The Pulse
Superwise Ships Sentinel Gateway for Enterprise AI Traffic
Superwise has launched Sentinel, an enterprise AI gateway that inspects and governs large-language-model traffic at the network edge. The company says Sentinel can redact secrets, personally identifiable information and toxic content before
AI.info Team ·
Superwise says its new Sentinel gateway can capture 100% of an organization’s active large-language-model traffic while keeping raw prompts and responses inside the customer’s own network perimeter. The company launched the product on September 16 as an enterprise control point for AI requests moving between employees, applications and providers such as OpenAI, Anthropic and Google Gemini.
Sentinel operates as a lightweight proxy deployed at the network edge or on individual computers. Superwise says it can redact secrets and personally identifiable information, identify toxic content and apply other guardrails before a request reaches an external model provider. The company positions the product as a response to employees and software teams adopting AI tools without a central view of what data those systems are sending.
Superwise puts the gateway between applications and models
The product’s basic operating model is simple: deploy a Sentinel gateway, route model traffic through it, then capture, inspect and log the resulting actions. Superwise says organizations can install the gateway through a Python-based command-line interface, Docker or a container cluster. Its documentation also describes a managed Helm deployment for production Kubernetes environments that need high availability and horizontal scaling.
Applications can connect without changes to their underlying AI libraries. Superwise’s documentation instructs customers to redirect standard provider base-URL settings to the Sentinel proxy, allowing compatible applications and SDKs to send requests through the gateway. The supported routes cover OpenAI chat completions, Anthropic messages and Gemini content-generation requests, along with custom OpenAI-compatible providers.
Superwise says Sentinel is available immediately and can be installed from the command line or deployed as an enterprise gateway. The company’s setup documentation lists Python 3.11 or later as a requirement for the CLI installation.
Raw prompts stay in the customer environment
Sentinel uses a split-plane design. The local gateway handles traffic interception, guardrail execution and routing to upstream models, while a cloud control plane manages gateway registration, policy configuration and operational reporting. Superwise says the cloud service receives anonymous operational metadata rather than the text of prompts, model responses or system instructions.
According to the company’s technical documentation, the metadata can include request volumes, remediation counts, triggered guardrail events and latency measurements. Sensitive content is redacted locally before the request is forwarded to the model provider. The gateway maintains an outbound connection to the control plane and does not require inbound exposure to the public internet.
That arrangement gives Sentinel a different operating boundary from a conventional centralized logging system. Superwise is asking customers to place inspection and enforcement close to the applications that generate AI traffic, while retaining a central view of policy status and activity without storing the underlying conversations in the company’s cloud.
Sentinel targets prompt injection and data leakage
Superwise says Sentinel addresses risks listed in the OWASP LLM Top 10, including prompt injection, sensitive-data leakage, jailbreaks and unsafe output. The launch announcement does not provide independent test results, measured detection rates or latency figures for those protections, so the product’s effectiveness against those threats remains a matter for customer evaluation.
The gateway’s documented processing path currently focuses on inbound prompts. It extracts and evaluates text content, applies active guardrails and forwards the request upstream. Other request parameters and non-text fields pass through, while the upstream model response returns to the application in its original structure.
“Every team we talk to is already using AI, whether IT knows about it or not. That’s the reality we built Sentinel for,” said Russ Blattner, CEO of SUPERWISE, in the launch announcement. “Organizations don’t need another policy document telling people what not to do with AI; they need a system that sees the traffic and enforces the policy in real time.”
A network control point for unsanctioned AI use
Sentinel’s immediate audience is the security or compliance team trying to govern a mix of approved and unofficial AI use. A gateway placed in front of model traffic can provide a single enforcement layer across developer tools, internal applications and customer-facing agents, rather than relying on each application team to implement separate filtering code.
The approach also creates a practical trade-off. Redirecting traffic through a proxy is less invasive than rewriting every application, but it makes the gateway part of the organization’s model-serving path. Customers will need to assess its effect on availability, latency, provider-specific features and traffic that does not use standard API routes.
Superwise is extending its existing agent-management platform with Sentinel, alongside products and capabilities covering chat, agent development, guardrails and observability. The company’s announcement provides no customer deployment figures or pricing details. For now, the concrete offering is a locally deployed gateway, a cloud control plane and a promise that the raw AI conversations remain on the customer’s side of the boundary.