Memory Tools

Atlas has a persistent memory system that lives in your vault. Memory tools let Atlas save facts you share, recall things from past conversations, update what it knows about you, and read specific memory files on demand. Everything is stored as plain markdown files in atlas/memory/ — you can read and edit them directly.

remember_fact

The most natural memory tool. When you share something important, Atlas uses remember_fact to save it to the right place in your long-term memory.

Key parameters:

• fact — the information to save
• category — where it goes:

Category	What belongs here
facts	Personal details, background info, general knowledge about you
goals	Things you’re working toward, ambitions, priorities
preferences	How you like things done, style preferences, habits
relationships	People in your life, how you know them, details about them
projects	Active projects, their status, key details

Try saying:

• “Remember that I prefer concise bullet-point answers”
• “Note that my manager is Sarah Chen and she values data-driven decisions”
• “Save the fact that I’m targeting a promotion to Staff Engineer by Q3”
• “Remember that the Henderson project deadline is March 15th”
• “My goal this year is to build a consistent writing habit — save that”

Tip

You can also just share information naturally and let Atlas decide whether to save it. If you say “By the way, I just moved to Denver,” Atlas will often use remember_fact automatically to capture that for future conversations.

---

memory_search

Searches across your stored memories using the same hybrid search that powers vault search. Use this when you want to recall something specific but don’t know exactly where it’s stored.

Key parameters:

• query — what you’re looking for
• memory_type — filter by memory location:
  • long-term — search your facts, goals, preferences, relationships, projects
  • sessions — search past conversation summaries
  • all — search everything (default)
• limit — maximum number of results to return
• since — only return memories newer than this date (e.g., 2024-01-01)

Try saying:

• “What do you remember about my goals?”
• “Search my memory for anything related to the Henderson project”
• “What have I told you about my preferences for note organization?”
• “Look through my recent sessions for anything about the API integration”
• “Search my memories from the past month about my team”

Note

Atlas uses memory_search automatically when building your context in lean context mode. It searches for things relevant to what you’re asking about, rather than loading all your memories upfront.

---

update_memory

Lets Atlas (or you) make targeted updates to memory files — editing specific entries, replacing sections, or appending new information. This is more surgical than remember_fact, which only appends.

Key parameters:

• category — which memory file to update: facts or user
• new_content — the replacement or addition
• mode — how the update is applied:
  • append — adds content to the end of the file
  • replace_entry — replaces a specific entry (use old_content to identify it)
  • replace_section — replaces an entire ## Section heading block
  • append_to_section — adds content under a specific section heading
• old_content — the existing text to replace (required for replace_entry)

Try saying:

• “Update my preference for code formatting — I’ve switched from tabs to spaces”
• “I moved from San Francisco to Denver — update that in my facts”
• “Replace the Henderson project entry with the new status: completed”
• “Add a new goal: launch the beta by end of quarter”

Tip

If you notice Atlas acting on outdated information — like an old job title or a completed project — just tell it. Atlas will use update_memory to correct the record. You can also say “actually, that’s changed” and Atlas will prompt you for the update.

---

read_memory_file

Reads the full contents of a specific memory file. In lean context mode, Atlas doesn’t load all your memory into every conversation — it reads files on demand when they’re relevant.

Key parameters:

• file — which file to read:
  • persona — Atlas’s personality and communication style
  • user — your user profile (name, role, location, preferences, snapshot)
  • facts — your long-term facts (facts, goals, preferences, relationships, projects)
  • rules — behavioral rules Atlas follows
  • tools — tool usage guidance
  • schedule — your configured scheduled tasks
  • drumbeat — your ambient awareness check configuration
• section — optionally read just one ## Section within the file

Try saying:

• “Read my user profile”
• “What rules are you following?”
• “Show me my long-term facts”
• “What scheduled tasks do I have configured?”
• “Read the ## Goals section of my facts file”

Note

You don’t usually need to ask Atlas to read memory files — it does this automatically when it needs context. But if you want to see what’s stored, or ask Atlas to work with specific memory content, these phrases will trigger a read_memory_file call.

---

Where Memory Files Live

All memory is stored as plain markdown files in your vault under atlas/:

atlas/
├── identity/
│   ├── user.md          # Your profile (read via file: "user")
│   └── persona.md       # Atlas's personality (read via file: "persona")
├── rules/
│   ├── rules.md         # Behavior rules (read via file: "rules")
│   └── tools.md         # Tool guidance (read via file: "tools")
└── memory/
    ├── long-term/
    │   └── facts.md     # Your facts, goals, preferences, etc. (read via file: "facts")
    └── functions/
        ├── schedule.md  # Scheduled tasks (read via file: "schedule")
        └── drumbeat.md  # Ambient checks (read via file: "drumbeat")

You can open any of these files directly in your vault — they’re standard markdown. If you edit them by hand, Atlas will pick up the changes on the next conversation.

---

Next: Daily Notes and Task Tools