What is SentinelGate?
SentinelGate is an in-memory firewall for autonomous AI agents, designed to check prompts before an agent executes actions. It evaluates prompt kinematics and can stop adversarial injections or tool hijacks before execution and token spend. Actions are accompanied by tamper-evident Ed25519 evidence bags. Developers can put it in front of an OpenAI-compatible client by changing the base URL while keeping their existing SDK, with offerings for prototyping, production, and enterprise deployments.
SentinelGate Features:
Pre-Execution Veto: SentinelGate evaluates prompt kinematics in memory and can halt adversarial injections or tool hijacks before execution and token spend.
Ed25519 Evidence Bags: It seals actions with tamper-evident Ed25519 evidence bags and provides .sent attestation hashes for recording those actions.
OpenAI-Compatible Routing: Teams can route an OpenAI-compatible client through SentinelGate by changing the base URL and continuing to use their existing SDK.
Public Sandbox: A public kinematic preview runs without an API key, alongside an input console, realtime telemetry HUD, and inspectable .sent evidence bag.
Request Tiers: The Developer offering includes 50,000 requests per month, while Pro includes 500,000, priority routing, and unlimited public notary verifications.
Enterprise Deployment: Enterprise options include dedicated VPC or on-premises deployment, vLLM and TensorRT-LLM middleware, and custom HSM Ed25519 signing.
SentinelGate Benefits:
Less Wasted Processing: Stopping a request before execution can spare downstream models from processing adversarial inputs and avoid token spend on requests that are vetoed.
Reviewable Actions: Tamper-evident evidence bags give teams a record they can inspect when reviewing what actions an AI system took.
Familiar Integration: Keeping an existing OpenAI-compatible SDK reduces the changes developers need to make when placing SentinelGate in their request path.
Early Evaluation: The public preview lets builders examine kinematic decisions before provisioning an API key or moving a prototype into a live environment.
Deployment Flexibility: Different deployment choices let organizations assess SentinelGate for prototypes, production traffic, or environments with dedicated infrastructure requirements.
Use Cases:
Prototype Validation: Individual builders can test prompt screening in a staging environment or prototype using the Developer offering and its monthly request allowance.
Agent Tool Calls: Teams building agents that invoke tools can place SentinelGate before execution to screen for adversarial injections and tool hijacks.
Production Screening: Operators of live AI systems can route requests through the Pro offering for pre-execution vetoes and priority routing.
Financial Workflows: Teams working with financial-rail systems can use cryptographic action evidence designed for ISO 20022 financial rails.
Private Infrastructure: Organizations that need dedicated infrastructure can evaluate an Enterprise deployment using a dedicated VPC or on-premises bare metal.
