UnfoldAI

Your AI agents break things. You find out too late.

UnfoldAI is the control surface for AI coding inside VS Code. It watches agent work, catches failures early, captures session checkpoints, and helps you compare, recover, and commit with confidence.

Install from Marketplace | Website | Docs | Discord | Pricing

VS Code Marketplace: Version 3.3.1 - Works with Claude Code, Codex, Copilot, Cursor, and more

AI agents move fast. That is the point. But when they touch five files, rerun a broken command, change dependencies, or overwrite a manual fix, the hard part is no longer writing code. It is figuring out what happened and where it was still safe.

Without UnfoldAI: you dig through terminal output, git diffs, Problems, and half-finished agent work.
With UnfoldAI: you get live issue cards, agent attribution, session checkpoints, checkpoint compare, recovery context, and a pre-commit review before bad work ships.

Works with Cursor, Claude Code, GitHub Copilot, Codex, Continue, Codeium, Aider, Cline, Windsurf, and custom agents.

Why UnfoldAI is different

UnfoldAI is not another coding agent. It is the supervision layer for the agents you already use.

• Catch failures early with issue cards, terminal capture, drift detection, dependency awareness, and timeline context.
• Understand what changed with session checkpoints, semantic titles, checkpoint-vs-current compare, and checkpoint-vs-checkpoint compare.
• Recover safely by inspecting restore points and exact file diffs before rolling anything back.
• Review before you commit with risk signals and commit summaries grounded in the actual session.
• See across agents in one place, so Claude, Codex, Copilot, Cursor, and human edits do not become separate blind spots.

The unique value is simple: UnfoldAI connects failures, file changes, checkpoints, terminal commands, and agent attribution into one timeline.

Failures become actionable in seconds

When something breaks, UnfoldAI turns it into an issue with the story already attached: what failed, which agent was involved, what changed, and what to do next.

Go from a captured failure to a reviewable fix plan without rebuilding context by hand.

• Ask AI opens chat with the relevant session context already loaded.
• Fix plan proposes the smallest safe fix with a reviewable patch preview before anything is applied.
• Failure review explains likely root cause, evidence, and what to verify next.
• Checkpoints lets you jump straight to the repo state around the failure.

This is the core loop: the agent fails, UnfoldAI catches it, and you move from confusion to action without rebuilding context by hand.

Checkpoints: compare before you restore

Most AI tools can revert a chat turn. UnfoldAI gives you a repo-level checkpoint timeline you can inspect before you restore.

Compare checkpoints, inspect exact file diffs, and understand what changed before restoring anything.

• Compare a checkpoint vs current to see what changed since a known-good point.
• Compare two checkpoints to understand what changed between two moments in the session.
• Open native VS Code diffs from checkpoint compare, without restoring first.
• Scan older checkpoints faster with AI-generated semantic titles. Available in Pro and Ultra.
• Restore with context after reviewing the files and activity around that point.

This is where UnfoldAI becomes more than alerting. It gives you a readable history of agent work and a safer path back when the repo starts drifting.

Review before you commit

Before you merge agent-heavy work, UnfoldAI helps you review what actually happened in the coding cycle.

Turn agent changes into a risk summary, attributed file list, and ready-to-use commit message.

• Pre-commit review surfaces meaningful changes at the right time instead of making you remember to check.
• Risk signals highlight failed commands, unresolved issues, dependency changes, and suspicious drift.
• AI commit messages are grounded in real diffs and session evidence, not guessed from filenames.
• Use in Commit pushes the generated message into VS Code Source Control in one step.

This makes agent output easier to trust, easier to verify, and harder to commit blindly.

Quiet safeguards in the background

UnfoldAI keeps watching even when you do not click anything.

• Drift detection catches agents stuck in loops.
• Dependency change awareness makes package and lockfile changes visible.
• Welcome Back summaries show what happened while you were away.
• Human overwrite protection surfaces when an agent touched files you were editing manually.
• Playbooks let you save successful fixes so repeat issues are easier to handle.
• Unfold Terminal gives cleaner command lifecycle tracking and better failure capture with Ctrl+Shift+U.

You do not need to change your workflow. The product becomes useful the moment an agent starts touching the repo.

MCP for agent-aware workflows

UnfoldAI ships with an MCP server on every plan, including Free. Your agents can query the live session before they keep editing, so one agent can check what another already broke.

Example tools include unfoldai_get_recent_issues, unfoldai_get_risk_signals, unfoldai_get_run_report, and unfoldai_search_playbooks.

Pricing

Start free. Upgrade when you want unlimited AI reviews, fix plans, semantic checkpoint titles, and auto-apply.

MCP server access is free on every tier.

Pro and Ultra include a 3-day free trial. Payment details are required. Cancel before the trial ends to avoid charge.

Get started in 30 seconds

1. Install UnfoldAI from the Marketplace and open the sidebar.
2. Let your preferred agent work, or open the Unfold Terminal with Ctrl+Shift+U.
3. Watch UnfoldAI surface failures, build session checkpoints, and track what changed.
4. Use Checkpoints to compare a clean moment against current, then use Pre-commit to review before you commit.

That is the loop: supervise the session, understand what changed, recover faster, and commit with confidence.

Security and privacy

• Session capture stays local by default.
• Backend calls are used for AI features such as chat, fix plans, reviews, commit messages, and semantic titles.
• Secrets are redacted before data leaves your machine.
• unfoldai.privacy.localOnly disables backend calls if you want pure local monitoring.
• No personal API key is required.