Make it yours
Memory, skills and MCP
What it learns about you is a file you can read. What it knows how to do is a shelf of names it opens on demand.
Memory: nothing is kept without you
When something is worth keeping, Aetox asks with a card: Let me remember this? It has not written anything yet. Approve writes it, No drops it and it will not be asked again in different words, Edit lets you fix the wording first, and Store elsewhere picks which file it goes into.
| File | What it holds | Who reads it |
|---|---|---|
USER.md | Who you are, what you do, how you like to be worked with | Every desk and every agent |
MEMORY.md | How this machine is set up, and what the work taught the assistant itself | The Assistant desk |
SHAPE.md | How you want answers built — length, order, what never to repeat | Every desk |
All three are plain Markdown in %APPDATA%\Aetox\memory\, and all three are editable inside the app: Settings › Memory lists every file with its lines, and Settings › About you holds USER.md on its own.
- A memory takes effect from the next message, not the one you are in.
- What was approved is never asked for a second time; what was refused is not asked again at all.
- Turn the whole thing off with one switch in Settings › General — *let Aetox learn from the work it does*.
Skills: what the model cannot guess
A skill is a folder with a SKILL.md in it. It carries what a model does not know: the frame size of a slide room, how a Thai tax invoice rounds, how an attacker breaks a path check. Aetox ships with a set, and adding one is a file — not a rebuild.
- Where
- Capabilities › Skills in the app: *your skills*, the library, and what each agent is allowed to open
- On disk
%USERPROFILE%\.aetox\skills\<name>\SKILL.md- Ways in
- Add a whole set from the library · install from a URL or a
.zip· ask the assistant to write one and approve the card - How it is used
- The model sees the names on the shelf and opens one with
skill_viewonly when the job needs it — fifty skills cost what five cost
MCP: tools someone else wrote
An MCP server adds tools Aetox does not ship: your own database, a tracker, a service you use. Add one in Capabilities › MCP servers — an address or a command, a key if it needs one — then press Test, which connects, initialises and lists the tools that came back.
- A server is listed for the desks and agents you tick, not for everything at once.
- Every tool that comes from an MCP server asks before it runs, at every permission level.
- The list is one file,
%APPDATA%\Aetox\mcp-servers.json, so it can be copied to another machine.
Importing from another app on this machine
If this machine has run Claude Code, Codex or another agent app, what you already set up there does not need setting up twice. The found in other apps on this machine band in Capabilities › MCP servers holds four kinds in one list: tick what you want and press Import selected once, instead of opening a card per row.
- Skills
- Copied from
~/.claude/skills,~/.codex/skillsand~/.agents/skillsonto Aetox's shelf — a name that already exists there is never overwritten. - MCP
- Read from the config files of Claude Desktop, Claude Code, Cursor, VS Code, Windsurf, Gemini CLI and Codex. A server that wants a key still opens its card so the key can be filled in first.
- Hooks
- Read from
~/.claude/settings.jsonand translated into Aetox's hooks — tool names are mapped (Bash → shell, Write/Edit → change) and the command lands in%APPDATA%\Aetox\hooks.json. - Programs
- Agent programs already installed (Claude Code, OpenCode, Devin, Antigravity) are added as code agents for the code desk — a program Aetox has no driver for says so and cannot be ticked.
Importing is a one-time copy, not a link: change the hook at the other app later and Aetox does not follow, and anything Aetox already has is not offered again.