Privacy Policy

1. Information We Collect Across the Platform

We collect and process data in several distinct categories to operate our conversational infrastructure:

A. Account, Tenant & Operator Data

• User Profiles: Full names, business email addresses, avatars, and encrypted authentication credentials. Passwords are irreversibly hashed using industry-standard Bcrypt.
• Organization & Tenant Metadata: Tenant names, billing addresses, selected subscription tiers, and custom role permissions.
• Enterprise SSO Data: SAML/OIDC metadata, issuer certificates, and authentication domain linkages.

B. Agent Configurations & Knowledge Base Documents (RAG)

• Behavioral Guidelines & Prompts: Natural language guidelines, agent personas, and conversational action rules.
• Uploaded Knowledge Assets: PDF documentation, TXT guides, and web knowledge sources uploaded to train agents. These assets are chunked and converted into vector embeddings stored in isolated vector databases for semantic retrieval during customer interactions.

C. Omnichannel Conversation Transcripts & Multimodal Media

• Text Transcripts: Real-time and historical chat logs across connected channels (Web Chat SDK, WhatsApp Cloud API, Instagram DM, Facebook Messenger, TikTok DM, Telegram, Email, and REST APIs).
• Audio Voice Notes: Voice recordings received from customers are transcribed into text using automated Speech-to-Text (STT) models. Temporary audio files are stored in secured media buckets and transcribed transcripts are routed to the AI agent graph.
• Images & Document Attachments: Inbound images (such as receipts or product photos) are processed via computer vision models for automated entity resolution.

D. Autonomous Lead & Field Capture (CapturedData)

• Structured Entity Extraction: When configured by the tenant, AI agents automatically extract structured business fields from unstructured conversational text (e.g. customer names, phone numbers, delivery addresses, order numbers, budget specifications, and survey feedback) and index them in the tenant’s CapturedData CRM repository.

E. Technical, Network & Threat Telemetry

• IP Addresses & Geo Headers: Client IP addresses (cf-connecting-ip, x-forwarded-for) are logged in security audit tables (AuditLog, IpBan) to enforce rate limits and prevent unauthorized access.
• Honeypot Decoy Telemetry: Edge proxies log automated reconnaissance requests targeting exploit endpoints (e.g. /.env, /.git, /wp-admin) to automatically block attacking IP addresses.

2. Meta Conversions API (CAPI) & Ad Attribution Disclosure

For tenants connecting WhatsApp Business Accounts (WABA) with click-to-WhatsApp marketing campaigns:

• When an end-user initiates a chat via a Meta Click-to-WhatsApp advertisement, Flowk extracts ad referral parameters (ctwa_clid, source_id).
• To measure marketing conversion performance, Flowk normalizes and generates one-way SHA-256 cryptographic hashesof the customer’s phone number and email address.
• These pseudonymized hashes, along with transaction metadata (e.g. Lead, Contact, Purchase events), are securely transmitted server-to-server to the Meta Graph API (Facebook Conversions API)under Flowk’s authorized Tech Provider permissions.

3. Human-in-the-Loop (HITL) Real-Time Monitoring & Operator Notes

Flowk operates a collaborative Human-in-the-Loop architecture:

• Live Chat Oversight: Authorized human operators in your workspace can view ongoing AI conversations in real-time via secure Server-Sent Events (SSE) streams.
• Manual Takeover & Handover: Operators may intervene in conversations at any time (agent-takeover) to converse directly with customers and return control to the AI when resolved (agent-handover).
• Internal Session Notes: Operators can attach internal, non-customer-facing commentary (SessionNote) to conversation threads for quality assurance and collaborative resolution.

4. AI Processing & Zero Data Retention for Model Training

Flowk connects to leading Large Language Model (LLM) providers via enterprise-grade API endpoints.

5. Third-Party Sub-Processors

To deliver our services, we engage vetted third-party sub-processors under strict data processing agreements:

6. Controller vs. Processor Roles (GDPR & Data Protection)

• You as Data Controller: Your organization acts as the Data Controller for all end-user customer data processed through your deployed agents. You are responsible for maintaining your own privacy disclosures to end-users.
• Flowk as Data Processor: Kemonai Technologies Limited acts strictly as the Data Processor, executing automated instructions, vector searches, and messaging workflows on your behalf.

7. Technical Security Architecture & Encryption Controls

We deploy robust cryptographic and infrastructural controls to protect your data:

• AES-256-GCM Credential Encryption: Sensitive third-party API credentials (WhatsApp access tokens, bot tokens, OAuth tokens, and webhook secrets) are encrypted at rest using AES-256-GCM with authenticated integrity tags.
• TLS 1.3 in Transit: All web, API, and inter-service communications enforce TLS 1.3 encryption with strict HTTP Strict Transport Security (HSTS).
• Tenant Isolation: Logical multi-tenant boundaries isolate all database records, vector chunks, and conversation histories by tenant and business IDs.
• Timing-Safe Authentication: Internal machine-to-machine RPC communications utilize constant-time cryptographic comparisons to prevent timing attacks.

8. Data Retention and Deletion

You may request complete deletion of your workspace and tenant data at any time. Upon verified workspace termination:

• All associated AI configurations, vector embeddings, customer chat transcripts, audio uploads, and captured lead data are permanently purged from active databases within thirty (30) days.
• Encrypted database backups are cycled out and purged in accordance with standard 30-day disaster recovery retention schedules.

9. Data Subject Rights

In accordance with applicable privacy regulations (such as GDPR, CCPA, and NDPR), end-users and operators have the right to request access to, correction of, or permanent erasure of their personal information. Because Flowk is a data processor for customer conversations, end-users should submit privacy requests directly to the respective business deploying the agent.