Technical Visionaries

The Developer Portal — Infrastructure for the Next Generation of AI

"Built by developers, for developers. Deploy enterprise-grade governance in an afternoon."

SASI is a model-agnostic Python SDK designed to sit at the application layer. It provides the "Governance Airlock" your app needs without requiring you to retrain models or rewrite your existing prompt engineering.

SDK Quickstart: 2-Hour Integration

SASI is designed as a drop-in middleware solution. Most teams move from installation to a production-ready governance layer in under 4 hours.

  • Install the SDK: Simple Python-based installation into your existing environment.
  • Declare Your Mode: Select from 12 operational modes (e.g., child, patient, hr_recruiting) to automatically set thresholds and compliance rules.
  • Route Your Pipeline: Place SASI in front of your LLM call. SASI analyzes the input, enforces PII redaction and clinical boundaries, and provides a "Message for LLM" alongside a deterministic "Governance Result."
  • Handle the Metadata: Use the structured JSON response to trigger webhooks, clinical escalations, or human-in-the-loop workflows.

The Developer Toolkit

We provide the blueprints to ensure your engineering team has a rapid, reliable integration experience.

  • Real-World Integration Guide: Documentation based on the production architecture used by MyTrusted.ai and Bibbit.ai.
  • 5 Production-Ready Examples: Including a HIPAA-compliant patient mode API and a highly-regulated therapeutic companion implementation.
  • Integration Patterns Guide: Best practices for fine-tuning LLM profiles and advanced configuration strategies.
  • SASI Canary (Drift Monitor): An Enterprise SLA feature that monitors model drift and alerts you if the underlying LLM's performance begins to degrade against your governance baselines.

API & Performance Specs

  • Tiered Processing Latency: Sub-50ms (target <25ms) for Layer 1 Core Enforcement (Active Mediation).
    Cryptographic signing, chain-of-custody logging, and higher assurance tiers (Layers 2-5) introduce deterministic overhead or can be configured for asynchronous processing to protect user experience.
  • Model Compatibility: Works seamlessly across OpenAI, Anthropic, Llama, and hybrid or open-source stacks.
  • Tamper-Evident Metadata: Every response includes action_rationale, principle_triggered, and a cryptographic decision_tree_path (SASIEnvelope) for instant auditability.
  • Event Hooks & Escalation: Dedicated webhooks for Crisis, Monitoring, and Override events to trigger backend workflows or mandate immediate human-in-the-loop oversight.

Developer Resources:
[ 📄 Download the SDK Integration Guide ]