How UnfoldAI Works
UnfoldAI is a monitoring and observability layer for AI coding agents. It sits in your VS Code sidebar and watches what your agents do — so you can supervise confidently instead of watching every command.
The Supervision Model
Unlike AI coding assistants that do the work, UnfoldAI watches the work. It runs passively in the background, monitoring activity across all your AI agents simultaneously.
┌─────────────┐ ┌───────────┐ ┌──────────────┐
│ You │◄────│ UnfoldAI │◄────│ AI Agents │
│ (Supervise) │ │ (Monitor) │ │ (Do the work)│
└─────────────┘ └───────────┘ └──────────────┘
│
Detects issues
Surfaces risks
Builds memory
UnfoldAI does not:
- Intercept or modify agent commands
- Slow down your agents
- Require changes to how you use your agents
- Need API keys from your agents
What Gets Monitored
UnfoldAI tracks four categories of activity:
- Terminal activity — Commands executed, their output, exit codes, and timing
- File changes — What files your agents create, modify, or delete
- Error patterns — When things go wrong and how they relate to each other
- Behavioral patterns — Agent loops, drift, and non-productive sequences
Key Concepts
Episodes
An Episode is a grouped sequence of related events, typically centered around an error or issue. When UnfoldAI detects a problem, it bundles the relevant commands, file changes, and error output into a single Episode that you can investigate, chat about, or resolve.
Playbooks
A Playbook is institutional memory — a saved solution to a problem that worked before. When UnfoldAI detects an issue similar to one you've resolved in the past, it surfaces the relevant Playbook so you don't have to re-diagnose from scratch.
Proactive Alerts
Proactive Alerts are contextual notifications that appear in UnfoldAI's conversation stream when something needs your attention. Unlike noisy notification systems, proactive alerts consolidate information and offer one-click actions — analyze, restore, generate a report, or dismiss.
Pre-Commit Summary
The Pre-Commit Summary is a review checkpoint before you commit agent-written code. It shows what changed, highlights risks, surfaces dependency modifications, and generates an AI commit message — so you commit with confidence.
Restore Points
Restore Points are snapshots of your workspace state that let you revert to a known-good state if an agent makes destructive changes. They can be created manually or automatically before risky operations like dependency installs.
The Dashboard
The Dashboard is your home screen — a single-pane view of workspace health, active issues, agent activity, and recent events. It's designed to give you a complete picture without scrolling.