Terms and conditions.
Plain-language terms for Policy Audit audits, pilots and this website. Last updated 10 October 2026.
Who we are
Policy Audit is a service provided by Manan Awasthi, a sole proprietor based in India (“we”, “us”). Our contact details are on the Contact page. By buying an audit or pilot, or using this website you agree to these terms.
The service
Policy Audit tests whether an AI agent's tool calls follow a written policy on reviewed fixture cases. Before payment, we confirm that the policy for your action tool fits the audit, agree the rules and capture requirements, and set a delivery date. You run your current agent and a changed version against simulated tools in your environment, and we deliver:
- reviewed rules and test cases, each rule shown beside the policy sentence it comes from;
- a comparison of the two agent versions, with case-level evidence for unsafe tool calls and missed eligible actions in the reviewed test suite;
- what the agent does when a required fact is missing;
- an offline report package and JUnit results for your CI.
The public sample and the free refund policy check on this website are free.
The hosted workspace supports a limited set of English refund-policy rules: windows, amount limits, confirmation, ownership, duplicate refunds and currency. Other accepted action tools use manually mapped rules, a reviewed local test bundle and agreed report delivery. They do not use the hosted refund importer. More than one action tool, or more setup help than the audit includes, needs an assisted pilot. Anything beyond an assisted pilot is quoted in writing first.
Prices
- Tool-call audit: USD 499. One action tool, up to ten reviewed rules, one baseline and one changed-version run, one retest within 30 days of payment, and up to 30 minutes of setup help by email or call.
- Assisted pilot: USD 2,500 for up to three action tools. Up to 30 reviewed rules in total, a baseline and a changed-version run and one retest within 45 days of payment for each tool, and up to 4 hours of integration help. Each further action tool, with up to ten more rules, adds USD 499. Anything else is quoted in writing before work starts; nothing is charged without your written approval.
These are one-time fees. There is no recurring subscription. The audit’s 30-day and the pilot’s 45-day retest periods begin when payment is received.
An action tool is one function or MCP tool your agent calls to take one kind of action, such as issuing a refund; a tool that can take several kinds of action counts once for each kind. A rule is one condition from your written policy, such as a refund window or an amount limit. The written scope and the invoice state the tools, rules and price before you pay.
Prices exclude taxes. Applicable taxes are added where the law requires, and you are responsible for any bank or transfer charges on your side.
Payment
We confirm the scope and delivery date in writing before sending a PayPal invoice. You pay that invoice in advance using the payment methods PayPal makes available to you. Work starts once payment is received. Assisted pilots may also be paid by an agreed bank transfer. The invoice shows the amount and currency due. Payment providers process your payment; we never see or store your full card number.
Delivery
Everything is delivered digitally, by download, by email or through the supported refund workspace. There is no physical shipping. We send the agreed setup steps, and any applicable workspace invitation, within one business day of payment. Hosted workspace access lasts 30 days. For other accepted tools, we prepare the local test bundle with you and deliver the report by the agreed date after receiving the required fixture captures. Your agent code and model calls stay in your environment; model-provider charges are paid by you.
Cancellation and refunds
A tool-call audit can be refunded in full on request within 7 days of payment, including if a report has already been delivered. An assisted pilot can be refunded in full before its kickoff session. The full terms are in our Cancellation and refund policy, which forms part of these terms.
Your responsibilities
Give us accurate policy text, have a responsible reviewer approve the rules and expected outcomes, and use synthetic test records. An engineer on your team must run the agent against simulated tools and provide the agreed capture evidence. Do not send production customer data, credentials or API keys. You remain responsible for your agent and for decisions you make using our results.
Confidentiality
We keep your policy text, test data and results confidential and use them only to deliver your audit or pilot. We will not name you as a customer or publish your results without your written permission.
Ownership
You own your policy, your data and the report we deliver to you. You may use the report internally and share it with your customers or auditors. We keep the Policy Audit software, methods and templates.
What results mean
Results show how your agent behaved on the tested cases. They do not guarantee how it will behave in production, and they are not legal advice or a certification of compliance with any law or standard. The public sample uses made-up refund cases and the recorded comparisons use fictional cases.
Liability
To the extent the law allows, our total liability for an audit or pilot is limited to the fee you paid for it, and we are not liable for indirect or consequential losses. The website, the public sample and the downloadable run data are provided as is, without warranties of any kind.
Using this website
Please do not attempt to overload, probe or disrupt the site. You may inspect and share the downloadable run data with attribution. The Policy Audit software, name and site content remain the property of their author.
Governing law
These terms are governed by the laws of India, and the courts of India have jurisdiction over any dispute.
Changes
We may update these terms; the date at the top shows the latest version. A written pilot agreement takes precedence over this page.