Conceptual Overview
Vigilator is the Human-In-The-Loop (HITL) layer for AI agents, providing observability and human decision-making capabilities. In a large organisation running many agents, situations regularly arise where a human needs to be involved, for instance:
- An agent needs approval before running a tool or performing an irreversible action.
- An agent is unsure how to proceed and needs a human to make the call.
- You want to watch what an agent is doing in real time and step in when needed.
Vigilator sits between your agents and your employees, capturing these moments and routing them to the right people so decisions are made quickly and safely.
Key Concepts
Interrupts
When an agent encounters a problem or needs a decision, it raises an interrupt. An interrupt pauses the agent and hands control to a human, who reviews the situation and decides the next steps before the agent continues.
Inbox
The Inbox is where interrupts land. Human operators work through the inbox, handling each pending interrupt and governing what the agent does next.
Classifications
Classifications give interrupts a broad categorisation and a priority, so operators know what kind of issue to expect and the highest-stakes work surfaces first in the queue.
Live View
Live View streams your agents' conversations into the dashboard in real time. Operators watch transcripts as they grow, and can assign, escalate, or step into a session the moment something looks off.
Organisations and Teams
An organisation contains your members, agents, and settings. Members can be grouped into teams, which are used for analytics and for routing interrupts to the right group of people.
Argus
Argus is Vigilator's built-in AI assistant. It can automatically classify incoming interrupts, summarise them into concise blurbs, and suggest responses to help operators work faster.
Where to Next
Once you're comfortable with these concepts, head to the Guides to learn the dashboard, or the Technical section to integrate Vigilator into your application.