AgentWing

Enterprise AI compliance

Gate agent actions against internal security policies before they access sensitive files or systems.

Multi-agent workflows

Route actions to sandboxed environments when risk is high, preventing unintended side effects.

Human-in-the-loop approvals

Require manual approval for high-risk actions before execution proceeds.

Audit and forensics

Capture detailed receipts and logs for every agent action to meet regulatory or internal audit requirements.

Restore point creation

Automatically trigger restore points before actions that could modify critical state, enabling rollback.

Policy enforcement for LLM tools

Block actions that violate predefined policies (e.g., secret file access) in real time.

Pre-execution action checking

Every AI agent action is evaluated against policies before it runs, using the `POST /api/v1/check-action` endpoint.

Decision outcomes

Returns clear decisions: `allow`, `block`, `sandbox_required`, `approval_required`, or `restore_point_required`.

Risk classification

Each action is tagged with a risk level (e.g., `high`) to inform the appropriate governance response.

Policy identification

Specific policy names (e.g., `block-secret-file-access`) are attached to decisions for precise rule tracking.

Audit receipts

Every check generates a unique `receiptId` (e.g., `aw_receipt_...`) for full traceability and compliance documentation.

Sandbox routing

Automatically redirects risky actions to isolated sandbox environments when required.

Approval workflows

Flags actions needing human approval before execution, enabling safe delegation.

Restore point triggers

Creates restore points automatically for actions that may alter system state, supporting rollback.