Exogram Documentation
Welcome to the technical hub for Exogram. Here you will find the architectural blueprints, concepts, and API references required to deploy deterministic verification layers across your AI applications.
Quick Install
npm install @exogram/sdkArchitecture
Schema Integrity Engine
Prevent AI parser collapse, payload mismatch, and bad JSON structure errors.
ArchitectureBoundary Control Protocol
Stop autonomous agents from going rogue and draining session budgets.
ArchitectureThreat Prevention Layer
Block prompt injections, jailbreak overrides, and outbound PII/IP leakage.
ArchitectureMemory Integrity System
Stop memory hallucinations and prevent signed state drift in conversational context.
ArchitectureShadow Agent Mitigation
Detect and quarantine undocumented AI bots operating inside your network boundaries.
ArchitectureSemantic Caching & Edge Filtering Engine
Slash runtime API spend by 50%+ using sub-millisecond edge code filters and vector similarity caches.
Core Concepts
State Hashing
Attest context windows cryptographically to eliminate LLM execution drift.
Core ConceptsAction Admissibility
Evaluate tool calls against budgets and policies before executing dangerous updates.
Core ConceptsBlast Radius Containment
Limit the maximum operational and financial damage of a hijacked agent.
Core ConceptsDeterministic Fallback
Recover cleanly and trigger safety fallbacks when AI validation checks fail.
Core ConceptsAudit Ledgers
Immutable logging to prove AI safety and audit compliance for SOC 2.
Core ConceptsLatency Budgets
Run full AI firewall validation checks under 5ms without bottlenecking latency.
Core ConceptsAgentic Identity & RBAC
Separate human credentials from AI swarm machine keys to prevent RBAC privilege escalation.
Integration
API Reference
HTTP API reference for real-time AI security checks and structure verification.
IntegrationSDK Integration (TS/Python)
Drop-in SDK setup for real-time validation and guardrail checks in TS/Python.
IntegrationLocal Deployment (Docker)
Run the verification engine locally in air-gapped VPCs for maximum data sovereignty.
IntegrationDefining Custom Rulesets
Write Rego policies to restrict agent refund limits and unauthorized tool execution.
Runtime Integration Principle
Exogram operates adjacent to your existing LLM pipelines. You do not need to rewrite your agent execution loops: simply pass inputs and outputs through the verification boundary before final tool execution.