1. Who we are and what this covers
This policy applies to the A2A Firewall hosted service and website operated by Panchal Manan Jayeshkumar, sole proprietor, trading as "A2A Firewall" (the “Operator”, “we”, “us”). The Operator is the data fiduciary in respect of personal data processed through the hosted service.
A2A Firewall is built on privacy by design, zero-trust, and data minimisation. It is also consistent with the Digital Personal Data Protection Act, 2023 (India), the Reserve Bank of India cybersecurity framework, HIPAA, and the EU General Data Protection Regulation and California Consumer Privacy Act.
If you self-host the open-source software, you act as the data controller and this policy does not govern that deployment; instead, apply your own retention and privacy controls, which the software provides.
2. Guiding principles
- Tenant data isolation. Workspaces are isolated by UUID namespace, dedicated cryptographic keys, and database foreign-key constraints.
- Automated PII scrubbing. Personally identifiable information — including Aadhaar numbers, Indian PAN identifiers, social security numbers, payment card numbers (Luhn-validated), and contact details — is detected and masked or redacted before logs are persisted.
- Data minimisation. Inspection extracts only the features required to make a security decision (intent, tool arguments, routing metadata) rather than retaining whole conversation contexts.
- No model training. We do not use your data to train our own models. Where your content reaches an LLM provider, that provider processes it under its own terms and retention commitments rather than ours.
3. Information we collect
3.1 Information you provide
- Account data. Name, email address, plus workspace names and membership. Access is granted by verifying your email address — we do not ask you to set or transmit a password, so we do not hold a password for your account.
- Configuration data. Agent definitions, security policies, allowlists, and inspection settings you configure.
- Support correspondence. Messages you send us, including anything you choose to attach.
- LLM provider credentials. If you configure your own LLM provider (bring-your-own-key), the key is encrypted at rest using per-workspace derived keys and is decrypted only for the duration of an inspection.
3.2 Inter-agent payloads
The service inspects messages exchanged between the agents you register. This necessarily includes the content of those messages. Payload content is the most sensitive category of data the service handles, and it is treated accordingly: it is inspected in memory, excluded from logs wherever possible, and subject to the retention limits in Section 7.
3.3 Technical data collected automatically
- Request metadata: timestamps, latencies, IP address, agent identifiers, risk scores, and violation metadata.
- An append-only audit trail: Ed25519 hash-chained records of administrative and security actions, designed for non-repudiation.
- Application error traces (optional, sampled at 10%) sent to our error monitoring provider.
4. How we use information
We use personal data only to: provide and operate the Service; authenticate users and enforce plan limits; inspect and score agent traffic for security purposes; detect and prevent abuse of the Service; maintain the audit trail required for compliance; diagnose faults and improve reliability; respond to support requests; and comply with legal obligations.
We do not sell personal data. We do not use Customer Data for advertising. We do not use it to train models.
5. Payment and billing data
We never see your card details
Subscription payments are processed entirely by Razorpay, our payment processor. Full payment card numbers, CVVs, and bank credentials are collected and stored by Razorpay, never by A2A Firewall. We receive only a customer identifier, plan identifier, subscription status, and transaction references. Razorpay is an independent controller for payment data and processes it under its own privacy policy and RBI-regulated security standards.
The personal data the Operator processes in relation to billing is limited to your name, email address, plan, subscription status, and billing dates. This is retained for as long as your subscription is active and for 8 years afterwards where required by Indian tax and accounting record-keeping obligations.
6. Sub-processors and third parties
Where your agent content goes: the LLM layer
Where the LLM detection layer is enabled — which your plan makes available on Pro, Team and Enterprise, but not available on Free, and which activates only when a LLM provider key is available to the service — up to the first 300 characters of each inspected payload are transmitted to our LLM provider (Groq), or to whichever LLM provider you have configured, so that the payload can be semantically classified for prompt injection.
This is the one point at which agent content leaves your infrastructure. The excerpt is sent unredacted; the automated PII scrubbing described in Section 2 applies to persisted logs, not to this transmission. If your agent traffic must never leave your perimeter, do not enable the LLM layer, or configure your own LLM provider so that the transfer is covered by your own agreement with that provider.
In addition to the LLM provider, we rely on the following processors. Full details are also listed in Section 8 of the Terms of Service.
| Processor | Purpose | Location |
|---|---|---|
| Groq / your LLM provider | Payload excerpt classification | United States |
| Razorpay | Payments and receipts | India |
| Render | Application hosting | United States |
| Neon | Managed PostgreSQL | United States |
| Sentry (optional) | Error monitoring | European Union (Germany) |
Where a transfer of personal data outside India is necessary, it is made subject to appropriate safeguards. You may request details of the safeguards applied by contacting the Grievance Officer.
