Skip to main content

ist β€” this one thing is all you need πŸ™‚

A guide for the friend who's just getting started. You don't need to know everything. From install to running an AI team, this one doc has you covered.

1. What even is ist? πŸ€”β€‹

In one line, it's a "command center for running several AIs at once."

When you put AIs (like Claude Code or Cursor) to work, ist lets you:

  • 🏒 Open several AI work "rooms" at the same time
  • πŸ‘₯ Group rooms into teams so they can hand each other work (like manager–worker)
  • πŸ‘€ Keep an eye on what each room is doing
  • πŸ€– Handle several AIs β€” Claude, Cursor, whatever β€” in one consistent way
  • πŸ“ And keep a record of the whole thing automatically

So it's not "grab one AI and chat" β€” it's a command tool for running a whole team of AIs.

Its real name is IST (Integrated Session Tools), but you don't have to memorize that πŸ˜„

Don't be intimidated! You don't need to use everything from day one. Learn just a few of the things below and you can get going, and the truly powerful part β€” "running an AI team" (TDA) β€” you'll bump into naturally once you're comfortable. πŸ™‚

2. Installing (from scratch) πŸ§°β€‹

πŸͺŸ On Windows? First you'll want to set up a Linux environment and VSCode. You can look this up online, or just ask an AI "how do I set up Linux (WSL) and VSCode on Windows?" and it'll walk you through it very kindly. Once that's set up, follow along from here exactly the same way.

The steps below are for Mac or Linux. Just paste one line at a time into your terminal.

β‘  Install Node/npm (via nvm)​

Node is the basic engine that these programs run on. Installing it with nvm makes version management easy. (Same for Mac/Linux.)

# 1) Install nvm (copy-paste this one line as-is)
curl -o- https://raw.githubusercontent.com/nvm-sh/nvm/v0.40.7/install.sh | bash

# 2) So the terminal recognizes what you just installed, close and reopen it, or run this one line:
source ~/.nvm/nvm.sh

# 3) Install the latest stable version of Node (LTS)
nvm install --lts

πŸ’‘ The v0.40.7 above is the nvm version. If it doesn't work or you want something newer, check the latest one-liner under "Install & Update Script" at github.com/nvm-sh/nvm.

Check that it installed:

node -v
npm -v

If both print a number, you're good! (If you get nvm: command not found, try fully closing and reopening the terminal.)

β‘‘ Install the AI CLI you want​

Just pick one AI to work with. A popular one is Claude Code:

npm install -g @anthropic-ai/claude-code

Cursor Β· Codex CLI Β· Gemini CLI and others work too. Just install one you like β€” that's plenty.

β‘’ Install tmux​

This is the tool used to keep several rooms (sessions) up at the same time. Pick the one for your setup:

# Mac
brew install tmux

# Linux (Ubuntu / Debian)
sudo apt install tmux

# For other distros, use your own package manager (e.g. dnf install tmux, pacman -S tmux)

Check that it installed:

tmux -V

β‘£ Get ready to install ist (snapshot)​

First, set up the installer tool that installs ist for you.

npm install -g @microwiseai/snapshot

Check that it installed:

snapshot --version

β‘€ Finish installing ist​

snapshot install @ist/beta

One final check that everything's done:

isesh --version
itda list

If a version number shows up and itda list runs without errors, you're almost there! πŸŽ‰

β‘₯ Turn on monitoring (just once)​

This is the switch that turns on auto-approve and the web dashboard. You only need to log in once.

detector-agent login

πŸ” You log in with a Google account. With just this β†’ the AI auto-clicks "yes" for you (auto-approve), and you can watch your sessions in the web dashboard. (More on that in section 7!)

3. Try starting your first manager session πŸ§‘β€βœˆοΈβ€‹

Once it's installed, give it a spin right away. It feels like placing one AI manager in a folder and handing it a job.

β‘  Make a folder to work in

mkdir ~/my-first-project

The AI needs a workspace to work in. (Any folder is fine.)

β‘‘ Start a manager session in that folder

isesh start session-name -p tda-manager -w ~/my-first-project

Here's how to read the options:

  • session-name β†’ the session's name (anything, e.g. myproj-mgr)
  • -p tda-manager β†’ open this session in the 'manager' role
  • -w ~/my-first-project β†’ the path to the folder to work in

πŸ§‘β€βœˆοΈ To put it as an analogy, it's "placing one AI manager in that folder and handing it a job." Now just tell the manager what you want done, and it takes care of it. (Actual instructions are in section 6!)

4. You only need to know 3 kinds of commands πŸ—‚οΈβ€‹

There look like a lot of commands, but don't panic. They split up neatly like this:

🟒 Just memorize kind 1. 🟑 For kind 2, just "know it exists." βšͺ For kind 3, don't worry about it at all.

🟒 Kind 1 β€” the ones you type yourself (just these!)​

The ones a human actually types are basically these three. Let me explain with everyday analogies.

