36 docs tagged with "Governance"

Default-deny access control for AI agents: declare business resources in a Resource Catalog and deny any operation that doesn't match a granted role.

Trace an AI agent from repository changes to registered runtime, governance configuration, sessions, and audit evidence.

Require human approval for risky AI actions: Build 3-step workflows with automated escalation, Slack notifications, and audit trails.

How OpenBox verdicts, approvals, and guardrails are enforced at runtime in DeepAgents.

How OpenBox verdicts, approvals, and guardrails are enforced at runtime in LangChain.

How OpenBox verdicts, approvals, and guardrails are enforced at runtime in LangGraph.

How OpenBox verdicts, approvals, and guardrails are enforced at runtime in Mastra.

How OpenBox verdicts, approvals, and guardrails are enforced at runtime in the n8n node.

Control what AI agents can do: Enforce policies, verify permissions, require approvals for high-risk operations before execution.

Define allowed AI agent behaviors: Specify permitted actions, data access patterns, API calls, and interaction boundaries per agent.

Define the exact commands an application may run in the sandbox.

Map governance activity to EU AI Act, ISO/IEC 42001, and NIST AI RMF compliance postures, backed by cryptographic proof and an immutable audit trail.

Learn how OpenBox routes governed commands through sandbox providers.

Inspect bounded sandbox execution evidence in the OpenBox session tree.

How does OpenBox governance work? Understand trust scores, runtime decisions, and the 5-stage lifecycle for controlling AI agents.

Handle OpenBox governance errors in DeepAgents: exception types, try/except patterns, and debugging guidance.

Handle OpenBox governance errors in LangChain agents: exception types, try/except patterns, and debugging guidance.

Handle OpenBox LangGraph SDK errors: governance blocks, guardrail violations, HITL rejections, and network failures.

Handle OpenBox startup failures, runtime verdicts, approval states, and guardrail validation errors in Mastra.

Handle OpenBox governance errors in n8n: node error types, Continue On Fail behavior, and debugging guidance.

Handle OpenBox errors gracefully: Manage policy violations, trust failures, and network issues with retry logic and fallback patterns.

See how governance decisions happen in real-time: Policy checks, risk assessment, and automated enforcement before agent execution.

Set hard limits for AI agents: Prevent prohibited actions, enforce safety boundaries, block violations at runtime - not after the fact.

Install and provision the default native sandbox provider on macOS or Linux.

What is OpenBox? Runtime governance platform for AI agents. Enforce policies, verify actions, ensure compliance - before execution, not after.

Provision the optional OpenShell microVM provider on macOS or Linux.

How a BLOCK verdict can carry a machine-readable fix that the SDK applies and retries automatically, instead of just failing the operation.

Build AI governance policies that actually enforce: Define rules once, apply to all agents, block violations before execution.

Connect repositories to OpenBox Projects and trace agents from code changes to governed runtimes, sessions, and governance snapshots.

Select, verify, and provision an OpenBox Sandbox provider and policy.

Provision the sandbox and run one Temporal activity inside it.

Declare the business resources your AI agents can access (APIs, databases, and queues) and grant agent roles against each one.

Route a constrained operation through an admitted sandbox command instead of the host action.

What counts as a trust incident in OpenBox: HALT decisions, critical behavioral violations, trust-tier demotions, and drift crossing into a critical band.

How OpenBox governs AI agents: The 5-stage trust lifecycle from identity verification to adaptive policies with runtime enforcement.

What are trust tiers? Learn the 4-level system from restricted to autonomous and how tier changes affect agent permissions.