
Ory Agent Security is an identity and access control plane for teams running AI coding agents. It authenticates agents, checks tool permissions before execution, and records who acted on whose behalf.
Controls operate inside the agent harness, where tools execute. Each agent has its own identity and a recorded delegation chain back to a user or upstream agent. That chain remains available after credentials expire, helping teams attribute actions beyond a shared service account.
Plugins support coding workflows in Claude Code, Codex, and Gemini CLI. SDK controllers cover CrewAI, LangGraph, LangChain, and Microsoft Agent Framework. Ory Keto checks whether a session may invoke tools such as Bash, Read, or Edit; OAuth 2.1 governs MCP server access.
Explicit denials block actions where the harness supports blocking and produce a trace. A critical limitation is fail-open behavior: when Ory is unreachable, rate-limited, or unconfigured, the tool call proceeds and the error is logged. Teams should test both explicit denials and service failures before relying on enforcement.
Evaluate denied commands, sub-agent attribution, and audit exports in your own harness. For sensitive workflows, assess whether allowing actions during an Ory outage fits your security requirements. A successful identity check alone does not demonstrate that every tool action will be blocked when required.
Confirm coverage and ongoing costs for your deployment before budgeting. General Ory Network subscriptions and free open-source components do not establish the price of this named product.
+4 more
+2 more