Governance
Security and governance, by design.
EnterpriseWeb enforces system-wide governance. It wraps all typed, immutable objects with security/identity and change management. All deployed services in inventory have lifecycle management.
All event-driven, asynchronous transactions are covered by security/identity, reliable messaging, transaction guarantees and state management with rich graph-connected logs and distributed traces. As part of dynamically contextualizing all system responses, the Semantic AI Harness enforces contracts and filters payloads based on permissions to ensure strong, systematic governance.
Governance includes multi-model, multi-agent management.
Onboarding, composing and registering — with governance throughout.
Typed, immutable objects with harness-based contracts.
Language model are onboarded as typed, immutable objects in a catalog with harness-based contracts that govern their use. The objects capture interface details and relevant properties and metrics including token costs, latency and quality that can support intelligent routing by team, workload or interaction context. Language models can be declaratively referenced in workflows and invoked during runtime orchestration. All interactions are through the platform as an intermediary; no root access is provided. EnterpriseWeb supports ModelOps in conjunction with cloud-based or on-premise model hosts.
Loosely coupled, without lock-in.
The Semantic AI Harness supports multi-model Generative AI. It works with cloud-based foundational models and open-source models deployed onsite with a library of modeled adaptors (e.g., Claude, Gemini, GPT, Cortex, Mixtral, Granite). Language Models are loosely coupled to the platform, allowing LLMs/SLMs to be readily updated or substituted without breaking existing workflows and ensuring common governance. EnterpriseWeb supports dynamic, contextual routing per use case, user or task based on declarative policies.
In EnterpriseWeb, the Knowledge Plane is the central business asset that customers extend and adapt to reflect their operations. It allows customers to use generic models out-of-the-box to realize immediate value, without lock-in and without the cost and effort of training and tuning each model.
Identity, personas and permissions.
Agents are registered as “clients” in the Gateway. They have identities, personas which scope their role, and permissions (i.e., role-based and attribute-based access control) that define their harness-based contract. Personas align with system roles (e.g., IT Admin, System Admin and Model Manager) and DevOps roles (e.g., Application Manager, Resource Manager, Service Manager, Network Manager, Data Manager). Registered agents can be assigned to workflows with hooks back to EnterpriseWeb for Context-as-a-Service. EnterpriseWeb supports AgentOps in conjunction with the agent provider or platform.
Designed and deployed in minutes.
EnterpriseWeb’s design environment also provides rapid declarative modeling of agents (e.g., LangGraph, Snowflake Cortex agents). New agents can be designed and deployed in minutes. The Semantic AI Harness provides AgentOps for agent lifecycle management. At design-time, new agents are tied to their creators for authority and accountability. At run-time, agent permissions are further reduced by the rights of the user that invoked the agent to ensuring “least privilege” security by design. As with any registered agent in EnterpriseWeb, all agent activities are logged and traced.