Loading...
Loading...
Governance, Guardrails & Audit Framework for AI Agents
AgentsOS is Digital Response's governed AI agent runtime. Agents execute real business workflows — order entry, quoting, payment chasing — but every action is scoped, logged, and subject to policy before it can affect a live system. Nothing an agent does is invisible or unattributable.
Every agent operates inside an explicit policy boundary. An agent cannot take an action, call a tool, or spend beyond what its policy allows — and high-impact actions always route to a human before execution.
Agents can only invoke pre-registered tools and connectors — there is no open-ended code or command execution surface.
Configurable approval thresholds (spend, customer impact, irreversibility) route an agent's proposed action to a named approver before it executes.
Any running agent or workflow can be halted instantly by an administrator, with completed actions traceable for manual reversal.
Untrusted content (emails, supplier documents, web data) is treated as data, never as instructions — input sanitisation and tool-scope isolation prevent injected instructions from expanding an agent's authority.
AgentsOS routes model calls through a governed gateway rather than binding workflows directly to a single provider.
Azure AI Foundry primary, with pluggable alternate providers per customer requirement
Customer content is never used to train or fine-tune shared foundation models
Sensitive fields are redacted before leaving the tenant boundary where the workflow allows it
Every prompt, tool call, decision, and approval is recorded and replayable
Agent memory, connector credentials, and logs never cross tenant boundaries
Anomalous agent behaviour (spend spikes, repeated failures, unusual actions) triggers alerts
For the full breakdown of AgentsOS's orchestration, guardrail, connector, and model-gateway layers, alongside available hosting models, see the platform-wideSolution Architecture.
View Solution ArchitectureHave questions about agent governance, model providers, or approval configuration? Our dedicated trust team can walk through the specifics of a deployment before you commit.
security@mydesk.digitalresponse.com.au
Available for security enquiries and incident reporting
compliance@mydesk.digitalresponse.com.au
For audit requests, model provider agreements, and regulatory questions
Last updated: July 2026 • This document is subject to change without notice. For the most current information, please contact our security team.