Agentic AI
WSO2 Launches Open-Source Agent Manager to Rein In Enterprise AI Agent Sprawl
A new governance platform aims to provide centralized identity, security, and operational control for AI agents running across diverse models and frameworks.
InfoQ
WSO2 has made Agent Manager generally available—an open-source platform built to address a mounting tension in enterprise AI: organizations can now build agents using many different models and frameworks, but the infrastructure for managing their identity, permissions, behavior, and lifecycle remains scattered and fragmented across systems.
The Governance Challenge#
As teams deploy agents at scale, each built with different technologies and running in different environments, they face overlapping questions: Who gets to invoke what? What can an agent access? Can it delegate tasks? How do we revoke permissions or suspend misbehaving agents? Traditional approaches lock governance to specific AI providers, making it difficult to swap models or migrate workloads without rewriting controls.
WSO2's core insight separates agent governance from agent logic—allowing enterprises to apply consistent policies regardless of whether an agent runs on LangChain, CrewAI, Amazon Bedrock, Azure, or a custom framework.
What's New in General Availability#
Moving beyond its June 2026 beta launch, the GA release adds deeper agent identity controls, governance rules for Model Context Protocol (MCP) interactions, and a Kubernetes-native sandboxed runtime. The platform operates across cloud, on-premises, and hybrid setups, letting organizations manage agents without vendor lock-in.
Key capabilities include:
- Agent identity and authorization: centralized inventory, role-based access, delegation, token exchange, and immediate revocation
- Lifecycle controls: movement through development, staging, and production stages, plus suspension when needed
- 40+ built-in policies: personally identifiable information masking, rate limiting, and others—applied at the agent, MCP, and LLM layers
- Sandboxed execution: isolated Kubernetes-native runtime with activity monitoring and management
- Observability: OpenTelemetry tracing and rule-based or LLM-based evaluations to catch unexpected token consumption, behavioral drift, and quality issues
