SUPERWISE

• Two of the three guardrails in the Sentinel catalog work by matching known credential and identifier formats, and all three replace what they find with a placeholder, so the cited record shows no retained asset behind them that a funded rival would have to rebuild.
• Every customer name in the record comes from the company itself, either a trusted-by row on its homepage or an interview its chief executive gave, so a buyer asking for a reference has nothing independent to check it against.
• Sentinel's published plans stop at $50 a month and the chat ladder at $69 per user, and the reviewed record states no contract value for the contact-sales path beside them, so what a large account is worth to the company is not on the record.
• The about page carries recognition claims, including a Forbes AI 50 entry and a Deloitte Fast 500 placement citing 847% revenue growth, that the company's own analyst-recognition page does not list, so a buyer relying on the analyst story has two versions of it to reconcile.
• Marketing describes automatic blocking of policy violations while the guardrail documentation describes redaction, so a buyer who purchases on the blocking claim may find the shipped controls rewrite content instead of stopping it.

The problem SUPERWISE names is that nothing sits between an organization's applications and the language models they call. Prompts carry secrets, personal data and internal content out to a third-party provider, and the company frames this as a data and compliance exposure rather than a model-quality one.

The evidence for the problem is analyst opinion the company selects. An IDC finding reported in November 2025 says that 68% of companies are investing in agentic AI while 7.9% can govern those systems at scale, and the homepage adds an IBM breach statistic and a Gartner effectiveness multiple. Those are statements about the market, not about buyers searching for this product.

The framing has moved with the market. The business the company acquired sold model observability for production machine learning, and the current pitch is runtime control over language-model traffic. Both address trust in deployed models, but the buyer and the failure they fear are different. [s1, s2, s15, s17]

Sentinel is the part of the line that constrains what an AI system does. It deploys as a proxy through Docker, a command-line installer or a container cluster, applications point at it instead of at the model, and it inspects what passes through without any security code inside the applications themselves.

The documented enforcement action is redaction, and blocking is documented one layer up. The guardrails offered in the Sentinel catalog are secret and credential detection, personal-data redaction and a rudeness filter, and the docs describe each as remediating, meaning it replaces matched content in place so the request still reaches the model. The activity log then records each interaction as passed, remediated, blocked or flagged, so a blocking outcome exists in the product's own vocabulary even though no guardrail in the current catalog is described as producing one.

The two surfaces describe enforcement in different words. A rule check in the governance console resolves to a passed or violated result, the platform page says the same guardrails stop harmful outputs such as personal data and jailbreaks before they reach users, and the guardrails page promises automatic blocking, modification or flagging of violations in real time. What the detector documentation specifies for each shipped Sentinel guardrail is a redaction placeholder.

The privacy boundary is the most specific engineering claim on record, and it belongs to the gateway. Prompts and responses passing through Sentinel stay inside the customer's environment while only counts, latency measurements and which rule fired reach the SUPERWISE cloud. The governance console is described differently, as capturing requests, responses and metadata for observability, so a buyer should ask which product holds what. [s2, s3, s4, s5, s6, s7, s8, s24, s25]

An AI gateway that redacts secrets and personal data sits in a crowded part of the market. The Forrester landscape the company cites covers 22 vendors of AI governance solutions, and two of the three checks in the Sentinel catalog work by matching known credential and identifier formats, with the third a toxicity filter. The cited record shows no retained asset behind any of them.

The company competes on placement rather than on detection. Traffic is captured at the network edge with no code change in any application, the policy lives in one place across every model and team, and raw content never leaves the customer's perimeter. That combination is a reasonable buying reason, and it is also the design a funded rival can copy.

Analyst placement is the differentiator the company leads with, and it is reported by the company. The press page counts more than 57 analyst recognitions and names a Forrester landscape of 22 vendors, an IDC assessment alongside IBM, DataRobot and Collibra, and a Gartner Cool Vendor entry from September 2020. A buyer without analyst subscriptions can verify that those reports exist but not what they say about this vendor. [s2, s6, s11, s21]

The published motion is self-serve, and a contact-sales path runs beside it. Sentinel's launch plans start free for 30 days and then cost $10, $25 or $50 a month, with the upgrade buying the right to operate rather than extra features, while the same page invites a buyer who needs Sentinel deployed for a business to get in touch. The chat product runs its own ladder up to a $69 per-user Business Plus plan.