7. Data handling by deployment model
| Data category | Self-hosted (Docker / Kubernetes) | Cloud-managed (hosted Service) |
|---|---|---|
| Inter-agent payloads | Stored in your own database under your retention rules | Held in memory during inspection and discarded immediately afterwards unless you enable the review queue; a truncated excerpt may be sent to the LLM provider as described in Section 6 |
| Signatures & audit records | Hashes and Ed25519 signatures stored locally | Stored in your isolated workspace database, hash-chained for tamper evidence |
| Telemetry & metrics | OpenTelemetry spans routed directly to your own collector (for example Jaeger or Datadog) | Aggregated latency and risk metrics without prompt content |
| Customer LLM API keys | Held in local container memory | Encrypted at rest; decrypted only for the duration of an inspection |
8. Retention and deletion
Workspace administrators control retention via Dashboard → Data Retention. Defaults are set by plan:
- Payloads: 1 to 90 days (default 7).
- Telemetry records: 7 to 180 days (default 30).
- Audit logs: minimum 365 days, enforced as a compliance floor for administrative and security actions.
- Account records: retained for the life of the account plus 8 years where required for tax and accounting purposes.
On expiry, sensitive payload fields are scrubbed before permanent deletion. You may also request immediate erasure of a workspace using the purge controls, which permanently deletes associated telemetry, logs, and agent metadata.
9. Cookies and analytics
The Service sets no advertising or third-party tracking cookies. Authentication state is held in a session token in your browser’s storage; we do not use it for cross-site tracking. We do not embed third-party analytics beacons. Aggregated performance metrics are generated from server-side telemetry rather than from visitor tracking.
10. How we protect data
We use TLS for data in transit, per-workspace derived encryption keys for secrets at rest, Ed25519 agent identities, macaroon capability scoping for delegated requests, and a hash-chained audit trail. Access to production systems is restricted to the Operator. There is, however, no system that is completely secure, and absolute security cannot be guaranteed.
To report a suspected vulnerability, email mananjpanchal11@gmail.com rather than disclosing it publicly.
11. Automated inspection and decision-making
The Service makes automated security assessments about inter-agent traffic, using deterministic rules and, where enabled, a large language model. It may therefore allow, flag, or block a message based on those assessments. Those decisions are advisory security controls you configure; you remain responsible for the agents and systems you operate.
Assessment outcomes are recorded in the audit trail so that decisions affecting your traffic can be reviewed and explained. We do not make decisions producing legal effects or similarly significant effects on natural persons.
12. Your rights
Subject to applicable law, you may exercise the following rights over your personal data:
- Access — obtain a copy of the personal data we hold about you.
- Correction — have inaccurate or incomplete data rectified.
- Erasure — request deletion, where no legal obligation to retain the data applies.
- Data portability — receive your data in a structured, commonly used format.
- Withdraw consent, where processing is based on consent.
- Nomination — under the Digital Personal Data Protection Act, 2023, nominate a person to exercise these rights on your behalf in the event of death or incapacity.
- Grievance — complain to us, and thereafter to the Data Protection Board of India.
To exercise any right, email mananjpanchal11@gmail.com from the email address on your account. We may ask you to verify your identity before acting on a request.
13. Grievance Officer
In accordance with the Digital Personal Data Protection Act, 2023, the Operator appoints a Grievance Officer. Any grievance may be raised at mananjpanchal11@gmail.com with the subject line “Data Protection Grievance”. A response will be provided within 30 days of receipt. If you are not satisfied with our response, you may escalate to the Data Protection Board of India.
14. Children's privacy
The Service is not directed to children under 18 and we do not knowingly collect personal data from them. If you believe a child has provided personal data, contact mananjpanchal11@gmail.com and we will delete it.
15. Changes to this policy
We may update this policy to reflect changes in the Service or in legal requirements. Material changes will be notified by email and by a notice on this page at least 30 days before taking effect. Prior versions are available on request.
16. Contact
For privacy inquiries, data-subject requests, or regulatory audit requests, contact mananjpanchal11@gmail.com, or write to:
Panchal Manan JayeshkumarSole proprietor, trading as “A2A Firewall”
F-104, Veda Apartments
TP-13, Near Chhani Canal Road
Vadodara, Gujarat 390024
India
See also the Terms of Service and the Cancellation and Refund Policy.