How Agent Tools Work

When you chat with Atlas, you’re not just talking to a language model — you’re working with an agent that can take real actions. Those actions are called tools. Atlas has 39+ tools it can use during a conversation to read your notes, create files, manage tasks, search your vault, save memories, and more.

What Tools Are

A tool is an action the AI can perform on your behalf. Instead of just telling you how to create a note, Atlas can actually create it. Instead of describing what your upcoming meetings look like, it can check your calendar and tell you exactly what’s scheduled.

Tools are the difference between a chatbot and a true personal assistant.

Tools Run Locally on Your Machine

This is an important point: all tools execute locally. When Atlas reads a note, searches your vault, saves a memory, or creates a file, that happens on your computer — not in the cloud.

The AI model (running on the Atlas cloud) decides which tools to call and what to pass to them. The tool itself runs in the desktop app, on your machine, against your local vault. Your files never leave your computer as part of tool execution.

How Atlas Decides When to Use a Tool

You don’t have to ask Atlas to use tools explicitly. It figures out when a tool is appropriate based on what you ask. A few examples:

“Create a note about today’s standup” — Atlas uses create_note
“What tasks do I have this week?” — Atlas uses get_tasks with date_range: week
“Remember that I prefer dark mode” — Atlas uses remember_fact
“What are my meetings tomorrow?” — Atlas uses get_upcoming_meetings

You’ll see tool calls appear in the chat as Atlas works — a small indicator shows the tool name and what it’s doing. Once all tool calls are complete, Atlas delivers the final response.

The Tool Call Limit

Each response can involve up to 5 tool-use iterations. Most requests need just one or two. Complex requests — like “summarize all my notes on Project X and create a brief” — might chain several tool calls together.

If Atlas can’t complete a task within 5 iterations, it will explain what it was able to do and what’s left. You can just ask it to continue.

Core Tools vs. Lazy-Loaded Tools

To save context window space, Atlas uses a smart loading system:

Core tools (6 tools) are always fully available with complete parameter schemas:

search_notes
create_note
read_note
remember_fact
memory_search
read_memory_file

Other tools are loaded on demand. When Atlas wants to use a non-core tool, it first calls get_tool_schema to fetch that tool’s full parameters, then uses the tool. This happens automatically — you won’t notice it.

The benefit is that Atlas can hold more of your conversation history in context instead of spending tokens on tool definitions you might not need.

Read-Only vs. Write-Capable Tools

Tools fall into two general categories:

Type	Examples	What it does
Read-only	`read_note`, `list_files`, `get_tasks`, `get_person`	Reads data, makes no changes
Write-capable	`create_note`, `update_note`, `remember_fact`, `add_task`	Creates or modifies data

Write-capable tools are not dangerous — they work with your normal vault files, and deleted files are moved to trash (recoverable). But if you want to be cautious, you can disable specific tools in atlas/rules/allowed-tools.md. See Enabling and Disabling Tools for details.

What You’ll See in the Chat

When Atlas uses a tool, a compact indicator appears in the chat thread:

The tool name (e.g., search_notes)
The key input (e.g., query: "project henderson")
The result (e.g., Found 3 matching notes)

This gives you transparency into what Atlas is doing so you’re never left wondering why a note appeared or a task was created.

Next: Note Tools