Privacy & Data

Vreko is intelligence about your code, not a copy of it. This page states plainly what stays local, what is metadata, and the exact conditions under which any metadata leaves your machine.

What stays local - always

Your code never leaves your device. Specifically, Vreko never transmits:

• File contents - the actual source in any file
• Diffs - what changed between edits
• Snapshots - local restore points are stored on disk under ~/.vreko, not synced

The intelligence layer reads your code on your machine to compute signals. The code itself is never the payload.

What is metadata

The signals Vreko computes are derived metadata, not content:

• File paths - e.g. src/auth/session.ts
• Change counts - how often a file is edited or reverted
• Timing - when sessions happen and how long they run
• Tool names - which AI tool was active (e.g. claude-code, cursor)

This metadata is what powers fragility scores, co-change patterns, and AI attribution. On the Free plan it never leaves your machine.

When metadata leaves your machine

Metadata leaves your machine only when you turn on a Pro feature that needs it - for example, the cloud dashboard. That sync is:

• Metadata only - paths, counts, timing, tool names; never file contents or diffs
• Explicit opt-in - off by default; you enable it, and you can turn it off

Plan	What runs locally	What syncs
Free	Everything - scoring, attribution, snapshots	Nothing
Pro	Everything	Metadata only, for features you opt into

How to verify

You don’t have to take our word for it:

• Inspect the local store. Vreko keeps its state under ~/.vreko. Your agents.workspace.json, snapshots, and learned patterns live there on disk.
• Watch the network. On the Free plan, the intelligence layer makes no outbound calls. Run Vreko behind a network monitor and confirm it.
• Read the file. agents.workspace.json contains only the metadata described above - open it and see exactly what Vreko knows.

If anything here doesn’t match what you observe, that’s a bug - tell us.