
Technology
|
AI Security & Identity
|
YC W26
|
Valuation:
Undisclosed

Last Updated:
March 24, 2026

Builds the identity, governance, and visibility layer for AI agents. Sits in the middle of every agent's calls, managing both agent and user identity, handling token exchange, enforcing least-privilege access, and logging everything for audit and compliance. Positioned as 'Okta for Agents.'
TypeScript and Python SDKs live. Centralized agent registry with per-agent/per-user permissioning. OAuth2/JWT token management. Real-time audit logging. SSO integration, role-based access controls, instant revocation. Demo available with OpenWebUI integration. Website has pricing page and feature descriptions.
'Okta for Agents' positioning signals intent to become default identity layer for agentic AI. Guardrails market projected $0.7B in 2024 to $109B by 2034. Nobody has agreed on what 'OAuth for agents' means yet, giving Agentic Fabriq a window to define the standard.
ML-powered anomaly detection that monitors every AI agent action in real time, flagging and blocking suspicious or out-of-policy behavior before damage occurs.
It's like a security camera that watches every AI agent in your company and sounds the alarm the instant one starts doing something it shouldn't.
It's like giving every AI agent in your company a parole officer who's read every rule book and never sleeps.
ML-assisted automatic generation and continuous refinement of least-privilege access policies for AI agents based on observed usage patterns.
It figures out exactly what permissions each AI agent actually needs by watching what it does, then locks everything else down automatically.
It's like a smart thermostat for permissions—it learns exactly how much access each agent needs and automatically dials everything else down.
ML-powered risk scoring that predicts the security and compliance impact of connecting a new AI agent to an enterprise tool before the integration goes live.
Before you plug a new AI agent into Salesforce or Slack, it tells you exactly how risky that connection is and what could go wrong.
It's like a credit score for AI agent integrations—before you approve the connection, you already know if it's trustworthy or trouble.
Paulina and Matthew met at MIT admit weekend and have been friends since. They dropped out before their second year to build Agentic Fabriq. Paulina was studying AI + Physics and doing CS/ML research at MIT Kavli Institute, MIT Haystack Observatory, and INAF Padua. Matthew dropped out of MIT at 19. They understand how autonomous agents reason, authenticate, and fail at a level that incumbents retrofitting human IAM systems fundamentally lack.