Generic AIRG integration recipes with placeholder keys and synthetic data. Sign in for organization-specific templates, policy workspaces, logs, traces, and live API testing.
Implementation recipes for integrating AIRG into agent applications, tool-calling frameworks, and governance workflows. Each recipe has a script and a Google Colab notebook version.
pip install airg-client
export GOVERNOR_URL=https://api.airg.nov-tia.com
export GOVERNOR_API_KEY="<your AIRG account API key>"Scenario notebooks from the airg-cookbook enterprise demos directory.
Single-agent healthcare workflow for patient intake, HIPAA-aware PII handling, output scanning, verification, receipts, and audit evidence.
Trading and compliance agents operating in one AIRG tenant with shared settings, independent sessions, fingerprint baselines, traces, SURGE receipts, and cross-agent impact analysis.
Four security agents coordinate incident response while AIRG governs scope, escalation, chain detection, trace evidence, and autonomous containment.
End-to-end agentic scenarios you can run step-by-step in Google Colab. Click a badge to launch.
A support chatbot with PII protection, injection blocking, and output scanning. Covers normal flow, PII leakage prevention, prompt injection, and dangerous tool blocking.
An autonomous research agent monitored by AIRG. Auto-engages the kill switch at 40% block rate when the agent goes rogue, then demonstrates incident response and resumption.
Three agents (Planner → Executor → Reporter) collaborate in a pipeline. AIRG detects recon → pivot → exfil attack chains across agent boundaries.
A coding assistant with triple protection: pre-execution governance, output scanning for secrets/PII in generated code, and post-execution verification.
8 real-world injection attack vectors — jailbreaks, DAN, indirect injection, unicode obfuscation, nested prompts. Tests the semantic + regex firewall against each.
Agent handles API keys, DB passwords, JWTs — then attempts to exfiltrate via HTTP, curl, encoded URLs, email. Every exfil channel blocked.
6-step APT attack: read_file → list_dir → write_file → shell_exec → C2 callback → sudo. Risk climbs, chain detected, auto kill switch triggers.
Agent fires 30 rapid calls to exhaust budget and overwhelm rate limiter. Shows hard-deny at ceiling, per-tool cost tracking, and 429 responses.
Healthcare agent processes patient PHI while under injection attack. Tamper-proof evidence hashes, PII redaction, post-verification, and audit trail for regulators.
Agent restricted to [search, summarize] probes 15 unauthorized tools. Scope enforcer blocks each, deviation count climbs, escalation triggers.
A healthcare support bot attempts to email patient records containing SSNs, credit cards, and phone numbers. AIRG's PII scanner detects every entity, boosts risk scores, and prevents the data leak before it leaves the system.
An analytics agent normally reads CSVs and runs SQL queries. After a prompt injection, it pivots to reading SSH keys and exfiltrating data. AIRG's fingerprinting engine detects the behavioural drift and raises alerts before damage occurs.
Walk through a full compliance audit: verify Ed25519-signed governance receipts, validate the hash chain for tamper detection, and export a compliance bundle ready for SOC2 / ISO 27001 review.
Simulate a 3-stage attack: reconnaissance (read /etc/hosts), credential theft (read SSH keys), then data exfiltration (POST to attacker.com). Watch risk escalate across the session as AIRG detects the chain pattern and blocks the final stage.
Generate a board-ready security report: risk percentiles (p50/p90/p99), decision breakdowns with visual bars, per-agent and per-tool drill-downs, and chain pattern analysis across 30 days of governance data.