Onyx Platform (Onyx AI Control Plane)
The Onyx Platform is a unified AI control plane product positioned as a single console for observability, security, governance, orchestration, and ROI tracking across enterprise AI deployments. Marketing language calls it a “Secure AI Control Plane for Enterprises.” The platform is built around a flagship “Onyx Guardian Agent” that operates across the platform’s domains.
The product fits the Guardian Agent vendor category as defined by Gartner (Feb 2026 Market Guide) — supervisory AI that monitors and governs other AI agents.
Five product surfaces
Per Onyx’s marketing site, the platform is organized into five concurrent capability surfaces, each presented as a co-equal pillar of the offering:
| Surface | What’s claimed |
|---|---|
| AI Observability | Real-time visibility into prompts, responses, agent interactions; full session replay; shadow AI detection; multi-cloud / multi-agent unified view; anomaly detection and behavioral baselining |
| AI Security | AI-SPM; supply-chain risk for agents/MCP/models/AI assets; automated red teaming; real-time prompt/response/action protection; SIEM/SOAR integration |
| AI Governance | Policy templates aligned to MITRE / NIST / OWASP / EU AI Act; natural-language policy creation; tool sanctioning + MCP server access control |
| AI Orchestration | Centralized AI traffic on a fully-managed OSS AI gateway; smart LLM routing for cost / latency / accuracy; inline MCP gateway; A/B testing; cost optimization |
| AI ROI | Adoption tracking by department/team/individual; productivity metrics; cost-benefit analysis; executive dashboards |
The combined-surface positioning is broader than any single specialist competitor. Closest single-vendor analogues: Prisma AIRS (security + posture + red-team in one), Wiz AI-SPM (posture + observability), AgentGateway (orchestration). Onyx claims to span the union of those scopes plus governance + ROI tracking.
Onyx Guardian Agent
The product’s centerpiece is the Onyx Guardian Agent — described as a “supervisory AI that continuously works across the platform to identify risks and remediate issues.” Per Onyx’s marketing claims as of 2026-05-03:
- 137,000+ agents secured across enterprise deployments
- 593,000+ employees covered across deployments
- 10M+ sessions analyzed for threats in real-time
These numbers should be treated as vendor-published claims pending independent triangulation.
Deployment
| Property | Detail |
|---|---|
| Deployment options | Cloud, hybrid, or self-hosted (advertised: AWS VPC, Bedrock Gateway, custom proxy configurations) |
| Time to deploy | ”Hours” per marketing copy |
| Integrations | ”100+ pre-built” — claims AWS, GCP, Azure, OpenAI, Anthropic, browser, AI platforms, CNAPP, SASE, EDR sources |
Role in the RA
If Onyx delivers all five surfaces as advertised, the product would touch every plane in the Agentic AI Security RA:
| Plane | Onyx claim |
|---|---|
| Identity | Discovery integration with browser, AI platforms, CNAPP, SASE, EDR sources |
| Control | Natural-language policy creation; tool sanctioning; MCP server access control (governance / posture-side, not PDP enforcement) |
| Runtime | Runtime protection — real-time prompt / response / action interception; Guardian Agent intervention on detected risks |
| Egress | Managed OSS AI gateway; inline MCP gateway; per-request logging and guardrails |
| Data | Supply-chain risk for agents / MCP / models / AI assets |
| Observability | Session replay, audit trail, anomaly detection, behavioral baselining; SIEM/SOAR integration |
This breadth is its competitive positioning and its primary skepticism vector — single-vendor coverage of all six planes is unusual; specialist tools typically dominate any individual plane. Note the precision distinction: Onyx’s strongest specific claim is runtime protection (in-line interception of prompts/responses/actions), which is much narrower than the generic “AI control plane” framing on the marketing site. The latter is positioning language; the former is the load-bearing technical capability.
Comparison with peers
| Comparison | Onyx Platform | Alternative |
|---|---|---|
| vs Prisma AIRS | Broader surface (adds orchestration + ROI); newer/smaller vendor | More mature; backed by PA portfolio integration |
| vs Wiz AI-SPM | Adds runtime protection + orchestration | Deeper graph + multi-cloud coverage |
| vs AgentGateway | Includes a managed AI gateway and an MCP gateway as one of five surfaces | OSS, Linux Foundation governance, narrower scope |
| vs Single guardian-agent products | Combines guardian-agent role with AI gateway and ROI tracking | Most guardian-agent-only products focus narrowly on supervision |
Critical assessment
Vendor marketing vs validated capability
The Onyx product page is the primary public source for capability claims; independent third-party validation (analyst write-ups, customer case studies, security research) is limited as of 2026-05-03. Treat the five-surface positioning as the vendor’s ambition rather than confirmed delivery. Buyers should validate (a) which surfaces are GA vs roadmap, (b) integration depth claims (100+ integrations is significant if accurate), (c) Guardian Agent capability vs marketing description, and (d) the customer count and session-volume claims.
Nomenclature is unsettled
Onyx’s marketing uses “AI Control Plane” as the umbrella positioning and “Guardian Agent” as the centerpiece, but neither term is being defined by Onyx — both are picked up from elsewhere. “Guardian Agent” specifically tracks Gartner’s Guardian Agent vendor category from the Feb 2026 Market Guide. The fact that the company has reached for an analyst category rather than a category of its own is itself a signal: the positioning is still being defined. The architecturally precise label for what Onyx actually does is closer to runtime protection + AI-SPM + AI gateway, layered together. Read the product capability claims (per-prompt/response/action interception, MCP gateway, behavioral baselining) as the load-bearing description; read the “AI control plane” wrapper as marketing scaffolding that may or may not survive into the eventual mature category language.
Open questions about Onyx
- Founding team and funding round details — not in the marketing source clipped to
.raw/- Specific GA dates for each of the five product surfaces — single-page marketing does not enumerate
- Pricing and licensing model — not published
- Customer references — published metrics aggregate but do not name customers
- Relationship to OSS AI gateway — the marketing claim says “fully-managed, OSS AI gateway” but does not name which OSS gateway is the upstream
CMM positioning
If the platform delivers as advertised, an Onyx-anchored deployment would target:
- D2 (Identity & Access) L3+ via discovery and policy enforcement
- D3 (Runtime Guardrails) L3+ via real-time protection
- D5 (Human Oversight Architecture) L3 via Guardian Agent
- D6 (Supply Chain) L3 via supply-chain risk coverage
- D7 (Observability & Audit) L4 via session replay + behavioral baselining
- D8 (Audit, Accountability, Forensics) L3+ via SIEM/SOAR integration + audit trail
The product is a candidate for the enterprise recommended stack as a “single-pane-of-glass” alternative to assembling Wiz AI-SPM + Prisma AIRS + AgentGateway separately. Validation of that positioning requires independent assessment beyond the marketing source.