Long-Term Memory

Long-term memory is where Atlas stores everything it learns about you over time. It lives in <your vault>/atlas/memory/long-term/ as plain Markdown files. When you tell Atlas something important — a goal, a fact about yourself, a key relationship — it saves that information here so it’s available in future conversations.

facts.md

Path: atlas/memory/long-term/facts.md

This is the main memory file. It’s organized into four sections:

Facts

General information about you — your background, current situation, preferences, and anything you’ve explicitly asked Atlas to remember.

## Facts
- Works as a freelance designer based in Chicago
- Prefers dark mode in all apps
- Has a standing desk setup with two monitors
- Allergic to shellfish

Goals

Your objectives, aspirations, and things you’re working toward.

## Goals
- Launch personal portfolio site by end of Q2
- Read 20 books this year (currently at 6)
- Learn enough Spanish to travel to Mexico City in November

Relationships

People and connections that Atlas should know about — colleagues, clients, friends, family.

## Relationships
- Sarah: business partner, co-founder of the design studio
- Marcus: client at Acme Corp, main point of contact
- Dr. Chen: primary care doctor

Projects

Ongoing work and initiatives.

## Projects
- Acme Corp rebrand: due March 15, currently in wireframe phase
- Personal website: paused, needs content before development continues

How Atlas Writes to Memory

Atlas uses the remember_fact tool to add new entries to facts.md. You can trigger this explicitly:

“Remember that I prefer to be addressed as Alex, not Alexander.”

Or Atlas may save things automatically when you mention something important in conversation. When this happens, you’ll see a tool-call indicator in the chat showing what was saved.

You can also write to memory yourself — just open facts.md in your vault and edit it directly. The format is flexible; Atlas reads the entire file and understands natural language.

Tip

If you want to give Atlas a lot of context about yourself quickly, just edit facts.md directly and add whatever you’d like it to know. You don’t have to tell Atlas things one by one through chat.

Session Summaries

Path: atlas/memory/long-term/session-*.md

At the end of each conversation, Atlas writes a session summary file. This is a short record of what was discussed, any decisions made, and anything important that came up.

Session summaries serve as long-term context — even after the full conversation transcript is compacted or gone, the summary keeps the key points available for future sessions.

These files are named by date and session ID, for example: session-2026-02-24-a3f7.md

The User Snapshot

Your user snapshot is a 3–5 sentence summary of who you are, stored in the ## Snapshot section of atlas/identity/user.md. It’s automatically regenerated at the end of sessions when Atlas has learned something new about you.

The snapshot is injected into every conversation as part of your profile, so Atlas always has a current, accurate high-level picture of you even in lean context mode.

Memory Consolidation

When a conversation ends, Atlas goes through a consolidation process:

1. Extract key facts — important information from the session is checked against what’s already in facts.md
2. Update or append — new or changed facts are added; outdated information is replaced
3. Refresh the snapshot — if your situation has changed, the user snapshot is updated

This happens automatically and silently. You’ll notice that future conversations feel more informed as Atlas accumulates knowledge about you over time.

Caution

Long-term memory files are plain text and not encrypted. Don’t store sensitive information like passwords or financial account numbers in your vault.

---

Next: Sessions & History