Privacy Policy: cortex-viz

Last updated: 2026-07-22

cortex-viz is a local visualization server for Cortex and Claude Code session data. This policy states exactly what it reads, what it serves, and what leaves your machine.

What the server processes

• Your local Cortex database (PostgreSQL or SQLite), when present. Your memories, entities, and relationships are read and never written. cortex-viz does write five tables of its own into that same database: workflow_graph_snapshot, workflow_graph_snapshot_scoped, workflow_graph_layout, workflow_graph_layout_lod (derived layout and graph caches, so a view opens without re-running a long build), and session_activity (a record of tool calls, file accesses, and skill invocations, which is what lets the graph stream live). All of it stays in your local database. Nothing is transmitted.
• Claude Code artifacts under ~/.claude/: session transcripts (projects/**/*.jsonl), wiki pages, profiles, and local git state.
• In --no-db mode, only the ~/.claude/ artifacts and git.

The UI is served on 127.0.0.1 only, with host-header and same-origin guards; it is not reachable from your network.

What leaves your machine

• Your data: nothing. No memory content, session content, code, or metadata is transmitted to the author, to Anthropic, or to any analytics service. There is no telemetry.
• One disclosed exception, a CDN asset fetch: the 3D brain view loads the three.js library (and related loaders) from the public unpkg CDN at page load. This transfers a standard library request to the CDN (your IP and a static file path, no user content). The other views use only vendored, locally-served assets and work fully offline.

Your controls

• The activity-capture hooks no-op when no visualization instance is open.
• Close the server (idle watchdog also stops it) to end all processing.
• Avoid the brain view to avoid the CDN fetch entirely; all other views are offline-safe.

Contact

admin@ai-architect.tools · https://github.com/cdeust/cortex-viz/issues