# Trace Memory Architecture

Trace is TronVibes' consent-based relationship memory layer.

The original Trace materials found on this computer describe an Advanced Memory Management pattern: never blindly delete memory; compress, summarize, encode, and retrieve. TronVibes adapts that idea for profiles so returning users can feel remembered without exposing secrets or flooding AI prompts with every old message.

## Core Pattern

- **Episodic memory:** Safe user moments are stored as a relationship thread.
- **Compressed memory:** Older context is summarized into a smaller, careful memory.
- **Recursive continuity:** The existing summary and new safe moment are compressed together, so the relationship can mature over time.
- **Future retrieval:** Later releases can retrieve relevant old memories by meaning instead of loading everything.

## What Trace May Remember

- Preferred tone and language.
- Encouragement that helps the user feel steady.
- Personal goals the user chooses to share.
- Boundaries the user wants respected.
- Safe context for support, and future reflections.

## What Trace Must Never Remember

- Seed phrases or recovery phrases.
- Private keys, passwords, or two-factor codes.
- Full card numbers, bank details, or government IDs.
- Stripe, B.AI, TronGrid, hosting, or server secrets.
- Anything the user did not consent to store.

## Safety Rule

Trace exists for continuity and care, not surveillance, manipulation, profiling, or pressure. The memory should produce truth, courage, peace, mercy, self-control, and practical next steps.

The current launch implementation stores a compact profile memory summary, safe key facts, and a capped set of sanitized episodic notes. The deeper AMM roadmap can later add separated long-term summary files, semantic retrieval, richer user export/delete controls, and stronger encryption around sensitive profile data.

The launch profile controls already include basic export and clear actions for the current compressed Trace Memory. Export lets the user inspect what is remembered, including safe episodic notes. Clear resets the summary, safe facts, and episodic notes for that profile.

Trace notes pass through a redaction layer before storage. If a user accidentally types language that appears to include a seed phrase, private key, password, card-like number, or API secret, the app attempts to remove it before memory is saved. This is a safety net, not permission to share secrets.