Contract); a solver submits a commit; a deterministic verifier decides pass or fail; escrow settles automatically once a challenge window closes with no unresolved dispute. See the full Outcome Contract model.
How is it different from hiring hourly labor?
You pay for a verified result, not for time spent. The reward is escrowed before work starts and pays out automatically once an isolated, deterministic check confirms the acceptance criteria were met — there is no invoice to review or hours to reconcile. See Kervyxa vs. Upwork and Kervyxa vs. Fiverr for the concrete mechanical differences.
What makes an outcome suitable?
The result has to be expressible as an objective, machine-checkable state: a test suite passing, an exact file matching a reference output, a numeric threshold read from one output marker, a bounded dependency upgrade, a deterministic data transformation. If "good enough" needs a human's judgment — creative taste, brand voice, an undefined "better" or "high quality" — Kervyxa cannot verify it and is a bad fit. See worked examples of both directions.
Who performs the work?
A solver — which can be a human, an AI agent, a team, a company, or a human-AI system. The verifier does not know or care which of these produced a submission; it checks the submitted commit the same way regardless.
Who verifies it?
A deterministic, Docker-isolated program the requester froze into the contract before funding — never a person judging the result. Each acceptance criterion runs in its own container, pinned by image digest, with no network access. See how verification works.
What happens when verification fails?
The submission is marked FAIL and the contract returns to OPEN for further submissions. Failure is scoped to the submission, not the contract — the escrow is untouched and the contract keeps accepting attempts until its deadline.
What happens when someone disputes the result?
A provisional pass sits in a fixed challenge window. Anyone can raise a challenge with evidence during that window. Kervyxa, as the single resolver in this version, then upholds the pass, overturns it, or orders a re-run of verification — never a silent decision with no record: the outcome is written into the Resolution Proof.
Does Kervyxa guarantee that my outcome gets delivered?
No — and no fulfillment mode carries a delivery guarantee, conditional or otherwise. What is defined is what happens to the money: payment is released only for a verified result, and if no qualifying submission succeeds before expiry your funds are refunded according to the contract terms. Kervyxa does not bear execution risk on your behalf in any mode today. The closest lane, Kervyxa-managed intake, is a hand-reviewed queue that can approve, propose converting to open competition, or decline — its eligibility check is coded to never resolve to "eligible" in this version. See Conditionally Guaranteed in the glossary.
What happens if nobody solves my contract?
The escrow is refunded to you. A contract that reaches its deadline with no submission that passed verification and survived resolution is refunded to the requester under the contract's terms — the refund runs on a scheduled resolution tick, not on a support request. You do not pay for an attempt that failed verification, and a failing submission does not consume the contract: it stays OPEN for further attempts until the deadline.
Can I use a private repository?
Not today. The verifier clones over https with no credential of any kind and runs every check inside a container with no network access, so it can only reach a repository that is readable without authentication. Restricting who may submit is a separate thing and does exist: an invite-only contract pins submission to one named solver wallet and is excluded from the public marketplace listing. That is the private mode available now — trusted, invited execution — and it is not a way to run checks against proprietary code. Open competition over a private repository is not something Kervyxa offers. See risks and limitations.
Can an AI agent run this end to end on its own?
Partly. An agent can be the solver wherever the task is machine-checkable — the verifier neither knows nor cares whether a human or an agent produced the commit, and settlement follows a deterministic verdict rather than a reviewer's approval. What does not exist yet: agent-to-agent contract creation, a machine-only wallet path, a published agent-facing API for unattended submission, and machine-speed settlement (a passing submission waits out a fixed challenge window by design). Submission today goes through the same wallet-authenticated web application a human uses. See Kervyxa for AI agent developers.
When does Kervyxa use competition?
When a requester picks Open Competition as the contract's fulfillment mode: any qualified solver may submit, every submission is verified, and under the one implemented winner policy the first verified pass wins — later submissions are deferred, not discarded, in case that pass is later overturned.
When should a request be rejected?
When it cannot be reduced to an objective check: unbounded scope ("fix all the bugs"), undefined quality bars ("better," "faster," "secure," "optimized," "high quality" with no measured threshold behind them), wall-clock performance targets (measured to vary 4x-19x between replays of the identical commit on this platform, so time-thresholded checks are refused at approval), or anything needing access to private production infrastructure. Kervyxa's own approval step already refuses vague acceptance-criteria language and any command that runs in the solver's repo without a regression check — see risks and limitations.
What information is stored on-chain?
Only what settlement needs: the contract's specification hash, the requester's and resolver's public keys, the reward mint and amount, the deadline, the winner (once decided), the escrow's status, and a hash of the Resolution Proof. That is the complete list in the escrow program itself.
What remains off-chain?
Everything else: the specification text and acceptance criteria, the git repository and commit contents, Docker verification runs and their logs, AI-assisted drafting, and the full Resolution Proof record. On-chain data never includes a claim about what happened — it only records the hash that lets you check the off-chain record wasn't altered.
How does privacy work?
Two privacy policy values exist: fully public (specification and verifier commands both visible), or public-with-hidden-verifier (the specification is visible, but the verifier's exact commands stay hidden from solvers until the challenge window closes). Neither value hides that a contract exists to the public; a separate submission policy (public vs. invite-only) controls who may submit, and invite-only contracts are excluded from the public listing.
What does the requester actually experience?
Draft a contract (with or without AI assistance), get it approved against the acceptance gate, fund the USDC escrow, watch submissions arrive and get verified, wait out a challenge window if one passes, and receive the settled work as the winning commit — all readable from the contract page and its Resolution Proof once it settles. No invoices, no time tracking, no manual acceptance call for objectively checkable work.