Auditor OS V50-S Core • Open Reference Architecture

The In-Memory Firewall for Autonomous AI Agents

Cryptographic execution control for high-stakes AI. SentinelGate halts adversarial injections and tool hijacks at Token 0—evaluating prompt kinematics in-memory with sub-millisecond overhead, before execution, and before token spend. Every action is sealed with tamper-evident Ed25519 evidence bags engineered for ISO 20022 financial rails.

Get API Keys View Live HUD
pip install sentinelgate
Zero Retention In-memory phase evaluation; prompts evicted at socket drop.
ISO 20022 Ready Attestation bags designed for financial SplmtryData rails.
Ed25519 Signed Hardware-notarized cryptographic non-repudiation.
OpenAI Compatible Drop-in base_url replacement across Python & TypeScript.

Interactive HUD Sandbox

Run a public kinematic preview — no API key required. Emerald = permit. Coral = Phase III topological collapse.

Input Console

Realtime Telemetry HUD

Resting state — Homeostatic Equilibrium (NOMINAL)
0.042 (NOMINAL)
0.110 (STABLE)
0.018 (CLEAN)
FALSE (PERMIT)
ed25519:7a8f...resting

Inspectable .sent Evidence Bag

{
  "status": "resting",
  "note": "Awaiting audit — Run Audit or select a preset"
}

Get API Keys

Provisioned automatically via billing webhooks — keys minted as sk-sent-XXXX into the Sentinel ledger.

Developer

$19/ month

For individual builders, staging environments, and prototype validation.

  • 50,000 requests / month
  • 0-token prefill veto
  • .sent attestation hashes
  • Public notary lookups
Checkout Developer

Pro

$99/ month

For live production systems requiring high throughput and strict veto enforcement.

  • 500,000 requests / month
  • Priority routing
  • Unlimited public notary verifications
  • Email alerts on Phase III collapses
Checkout Pro

Enterprise

CustomCustom pricing

Dedicated clusters, custom kinematic threshold tuning, and financial rail SLAs.

  • Dedicated VPC or On-Premises Bare Metal deployment
  • vLLM & TensorRT-LLM kernel-level middleware
  • Linux eBPF kernel driver (XDP_DROP sub-50µs socket severing)
  • Custom HSM Ed25519 key signing & private notary
  • Custom State Uncertainty Coordinate (Ωt) threshold calibration & 99.99% SLA
Contact Enterprise

or email directly: [email protected]

One-Line Integration

Point any OpenAI-compatible client at SentinelGate — swap the base URL and keep your existing SDK.

from openai import OpenAI

client = OpenAI(
    api_key="sk-sent-XXXX",
    base_url="https://api.sentinelgate.dev/v1",
)

completion = client.chat.completions.create(
    model="meta-llama/Llama-3.3-70B-Instruct",
    messages=[{"role": "user", "content": "Execute financial database reconciliation."}],
)

Developer FAQ

Differentiation, financial rails, integration, retention, veto mechanics, and key operations.

How is SentinelGate fundamentally different from standard LLM gateways?
Standard gateways only provide telemetry, caching, and brittle regex or secondary LLM guardrails that add latency and are easily jailbroken. SentinelGate is an active cryptographic defense layer. It evaluates prompts against thermodynamic kinematic phase boundaries in-memory, severing execution at Token 0 when instability is detected, and stamps every transaction with a verifiable Ed25519 attestation proof.
Is SentinelGate compatible with financial messaging standards like ISO 20022?
Yes. Autonomous AI agents processing financial transactions, payment approvals, or automated credit actions require strict non-repudiation and tamper-evident audit trails. Every transaction passing through SentinelGate produces a canonical Ed25519-signed evidence bag (including thermodynamic coordinates, process identifiers, and cryptographic hashes) designed to map directly into ISO 20022 Supplementary Data (SplmtryData) envelopes.
How do I integrate SentinelGate with my existing code?
It requires a single line of configuration. SentinelGate is fully wire-compatible with the official OpenAI SDK. Simply update your client's base_url to https://api.sentinelgate.dev/v1 and pass your sk-sent-... API key. Prompts, schemas, streaming, and tool calls work out of the box with zero code refactoring.
Does SentinelGate store or log prompt payloads?
No. SentinelGate maintains a strict zero-retention architecture. Prompts are evaluated in-memory against kinematic boundary states, forwarded directly to GPT-4o, and evicted from memory upon socket termination. Only cryptographic attestation hashes, Ed25519 signatures, and billing counters are committed to the PostgreSQL ledger.
What is the latency overhead of the Kinematic Security Veto?
Stage I Kinematic evaluations run in-memory with sub-millisecond computation time prior to upstream dispatch. Overall proxy routing latency is typically under 15ms—negligible compared to the standard latency variance of upstream LLM inference.
What happens when an adversarial prompt triggers a Veto?
The proxy halts generation immediately at Token 0, issues an HTTP 403 kinematic_security_veto, and drops the socket before upstream tokens are consumed. The client receives an Ed25519-signed evidence bag detailing the thermodynamic state coordinates and attestation hash for audit logging.
How does key provisioning work after checkout?
Immediately upon completing Stripe checkout, you are redirected to a secure, one-time reveal page where your secret key is displayed. Once rendered or copied, the raw secret is permanently purged from our database—we store only a salted SHA-256 hash.
How do I rotate an API key or request developer support?
If you ever misplace an API key or suspect an environment leak, contact [email protected]. We execute an immediate multi-cluster credential revocation and mint a secure 24-hour one-time re-claim token.