Kavia AI — Security & Architecture Overview
Purpose: Give engineering leaders and procurement teams confidence that Kavia can be safely evaluated, integrated, and scaled inside enterprise workflows.
1. Security by Design
- Data Privacy: Kavia does not train foundation models on customer code or data. All inputs and outputs remain tenant-scoped.
- Encryption:
- Data in transit: TLS 1.2+
- Data at rest: AES-256
- Access Control:
- SSO/SAML 2.0 support
- Role-based access controls (RBAC)
- Team/group level access control
- Granular tenant isolation
- Auditability: Full activity logs and traceable run history available for admin export.
- Compliance Alignment: SOC 2 Type II in process, GDPR and CCPA aligned practices, export-control awareness for defense/aerospace customers.
2. Flexible Deployment Options
- SaaS Cloud (default): Multi-tenant, US-hosted, with strict tenant isolation. Source code and knowledge graphs remain in customer SCM systems.
- Dedicated VPC (optional): Single-tenant deployment in customer’s AWS/Azure/GCP account.
- On-Premise (roadmap): Containerized deployment for air-gapped or regulated environments.
3. Architecture at a Glance
- Core Orchestrator: Governs Inspect → Plan → Build workflows and manages context exchange.
- Knowledge Graph (EKG): Indexes customer codebases, test suites, and documents for context-aware generation.
- Model Abstraction Layer: Pluggable access to LLMs (OpenAI, Anthropic, Azure OpenAI, Google Gemini, AWS Bedrock) — customer selects provider(s).
- Framework Support: C/C++, Python, Java, Android, and 60 popular languages and framework with connectors for CI/CD pipelines and Figma/spec ingestion.
- APIs & Integrations: REST/GraphQL APIs, GitHub/GitLab/Gerrit connectors, CI/CD (Jenkins, GitHub Actions), Jira integration.
4. Customer Control
- Customer owns and controls their data and code.
- Kavia retains no rights to foreground IP created during use.
- Customers can export data, logs, and artifacts at any time.
Contact: security@kavia.ai | www.kavia.ai