System Architecture.

The definitive technical foundation of Sifta. Engineered for bank-grade isolation, high-currency throughput, and total autonomous security.

Core Infrastructure

Built on Hono (Node.js) and TypeScript — strict data typing, high stability, and resilience to errors. Powered by Better-SQLite3 in Write-Ahead Logging (WAL) mode for instant reads and writes with no latency.

Hono FrameworkTypeScript StrictSQLite3 WAL Mode

Multi-Tenant Isolation

Designed strictly around a Shared-Application, Isolated-Data pattern. Every enterprise (Tenant) operates within physically and logically segregated data domains and directory structures.

Physical Directory SegregationIsolated Audit DatabasesZero Cross-Contamination

Cryptographic Security Stack

Multi-layered Identity and Access Management (IAM). All sensitive keys and KSeF tokens are encrypted at rest using AES-256-CBC, with PBKDF2 (SHA-512) for password hashing.

AES-256-CBC Rest EncryptionPartial JWT VerificationBrute Force Protection

DLP & Anomaly Guard

Integrated Data Loss Prevention engine scans all OCR extractions and text for leaked cloud secrets (AWS, Google) or financial data (BIP39 Seed Phrases, Cards) before they touch the database.

Real-time Cloud Secret ScanningAI Anomaly DetectionSterile Whistleblowing Route

Intelligence & RAG Layer

A continuous data pipeline that transforms raw uploads into corporate knowledge. Includes a unique 'Knowledge Purgatory' staging phase before AI semantic synthesis and tokenization.

Multi-phase Data IngestionKnowledge PurgatoryContextual Semantic Synthesis

Real-Time WebSocket Protocol

A bidirectional streaming layer built on WebSockets (Node-WS) — real-time streaming of AI responses and instant alerts across the organization.

Sub-millisecond LatencyAI Thought StreamingLive Event Broadcasts

Definitive Access Model (RBAC)

Sifta operates on a strict three-tier hierarchy: Employee, Manager, and Administrator. Data visibility is enforced mathematically at the database layer. Managers can generate performance summaries exclusively for their direct reports, while AI Agents respect the caller's execution policy boundaries.

Tier 1: Employee

Task Execution, Internal RAG Access

Tier 2: Manager

Team Orchestration, Performance RAG

Tier 3: Admin

System Topography, AI Spending Approvals