Customer evidence is self-presented. A trusted-by row on the homepage carries Bridgestone, Firestone AG, NuevoSono, Renova Health and Home Depot. Renova Health is the single name described in any detail anywhere else, in an interview the chief executive gave to a technology outlet. Nothing in the reviewed record states revenue, seat counts or deployment scale.

Independent coverage exists but is thin and product-launch shaped. SiliconANGLE wrote up the June 2025 agent-operations launch and the November 2025 guardrails release, and the trade press covered the 2023 acquisition. The funding record is two records that do not line up: both SiliconANGLE pieces cite an aggregator figure of $4.5M, while the SEC filings under Deep Insight Solutions report 9,000,000 dollars sold in one offering and 2,000,000 dollars in another. [s13, s2, s14, s1, s19, s15, s16, s17, s22, s23]

The leadership page separates the operators from the founder whose product this was. Russ Blattner is chief executive and a co-founder, John Leschorn is chief product officer and a co-founder, and Oren Razon, who co-founded and led the acquired Superwise, holds a senior director of product title.

Razon carries the in-domain record on this team. He built a model-observability product, took it to an acquisition in January 2023, and described in a 2022 interview the gap between using AI and trusting it that the product was built to close. The reviewed record puts no value on that sale.

The chief executive's public account is about operating enterprise technology rather than about securing it. He describes the acquisition as buying a mature governance engine and an Israeli research and development team, which is a candid statement that the technical core came from outside. [s10, s17, s18, s19]

The company publishes a trust page and gates the evidence behind it. SOC 2 Type II is listed as active with a December 2024 audit and ISO 27001 with a November 2024 audit, alongside GDPR and HIPAA entries. Both the SOC 2 report and the penetration-test summary require a non-disclosure agreement, and there is no downloadable audit report on the page itself.

Two of the company's own pages disagree about how firm those controls are. The chat product page badges the offering as SOC 2 Ready, HIPAA Ready and GDPR Ready, while the trust page lists SOC 2 Type II as an active certification with a dated audit behind it. A buyer's security reviewer will have to reconcile those before signing.

The recognition claims are uneven across the site. The about page carries a Forbes AI 50 entry and a Deloitte Fast 500 placement citing 847% revenue growth, and the analyst-recognition page, which sets out the Gartner, IDC and Forrester placements in detail with document numbers, names neither. A buyer weighing the analyst story should ask which of these the company will document. [s9, s14, s10, s11]

Add analyzed competitors to compare them side by side with SUPERWISE.

The stated target and the priced target are different buyers. The marketing names healthcare, financial services, legal and manufacturing as regulated industries it is built for, while the Sentinel price list addresses individual developers at $10 and $25 a month and business use at $50 a month.

The company's own statement of who buys is smaller than the industry list. It describes the chat product as serving small, midsize and mid-market organizations, and it draws the Sentinel segment line on use rather than on size. Developer use may stay on the cheaper tiers inside a company, and the business tier applies when Sentinel protects production or automated operational workflows. That is a usage-rights boundary of the kind a self-serve product uses, not a procurement boundary.

The names the company puts forward point at large enterprises. The homepage carries a trusted-by row listing Bridgestone, Firestone AG, NuevoSono, Renova Health and Home Depot, and the reviewed record does not say what any of them bought, or how such an account is sold, sized or supported.

Placement is the capability, and the placement is specific. Sentinel is a proxy at the network edge that captures the organization's model traffic without any security code in the applications, deploys through Docker, a command-line installer or a container cluster, and forwards requests upstream with sensitive content already removed.

The detection work itself is pattern matching plus a toxicity model. The secret guardrail scans payloads for structured string patterns matching known API keys, cloud credentials and private token formats, and the personal-data guardrail matches structured identifiers, some of them from a keyword next to the identifier. Each is described as a remediating guardrail whose action replaces the match with a redaction placeholder so the request still proceeds. The governance console adds jailbreak and topic rules whose checks return passed or violated.

The privacy split is the design decision with the most consequence, and it belongs to Sentinel rather than to the company. Raw prompts and responses from the gateway never reach the SUPERWISE cloud, which receives volumetric counts, which rule fired, and latency measurements. The governance console is described differently, as capturing requests, responses and metadata for observability, and the chat product routes every conversation through SUPERWISE infrastructure for governance and audit.

Latency is claimed rather than demonstrated. The platform page states policy evaluation in under 10ms and the activity log reports policy latency separately from end-to-end time, so a buyer can measure it after deployment, and no independent measurement appears in the reviewed record.

