Onyx is a security company that gives enterprises one control plane for the AI agents and models running across their business. Security, governance and infrastructure teams use it to see which agents exist, watch what they do and step in when an action looks risky.
The problem it addresses is control. As agents gain access to company data and tools, most organisations cannot say which ones are running, what they are allowed to reach or who answers for their actions.
At A Glance
| Vendor | Onyx Security |
|---|---|
| Category | Secure AI control plane for AI agents and models |
| Identities covered | AI agents, MCP servers, models and the employees who use AI tools |
| Deployment | Cloud, hybrid or self-hosted |
| Agentic AI | Agent supervision, MCP gateway, Guardian Agent, natural-language policy |
| Funding | Series B led by Bessemer Venture Partners (2026) |
| Market classification | Agentic AI Identity |
What Onyx Does
Onyx replaces a patchwork of point tools with one console that covers every agent and model across SaaS, cloud, endpoint and code. The approach is supervision rather than after-the-fact alerting: Onyx’s own models follow what an agent is doing and can approve or correct an action before it completes. Visibility, protection, policy and the AI traffic itself are handled by separate modules on the same platform, so a policy set once applies wherever the agent runs, whether that is a SaaS copilot or a coding agent.
Products
AI Observability
AI Observability is for teams that do not yet know how AI is being used in their organisation. It discovers sanctioned and shadow AI, including unapproved model deployments, and gives visibility into every prompt, response and agent action. Activity is monitored in real time across LLMs and frameworks, baselined so that anomalies stand out, and kept as a full session replay and audit trail for compliance and forensics.
AI Security
AI Security protects agents and models across their lifecycle. AI security posture management hardens configurations, supply chain checks identify risks in agents, MCP servers, models and other AI assets, and automated red teaming tests agents and models before attackers do. In production it protects prompts, responses and agent actions in real time, and sends findings to SIEM and SOAR tools.
AI Governance
AI Governance helps compliance and security teams set rules for AI use. It maps controls to standards from MITRE, NIST and OWASP and to regulatory frameworks such as the EU AI Act, lets teams sanction approved tools and control which MCP servers can be reached, and turns policies written in plain language into enforceable rules.
AI Orchestration
AI Orchestration is the infrastructure layer. A fully managed, open source AI gateway centralises AI traffic and routes it across LLMs for failover, latency and load balancing, while an inline MCP gateway proxies MCP traffic, logs every request and response and enforces guardrails.
Guardian Agent
The Guardian Agent is a supervisory AI that works continuously across the platform, identifying risks and remediating issues without waiting for an analyst. It sits alongside the four modules rather than inside one of them.
How It Works
Onyx is available in the cloud, as a hybrid deployment or self-hosted for strict data residency needs. Rollout supports AWS VPC, Bedrock Gateway and custom proxy configurations.
| Component | What it does |
|---|---|
| API connectors | Some AI platforms are connected through their own APIs with no proxy or agent. For Claude Enterprise, Onyx uses the Compliance API to pull conversation content, uploaded files, project data and activity logs. |
| Gateways | Traffic that needs inline control passes through the AI and MCP gateways described under AI Orchestration. |
| Analysis | Activity from connectors and gateways is analysed in near real time and combined with usage data from SaaS, cloud and endpoint environments to raise alerts on risky actions. |
Identities And Systems Covered
| AI agents | Agents across SaaS, cloud, endpoint and code, down to each step of their reasoning. |
|---|---|
| MCP servers | Supply chain checks, access control and gateway enforcement. |
| Models | Sanctioned and unapproved model deployments. |
| Human users | Employees using AI tools, with their prompts and responses. |
| AI platforms | Anthropic Claude Enterprise, OpenAI, Amazon Bedrock Agents, Microsoft Copilot Studio and Google Gemini Enterprise Agent Platform. |
| Cloud platforms | AWS, GCP and Azure, among more than 100 pre-built integrations. |
Company Background
| Founded | 2024. |
|---|---|
| Offices | Tel Aviv, Israel, and New York, United States, with more than 70 employees across Israel, the United States and Canada at launch. |
| Leadership | Maxim Bar Kogan, CEO and co-founder; Gil Elbaz, Chief AI Officer and co-founder. |
| Board | Includes Sarah Guo of Conviction. |
| Funding | $40M from Conviction and Cyberstarts at launch in March 2026; a $113M Series B in July 2026 led by Bessemer Venture Partners, with Cyberstarts, TCV, Conviction, FirstMark, Vintage, QuantumLight and G Squared. |
| Scale | More than 1.1M reported agents secured, 1.8M employees covered and 66.2M sessions analysed, with dozens of enterprise customers including Fortune 500 firms in banking, technology, insurance and energy. |
Note: This overview is compiled from Onyx’s publicly available information as of 3 November 2026. It is provided for information only. NHI Mgmt Group does not endorse any vendor, and nothing on this page represents the opinions of NHI Mgmt Group.