Neuruh
← All ten products
01 · AI / System Integrity
SERVICE-FIRST · QUOTE

Agent Integrity Suite

Know, with evidence, whether the agent you are about to ship still does only what you authorized.

Who it is forAI startups, agencies, internal AI teams, and operating partners shipping agents into real workflows.
What it costs$2,500 pilot · $1,500–$5,000/mo after proofTest commercial pricing · quoted · not market-verified
TurnaroundPilot target: 7 business days from intake to receipt for one workflow. Stated as a target, not a measured average.

How buying this works

  1. 1 · REQUESTEmail your request. You get a written scope and quote before any work starts.
  2. 2 · PROVIDEOne workflow the agent runs today, with its allowed actions written down
  3. 3 · WE ANALYZECapture the baseline → Write golden, held-out, and adversarial cases → Run candidate against baseline → Report delta, regressions, authority and evidence checks
  4. 4 · RECEIVEWorkflow map and authority boundary and the rest of the deliverable by email.

There is no checkout for this product. Intake is by email and payment is collected from you by invoice once the work is scoped.

01

The problem it solves

An agent can look like it works while nobody can prove what it did, what it was allowed to do, whether it regressed since the last release, or whether the evidence it cites is real.

02

What happens

01Intake and scope one workflow
02Capture the baseline
03Write golden, held-out, and adversarial cases
04Run candidate against baseline
05Report delta, regressions, authority and evidence checks
06Issue PASS / SHADOW / MORE_EVIDENCE / REJECT with a receipt
03

What you provide

  • One workflow the agent runs today, with its allowed actions written down
  • Access to a release candidate and the previous baseline (or we capture one)
  • Five to twenty real cases you consider correct, and any known failures
  • The person who can say “promote” or “reject”
04

What you receive

  • Workflow map and authority boundary
  • Baseline snapshot
  • Test corpus (golden, held-out, adversarial)
  • Failure and regression report
  • Acceptance or release receipt
  • Remediation queue

Delivery format. Scoped by written quote first. Delivered by email as a human-readable report plus a machine-readable receipt naming inputs, sources, and unknowns. Founder-operated.

05

What an engagement looks like

A team ships a support agent allowed to issue refunds up to $50. The pilot baselines twelve real tickets, adds six held-out and five adversarial cases (a $500 request, an “ignore your limits” injection), runs the new build, and reports one regression: it issued a $75 refund. Decision: REJECT, with the failing case and the diff in the receipt.

Illustrative scenario written to show the shape of the work. It is not a customer result; no customer has bought this product yet.
  • Pre-deploy agent release gate
  • Customer acceptance testing for an AI vendor
  • Due diligence on an agent you are about to buy
  • Regulated workflow evidence packet
  • White-label assurance for an agency’s clients
06

Why you can trust it

Real today

  • Neuruh gates its own releases with these harnesses: this site’s release court records 0 broken links and 0 serious or critical accessibility violations at the shipped SHA.
  • The DeedSonar production court records 1,510 tests green at the SHA the production health endpoint attests.
  • The harness repositories are private and internal. They are not externally callable today; a pilot is run by a founder against your workflow.

Demonstration only

The browser demonstration on this page runs a small set of structural checks on text you paste. It illustrates the stages and the shape of a receipt. It is not a production assurance result and it does not touch your systems.

Release court for neuruh.com — see /proof (broken links, accessibility, dated).

DeedSonar production court — 1,510 green at e2b71fa (2026-08-23), receipt vendored in-repo.

No external customer pilot has been run yet. Public hard zeros apply (see /refusals).

Every receipt cited here is dated and vendored in the public repository or compiled on /proof. The hard zeros on /refusals apply: no revenue, no customers, no outreach as of the dated snapshot.

07

Try the idea in your browser

A small, inspectable illustration of the reasoning. It runs on numbers or text you type and proves nothing about production.

BROWSER DEMONSTRATION — NOT A PRODUCTION ASSURANCE RESULTRuns entirely in your browser on numbers or text you type. Paid work runs in the governed systems described above and returns a receipt; this panel does not.
  1. 01INTAKE
  2. 02SCOPE
  3. 03BASELINE
  4. 04GOLDEN CASES
  5. 05HELD-OUT CASES
  6. 06ADVERSARIAL CASES
  7. 07RUN
  8. 08DELTA
  9. 09REGRESSIONS
  10. 10AUTHORITY CHECK
  11. 11EVIDENCE CHECK
  12. 12DECISION
  13. 13SIGNED RECEIPT
Ready.

Load the sample pair or paste your own, then press run. Each stage lights up as the browser walks the sequence; the result shows the checks, the delta against the baseline, the decision, and an unsigned digest.

Request a 7-day assurance pilot against one real workflow. Scope and quote come back by email.

No checkout exists for this product. The contact page carries your selection into the email; scope and a written quote come back before any work starts.

Related. If you do not yet know which of your agents is canonical, start with the estate audit.

Where this product stands on the commercial ladder
ProductIMPLEMENTED · NOT LIVE
OfferIMPLEMENTED · NOT LIVE
Checkout / quoteIMPLEMENTED · NOT LIVE
IntakeIMPLEMENTED · NOT LIVE
FulfillmentINCOMPLETE
DeliveryINCOMPLETE
ReceiptINCOMPLETE
OutcomeUNKNOWN

Product → Offer → Checkout or quote → Intake → Fulfillment → Delivery → Receipt → Outcome. A rung is PROVEN LIVE only when it has been exercised on the public site and courted. Fulfillment, delivery, and receipt are founder-operated and become PROVEN LIVE only with a delivered, receipted engagement. The test-mode tape for the four priority offers is in the public repository under docs/receipts-external.

Status labels distinguish live cashiers, founder-set prices without checkout, and test pricing. No customer results, counts, or automation are claimed.