Dobby
Human-in-the-Loop

Governance & Control

Agents stop and wait for human approval at configurable gates. Every action is policy-checked, every decision is auditable. You stay in control.

Approval Gates

Configure which agent actions require human approval — code changes, deployments, security-sensitive operations.

Policy Engine

Define rules: max files per PR, allowed file paths, cost limits, model restrictions. Agents cannot exceed boundaries.

Kill Switch

Instantly halt all agent activity at org or tenant level. Three scopes: all traffic, LLM only, new keys only.

Immutable Audit Trail

Every action, decision, approval, and rejection is logged with timestamp, user, and context. 365-day retention.

Key Capabilities

Configurable approval gates per agent and action type
Auto-approve conditions (max files, max lines, allowed paths)
Kill switch with 3 scopes (all, LLM-only, new-keys)
4-layer policy hierarchy (Platform → Org → Tenant → User)
Real-time approval notifications via MCP tools
SOC 2 compliant audit trail with 365-day retention

How It Works

1

Agent proposes

An agent completes its work and proposes an action.

2

Policy validates

The policy engine checks against org and tenant rules.

3

Human reviews

If approval is required, the action waits for human review.

4

Audit records

Every decision is permanently recorded in the audit trail.

Ready to get started?

Connect and manage AI agents with governance & control built in.