2. Data Security & Management
Data Storage Policy
At Macha, we prioritize data minimization and ensure that only essential information critical for the optimal functioning of our platform is stored. Our data storage policy focuses on maintaining a lean repository while upholding the efficiency and effectiveness of our services.
| Data Type | Description | Processor | Location |
|---|
 Shopify product data around descriptions, variants, SKUs | We store Shopify product data to serve as a knowledge base for answering product-related inquiries efficiently. | |  Frankfurt, Germany |
 Zendesk macros, Zendesk help center articles | We leverage Zendesk data, including macros and help center articles, to empower Macha AI in providing tailored solutions to agents and customers. | |  Frankfurt, Germany |
 Past Zendesk tickets | Zendesk ticket data is utilized to provide agents with insights into relevant tickets based on incoming queries. | |  Frankfurt, Germany |
Encryption of Sensitive Data
We prioritize the security and protection of sensitive data, even when its storage is kept to a minimum. For the sensitive data we do store, stringent encryption measures are implemented to safeguard it against unauthorized access.
- Macha employs field-level encryption for all sensitive ticket and agent data stored within our systems.
- We use AES-256-GCM encryption, a modern encryption standard that provides both confidentiality and integrity. Each field is encrypted individually, and all encryption keys are securely derived and managed using salted key derivation (scrypt).
- Communication between all systems, including browser clients, Zendesk, Supabase, MongoDB, and OpenAI, is secured via TLS 1.2+ or higher.
- Personally Identifiable Information (PII), such as agent email addresses and customer context, is redacted where possible, and encrypted at the field level when retention is necessary.
AI Data Protection & PII Anonymization
Macha's AI capabilities are powered by large language models (LLMs), primarily OpenAI, accessed through the provider's API. Because support tickets and conversations can contain personal data, Macha gives you direct control over what reaches the model provider.
Anonymizing PII before it reaches the AI
Macha includes an optional PII anonymization control (available on Pro plans). When it is enabled, structured personal identifiers are replaced with non-identifying tokens before any request leaves Macha for the LLM provider. The model reasons over the tokenized text and never receives the underlying values for these fields.
The following categories are automatically detected and tokenized:
- Email addresses
- Phone numbers
- Payment card numbers
- National identifiers (for example, U.S. Social Security Numbers and equivalents)
- IP addresses
- URLs
How it works, end to end:
- Detect and tokenize. As a ticket or message is prepared for the AI, Macha scans it for the identifier types above and replaces each value with a reversible placeholder token (for example, an email address becomes
[EMAIL_1]). The mapping between a token and its real value is held only within Macha and is never transmitted. - Send tokens only. The tokenized text, not the original PII, is what is transmitted to the LLM provider, over TLS 1.2+.
- Tools receive real values just in time. When the AI needs to act, for example, look up an order or update a record through a connected tool, the token is substituted back to its real value at the exact moment the tool executes. Functionality is fully preserved; the real value is used only inside your own connected systems, not sent to the model.
- Restore on return. When the model's response comes back, Macha re-substitutes the real values so that agents and customers see the correct, complete information.
Current scope and limitations, stated plainly. This version detects structured, well-formatted identifiers through pattern-based matching. It does not automatically detect personal names or other free-text identifiers written in prose, because those do not follow a fixed, machine-recognizable format. We document this openly so the control can be assessed accurately. PII anonymization is one layer of a defense-in-depth model, deployed alongside EU-only data residency, field-level AES-256-GCM encryption, short retention windows, and contractual protections (DPA), and is not intended to be relied upon as a sole safeguard.
Provider data handling. Macha accesses OpenAI through its API. Data submitted via the OpenAI API is not used to train OpenAI's models. In combination with the anonymization control above, the structured identifiers listed are tokenized before transmission, and the remaining content is processed by the API without being retained for model training.
Data Not Being Stored
We uphold a strict policy of data minimization, ensuring that sensitive information is neither stored nor retained within our systems unnecessarily. We are committed to protecting the privacy and confidentiality of our clients and their customers by abstaining from storing the following categories of data:
- Passwords: Macha does not store passwords in any form. We utilize secure authentication methods, such as email OTP authentication, to verify user identities without the need to retain password data.
- Client Customers' Financial Details: Financial details of clients' customers, including payment card information, banking details, and transaction histories, are not stored within our systems. We do not have access to or retain any financial data related to transactions conducted by clients' customers.
- IDs (Identification Documents): Macha does not store any form of identification documents, including government-issued IDs, passports, or driver's licenses, belonging to clients or their customers.
- Shopify Orders: We do not retain information regarding Shopify orders, including order details, customer shipping addresses, or payment information associated with orders processed through Shopify.
Data Retention
- Macha implements automated data retention policies to ensure that sensitive information is not kept longer than necessary.
- By default, AI-generated ticket content, agent responses, translations, and related metadata are retained for a period of 45 days from the time of creation.
- This is enforced using automated TTL (Time-To-Live) indexing in our databases, which securely and irreversibly deletes expired data.
- Customers on enterprise plans may request custom retention policies at the organization level (e.g., 30, 180, or 365 days).
- Data older than the configured retention window is automatically purged, and cannot be recovered.
- Onboarding and widget usage analytics are retained for a maximum of 12 months for product improvement purposes, unless deletion is requested earlier.
Data Deletion
At Macha, we prioritize the secure and responsible handling of user data, ensuring that data deletion processes are carried out efficiently and in accordance with user preferences. Our data deletion policy is structured to facilitate the timely removal of deactivated accounts while offering flexibility to our customers.
- Deactivation and Initiation: Upon account deactivation, we begin the deletion process. After 60 days, accounts enter "Pending Deletion" status.
- Permanent Deletion: 15 days later, data is permanently deleted from our systems.
- Accelerated Removal: Upon customer request, we can expedite data removal within hours.
Data Subject Rights (GDPR)
In accordance with GDPR, Macha supports full data subject rights. These include the ability for our clients (as data controllers) to request:
- Access to personal data stored by Macha
- Correction of inaccurate information
- Deletion of personal data (right to be forgotten)
- Export of personal data in portable format (JSON or CSV)
Macha has implemented internal tools and APIs to allow secure querying and management of user data by ticket ID, agent ID, or email. This enables precise, auditable actions in response to GDPR access or erasure requests.
Requests can be initiated via our support team, and are fulfilled within 30 days unless otherwise required by law.