The motion is product-led and starts free. Every tier carries 30 days free with no credit card and the same use rights as the paid tier, the gateway ships through Docker, a command-line installer or a container cluster, and the documentation is open to anyone. That lowers the cost of trying the product and says nothing about how a large account is closed.

Analyst placement is the company's chosen proof, and the company is the one reporting it. The press page counts more than 57 analyst recognitions and names a Forrester landscape covering 22 vendors, an IDC assessment placing the company alongside IBM, DataRobot and Collibra, and a Gartner Cool Vendor entry from September 2020. The Forrester report page confirms the landscape exists, covers 22 vendors and was published in April 2025 under Michele Goetz, and its vendor list sits behind a paywall.

Independent coverage tracks launches rather than adoption. SiliconANGLE reported the June 2025 agent-operations launch and the November 2025 guardrails release, and trade press covered the 2023 acquisition. Neither carries a customer, a deployment or a revenue figure.

Sentinel's published pricing is flat and small. Solo is $10 a month, Pro is $25 and Business is $50 a month, every tier carries the same core feature set, and 30 days of each are free without a credit card. The product page also invites a buyer who needs Sentinel deployed for a business to get in touch, so a contact-sales path runs beside the list. Chat is priced separately, billed per user per month with no seat minimum, and runs up to a Business Plus plan at $69 per user.

The upgrade sells permission rather than capability. The company states that upgrading buys the right to operate rather than extra features, that Business applies once Sentinel protects production or automated workflows, and that Solo includes one Sentinel deployment while Pro and Business include up to five each.

The model has a consequence the company acknowledges in its own copy. Interaction limits are described as soft during launch and pricing is described as a working model that may be refined, so the published numbers are a customer-acquisition posture rather than a settled commercial position.

Who runs the software depends on which product it is. Sentinel ships as a proxy the customer deploys into its own environment through Docker, a command-line installer or a container cluster, and the customer chooses which guardrails that gateway enforces, while the governance console is offered as a fully managed service with no infrastructure for the customer to provision.

The control plane is thin by design. Each gateway opens an outbound connection to the SUPERWISE cloud, pulls its guardrail policy on a heartbeat roughly once a minute, and reports operating metadata back. A policy change is live within about a minute, which the docs state plainly.

Operational visibility is metadata visibility. The activity log streams each interaction with its gateway, provider, model, outcome and latency, and it deliberately carries no prompt or response content, so an operator sees that a rule fired without seeing what it fired on.

The trust page lists current certifications and gates the evidence behind them. SOC 2 Type II appears as active with a December 2024 audit and ISO 27001 with a November 2024 audit, alongside GDPR and HIPAA entries, and both the SOC 2 report and the penetration-test summary require a non-disclosure agreement.

Two of the company's own pages describe the same posture differently. The chat product page badges the offering as SOC 2 Ready, HIPAA Ready and GDPR Ready, while the trust page lists SOC 2 Type II as an active certification with a dated audit behind it. A security reviewer will need the company to reconcile the two before the paperwork moves.

The recognition claims are uneven across the site. The about page carries a Forbes AI 50 entry and a Deloitte Fast 500 placement citing 847% revenue growth, and the analyst-recognition page, which sets out the Gartner, IDC and Forrester placements in detail with document numbers, names neither. A buyer weighing the analyst story should ask which of these the company will document.

The gateway is model-agnostic by construction. Its activity log records the upstream provider as OpenAI, Anthropic, Gemini or a custom provider, the console claims support for more than 50 providers through an SDK or API, and the customer keeps whichever models it already uses.

Agent frameworks are the second integration surface. The agent-operations release supports agents built with Flowise, Dify, CrewAI, Langflow and n8n, which places the company alongside open-source builder tools rather than in competition with them.

Neutrality cuts both ways. A gateway that any model can sit behind is also a gateway a customer can point at a different checker, since the applications call an endpoint rather than embed the vendor's code.

The founder with the in-domain record is not the one running the company. Russ Blattner is chief executive and a co-founder and John Leschorn is chief product officer and a co-founder, while Oren Razon, who co-founded and led the acquired Superwise, is listed as a senior director of product.

Razon carries the technical history on this team. He built a model-observability product, sold it in January 2023, and set out in a 2022 interview the gap between using AI and trusting it that the product was meant to close. The reviewed record puts no value on that sale.

The chief executive is candid about where the engineering came from. He describes the acquisition as buying a mature observability and governance engine together with an Israeli research and development team, which locates the technical core outside the founding company.