Memory

Memory is the fifth pillar of Lytos: Memory.

AI agents are stateless — they forget everything between sessions. Memory gives them persistent knowledge about your project, stored in files you own.

Structure

memory/
├── MEMORY.md              # Index — read every session
└── cortex/                # Specialized zones — loaded on demand
    ├── architecture.md
    ├── patterns.md
    ├── bugs.md
    ├── sprints.md
    ├── frontend.md
    ├── backend.md
    └── business.md

How it works

MEMORY.md — The index

Read at every session. Contains a table of contents pointing to cortex files, plus a short living summary of the project’s current state.

The agent reads the index, identifies which cortex files are relevant to the current task, and loads only those.

Cortex files — Specialized knowledge

Each cortex file covers a domain:

File	Content	Load when…
architecture.md	Tech choices, structure decisions	Any structural task
patterns.md	Code patterns that work well	Code review, new features
bugs.md	Recurring problems and solutions	Debugging, fixes
sprints.md	Sprint history and retrospectives	Planning
frontend.md	UI/UX decisions, component patterns	Frontend tasks
backend.md	API decisions, data model choices	Backend tasks
business.md	Domain knowledge, business rules	Any domain-related task

What to store in memory

• Architecture decisions and their rationale
• Patterns that work well in this project
• Recurring bugs and their root causes
• Domain-specific knowledge the AI needs

What NOT to store

• Code snippets (they live in the code)
• Git history (use git log)
• Temporary task details (use the issue board)

Why it matters

“Quality doesn’t come from the prompt. It comes from context.”

Without memory, the AI makes the same mistakes twice. It suggests patterns you’ve already rejected. It doesn’t know that you tried Redis and switched to PostgreSQL for good reasons.

Memory is what makes Lytos sovereign — your project knowledge lives in files you own, not in a vendor’s conversation history.

Learn more

• How to organize Claude Code memory across sessions — a practical walkthrough of the memory/ folder, with concrete file examples and how Claude Code, Cursor, and Codex pick it up at session start.