Openlayer

Openlayer operates a platform for governing AI systems across their lifecycle, from pre-deployment testing through production monitoring and policy enforcement. The product is organized in layers: discovery of AI systems running across an organization (including automatic shadow-AI detection and an org-wide inventory), a registry that records each system's owner, purpose, data sensitivity and risk tier via configurable intake and risk-assessment forms, evaluation suites that can be run pre- and post-deployment and gated in CI/CD, production observability that traces inputs, outputs and tool calls while monitoring quality, cost, latency and drift, real-time guardrails at a gateway layer, compliance mapping, and cost controls.

The testing layer ships with more than 175 out-of-the-box evaluations, regression and CI/CD testing, custom metrics and root-cause diagnostics. Guardrails cover prompt-injection and jailbreak defense, PII/PHI detection and redaction, content safety and gateway-level policy enforcement and routing. Compliance features map controls to external frameworks including the EU AI Act (Regulation (EU) 2024/1689), NIST AI RMF and ISO 42001, alongside internal policy, and produce audit-ready evidence, reporting and attestations. Cost controls track token spend by project, team and user, support hard budget caps per API key or team, provide finance-facing usage dashboards, and can automatically route high-spend users to cheaper models.

The platform supports large language models, retrieval-augmented generation, AI agents, traditional machine learning and multimodal systems, and integrates through SDKs, a gateway, a CLI and a REST API rather than requiring changes to a customer's existing AI stack.

Founding story

The founders — Gabriel Bayomi, Vikas Nair and Rishab Ramanathan — were machine learning engineers who worked together at Apple from 2018 on projects including core Siri natural language systems and Vision Pro. Because Apple's privacy-oriented structure meant no shared evaluation, observability or governance system existed, each team rebuilt the same evaluation infrastructure, and the founders repeatedly confronted the question of how to know whether a model would work across all users and real-world edge cases such as background noise or a HomePod responding from another room. They founded the company in 2021 to catch such edge cases before shipping. It launched under the name Unbox — a reference to unboxing the black box of a model — and was renamed Openlayer, referring to opening up the layers of an AI system, to avoid naming conflicts.

Business model

Openlayer sells a software platform to enterprises, with self-serve sign-up alongside an enterprise sales motion; deployment options include hosted use and on-premises installation (for example an official Helm chart for the Gateway on Kubernetes and support for external/managed ClickHouse).

Enterprise software platform sold to AI, engineering, security, compliance and finance teams; the company also publishes a free public LLM cost-comparison tool at llm-costs.openlayer.com.

Traction

The company states its platform is used by Fortune 500 AI teams and has named Amdocs and Telefónica as customers evaluating their AI models with Openlayer; Telefónica and KPN also participated as investors in the Series A. It holds SOC 2 Type II certification and states GDPR compliance. The about page lists roughly two dozen employees across engineering, design, sales, solutions and operations.

Latest developments

An August 10, 2026 changelog entry introduced AI-generated summaries for sessions, traces and test results with drill-down to underlying rows, embedding-based semantic search filters, and a remote OAuth-protected MCP connector usable in Claude Code and claude.ai. The same release window added Openlayer Agent Skills for AI coding assistants, governance and lifecycle tools over MCP (expanding from 44 to 53 tools), workspace-level cost and health dashboards, webhooks for test results, semantic model-based routing, a guided `openlayer init` CLI command and non-interactive CLI mode, a native IBM watsonx Orchestrate integration, Gemini/Vertex and Claude Agent SDK tracing, per-project data retention, a content-safety guardrail and multimodal gateway endpoints, US National Provider Identifier PII detection, a free public LLM cost explorer at llm-costs.openlayer.com, and on-prem additions including a Gateway Helm chart and a deployment admin area.