Hybrid
Overview
Modern enterprises face a critical trade-off: the need for advanced AI capabilities without compromising on data security and regulatory compliance. At Doti, we bridge this gap through a hybrid deployment architecture that delivers powerful AI features - without customer data ever being stored outside their environment.
Our hybrid model ensures the processing engine runs in Doti's infrastructure, while all customer data remains stored within the customer's own environment. This approach balances the best of both worlds: cloud-native AI capabilities and complete data residency.
What Is a Hybrid Deployment?
Doti's hybrid deployment splits the architecture between:
Customer Environment: Where all databases and data storage live.
Doti Cloud Environment: Where AI processing and orchestration take place, without persisting any customer data.
This separation enables enterprises to retain control over sensitive data while leveraging Doti's platform for real-time insights, AI-assisted querying, and collaboration.
Data Flow Architecture
1. ETL (Ingestion) Request Flow
Doti’s ingestion process happens through a dedicated service that respects your data boundaries:
Connection Initiation: The Doti Ingestor service connects to your chosen systems (e.g., Salesforce, Jira, internal databases) using secure, preconfigured access.
Data Processing: Retrieved data is cleaned, normalized, and vectorized using a stateless embedding API.
Storage: Both the original and vectorized representations are stored entirely in your environment.
At no point does Doti persist any data in its cloud.
2. Prompt (Query) Request Flow
When users interact with Doti — through Slack, browser extension, or the web UI — here’s how the system handles the request:
Prompt Submission: The user submits a question or task.
Prompt Vectorization: The prompt is sanitized and vectorized in Doti’s cloud using a stateless embedding service.
Data Pool Construction:
Doti retrieves only the relevant, permission-granted resources from the customer environment.
User- and team-specific access controls are enforced rigorously.
Answer Generation:
The vectorized prompt and retrieved context are passed to a chat-oriented LLM endpoint.
The LLM generates a response based on the vectorized context.
Response Delivery: The answer is sent back to the user in real time.
Security & Data Privacy
Data Residency and Control
All customer data is stored and queried within the customer's environment.
No raw or processed data is cached or stored in Doti's cloud or embedding services (e.g., Azure Cognitive Services).
The hybrid architecture ensures regulatory alignment with SOC 2, GDPR, HIPAA, and other compliance frameworks.
Stateless Processing
Doti performs stateless embedding and query orchestration.
No customer prompts or records are retained post-processing.
Permissions Enforcement
Doti enforces fine-grained access control for each user, ensuring only authorized data is included in the context used to generate responses.
Unauthorized resources are excluded entirely — not just hidden.
Network Architecture
To support this hybrid model, we establish a unidirectional connection from the Doti cloud environment to the customer’s infrastructure:
Connection Target: Only the customer-hosted Database API endpoints.
Recommended Setup: Use AWS PrivateLink or equivalent to ensure isolated, encrypted, and secure network traffic.
Customer Controls Access: Customers can revoke or modify access at any time via API gateway rules or firewall policies.
Why Hybrid Matters
Many AI platforms rely on full-cloud deployments, creating risk around:
Data residency violations
Vendor lock-in for storage
Insufficient visibility into how and where data is processed
With Doti’s hybrid approach, your enterprise retains ownership, control, and visibility, while still benefiting from a state-of-the-art AI platform.
Summary
Data Storage
100% customer environment
Data Processing
Stateless, cloud-based
Security
Unidirectional private connection, no data retention
Compliance
GDPR, HIPAA, SOC 2 aligned
💡 With Doti, you don't have to choose between innovation and control — the hybrid model gives you both.
Last updated
Was this helpful?