About Onyx Security
Onyx Security sells a secure control plane for AI agents and models. Onyx gives security, governance, and AI teams control to protect, optimize, and measure AI agent adoption across the enterprise. Onyx was founded around a single goal: to control the most advanced forms of AI. Onyx describes its secure AI control plane as a single platform that serves security, governance, and infrastructure teams as they oversee how AI agents are used, ensure they are secure and compliant, and measure their effectiveness.[1][2]
Onyx says enterprises were deploying autonomous agents faster than they could see, govern, or control them. Onyx offers one console for every AI agent and model across SaaS, cloud, endpoint, and code. Onyx discovers sanctioned and shadow AI across the enterprise, with visibility into every prompt, response, and agent action. Onyx identifies every agent, MCP server, AI tool, copilot, and embedded AI surface continuously, across SaaS, cloud, endpoints, and code. Onyx combines directory scanning, native platform APIs, and behavioral signature learning for browser-based AI to keep the inventory current. Onyx records every prompt, model response, and tool call by name, argument, and result, with the reasoning context preserved across turns. Onyx correlates agent activity to the invoking user and the agent owner, and captures the identity used on each tool and MCP call. Onyx forwards alerts and their session records to Splunk and CrowdStrike NG-SIEM.[1][2][4][5][8]
Onyx provides the runtime and posture layer of the Onyx Secure AI Control Plane. Onyx inspects every prompt, tool call, and model response inline across browser AI, coding assistants, desktop AI agents, cloud workloads, and MCP servers. Onyx applies five enforcement actions: alert, block, mask, steer, or ask (human in-the-loop). Onyx evaluates every agent on deployment and continuously across its lifecycle. Onyx red teaming continuously attacks deployed agents, generating attack plans across reconnaissance, jailbreaking, and weaponization. Onyx identifies supply chain risks in agents, MCP servers, models, and AI assets.[4][6]
Onyx gives an agent its own identity, separate from the invoker, when multiple users delegate to the same agent across tools. Onyx lets teams sanction approved tools and control MCP server access. Onyx offers pre-populated policy templates aligned to industry regulations. Onyx helps teams meet security standards from MITRE, NIST, OWASP, and more.[4][7]
Onyx centralizes all AI traffic on a fully-managed, OSS AI gateway. Onyx routes across popular LLMs to automate failover, minimize latency, and balance load by task complexity and cost constraints. Onyx proxies all MCP traffic with an inline gateway, logs every request and response, and enforces guardrails. Onyx evaluates every tool call an agent makes to an MCP server against scope, identity, data class, and policy before the call lands.[4][8]
Onyx lets teams define AI adoption targets by department, use case, or tool category. Onyx shows total AI spend, avoidable costs, and optimization opportunities. Onyx correlates AI usage, cost, risk, and outcomes to the same activity record. The Onyx AI Gateway captures per-request cost metadata at runtime: cost, tokens, model, and identity context. The Onyx Guardian Agent is a supervisory AI that continuously works across the platform to identify risks and remediate issues.[2][4][9]
Onyx offers cloud, hybrid, or self-hosted options for sensitive data residency requirements. Onyx supports deployment with AWS VPC, Bedrock Gateway, and custom proxy configurations. Onyx offers 100+ pre-built integrations that connect to agent platforms, frontier model providers, SIEM, and cloud platforms. Onyx pulls identity context from Okta and Microsoft Entra, and endpoint management context from Intune, Jamf, and Tanium. Onyx integrates with Claude Enterprise through the Compliance API, pulling conversation content, uploaded files, project data, and activity logs into the Onyx control plane. Onyx integrates with the OpenAI Compliance API, ingesting ChatGPT Enterprise workspace activity. Onyx integrates with Amazon Bedrock Agents through Bedrock Agent configuration APIs, agent invocation logs delivered to S3, and CloudTrail data events. Onyx integrates with Microsoft Copilot Studio through the Microsoft Graph API. Onyx Security is SOC 2 Type II and ISO 27001 certified.[1][4][6][7][8][10][11][12][13]
Onyx Security on the record