Papyrus Runtime (Public)

Papyrus is a governed agentic harness built for organizations that need automation to operate reliably inside their boundary, whether that is their VPC or an IL6 environment. Rather than relying on a vendor-hosted control plane, Papyrus runs within the customer’s own Azure environment and provides the runtime, governance, state management, scheduling, approvals, connectors, and execution controls required to support long-running agent workflows. Papyrus is designed for operational use cases such as document processing, spreadsheet generation and enrichment, internal application workflows, scheduled tasks, system integrations, review and approval processes, and other back-office or mission-support activities where an agent may need to work across multiple steps, tools, and systems over an extended period of time. ### Durable Agent Execution Papyrus maintains persistent session and workflow state so agents can continue operating across reloads, interruptions, delayed inputs, scheduled events, and external signals. The runtime manages sessions, memory, schedules, workflows, and signal delivery as part of the platform rather than requiring teams to assemble those capabilities separately. ### Customer-Controlled Identity and Access Microsoft Entra ID provides the identity and authorization boundary for Papyrus deployments. Application roles can be used to control access to the runtime, administrative functions, workflows, and approval actions using the customer’s existing identity environment. ### Governed Actions and Human Approval Agents can propose actions without being given unrestricted authority to execute them. Papyrus supports deterministic workflow policies and approval requirements so sensitive actions can be held for review and released only by authorized users. This allows organizations to introduce increasingly capable automation while maintaining explicit control over what an agent is permitted to do. ### Secure Integrations Papyrus can connect agents to the systems and data sources required for operational workflows while keeping credentials under customer control. Connector configuration is designed around customer-managed secret stores, certificates, managed identities, and other secure references rather than embedding reusable credentials directly into agent prompts or workflow definitions. ### Hardened Execution On supported Linux environments, Papyrus can apply operating-system-level isolation to agent execution using sandboxing and syscall restrictions. When required isolation cannot be established, execution can fail closed rather than silently falling back to an unprotected runtime. ### Bring Your Own Model Papyrus is model-independent and can work with compatible model endpoints selected by the customer, including Azure OpenAI and OpenAI-compatible inference services. This allows organizations to select models based on their own security, performance, cost, deployment, and data-handling requirements instead of coupling the agent runtime to a single model provider. ### Designed for Controlled Environments Papyrus is intended for teams that need the productivity benefits of AI agents without surrendering control of their infrastructure, identity, workflow state, integrations, policies, or operational data. The result is a governed foundation for deploying durable AI automation across business operations and mission workflows while keeping the runtime under the customer’s administrative control.