πŸ“‹ See which rooms exist β€” itda list

The "room status board" at the office door. See at a glance which AI work rooms are running right now.

itda list

# If you want to see every room, including stopped ones
itda list -a

πŸšͺ Open a new room β€” isesh start

"Open a new work room and turn on the lights." Create a room to start working with an AI.

isesh start mywork -w .

Put any name where mywork is; -w . means "I'll work in this folder right here."

πŸ§‘β€πŸ’» Go into a room β€” isesh attach

"Open the door and go in to work together." Enter a room you made (or one that's running).

isesh attach mywork

πŸ’‘ If you want to close a room you're done with, isesh stop mywork β€” that's all you need to know. (Cleanup is in section 8!)

πŸ” Monitoring login (just once, the first time) β€” detector-agent login

The switch that turns on auto-approve + the web dashboard. Do it once at install and you're done.

detector-agent login

After that you don't have to think about it. Check who you're logged in as with detector-agent whoami, log out with detector-agent logout.

🟑 Kind 2 β€” ones you only need to know exist​

You don't have to use these now, but if you just remember the names β€” "oh right, that was a thing" β€” it'll make life easier later.

  • smon β€” the auto-approve helper. It watches a room and clicks "yes" for you when needed. That said, this is mostly turned on and used by the AI (the manager) on its own. You'll almost never type it yourself, and if auto-approve doesn't seem to be working, you can just tell the AI "turn on smon."
  • detector-agent β€” the essential backend for monitoring/auto-approve (the engine that detects sessions behind the scenes). It's not optional β€” it's required β€” but once you've done detector-agent login once, the rest just runs on its own.
  • The web dashboard ism.microwiseai.com β€” where you watch your sessions live in a browser. (More in section 7!)
  • skit β€” the tool for installing extra bundles like a tutorial kit (used exactly once, in section 10).

βšͺ Kind 3 β€” ones the AI uses on its own (you don't need to know these)​

  • ilogsession, imessenger, detector-agent start, smon watch …

These are behind-the-scenes commands the AI uses on its own. For example, when a manager assigns work to a worker it uses imessenger, and when monitoring/auto-approve is needed it turns on smon β€” but that's all handled by the AI on its own. (detector-agent too β€” once you've logged in, it fires up on its own when needed.) You don't need to know any of this, and you don't need to memorize it. Feel free to not worry about it 😌

5. Coming and going from sessions (entering Β· slipping out) πŸšΆβ€‹

The key is going into a room and then slipping out without shutting the room down.

β‘  Going in

isesh attach mywork

β‘‘ Slipping out (detach) β€” the room keeps running!

Sessions run on top of tmux. To step out, use the keyboard:

Press Ctrl-b, let go, then β†’ d

πŸ”‘ The important part here: stepping out does NOT shut the room down. Even after you leave, the AI keeps working away inside. You can jump back in later with isesh attach mywork. (Unless you've changed your tmux config, Ctrl-b d is the default.)

β‘’ Going back in

itda list          # check which rooms are alive
isesh attach mywork # back into that room

❗ Note: only use isesh stop mywork when you really want to shut it down. Don't mix it up with detach!

6. Giving the manager work Β· dividing it among workers (TDA) ⭐​

This is where the real heart of ist is. And yet, what you have to do is surprisingly simple.

You give the "big directions," the AI team does the actual work​

β‘  Go into the manager room

isesh attach session-name   # the manager session you made in section 3

β‘‘ Just tell it in plain words. (Not a command β€” talk to it like you'd talk to a person!)

"Add a 'mark as done' feature to the to-do app"

Then the AI manager takes care of it on its own:

  • Breaks the work into small pieces
  • If needed, creates several worker (subordinate) rooms
  • Splits the work up β€” like a frontend person / a backend person
  • Gathers the results and reports back to you.

🏒 Analogy: you're the boss, tda-manager is the team lead, and the workers are the team members. The boss just tosses out the big picture β€” "get this done" β€” and the team lead divides it up among the members. The notes the members pass around (imessenger) are handled by the AIs on their own, so you don't need to know about them.

How you watch and step in (optional)​

Whenever you're curious, you can peek in:

# See how the manager and workers are grouped, as a 'team org chart'
itda tree -a

# The full room list (including stopped ones)
itda list -a

# Peek directly at what a specific worker is doing (after looking, slip out with Ctrl-b d like in section 5)
isesh attach worker-name

In other words, you just give the big directions and observe β€” the AI team does all the hands-on typing. 😌

7. Watching Β· auto-approve Β· dashboard πŸ‘€β€‹

Auto-approve β€” you don't have to click "yes" every time​

Once you're running several rooms, the AI asks "is it okay to do this? (yes/no)" every time it wants to do something. Clicking every time is a pain, right? So here's how it works:

  1. Do detector-agent login once at install (step β‘₯ in section 2) β†’
  2. The AI (manager) handles auto-approve on its own via smon β†’
  3. You don't have to click "yes" every time. 😌

πŸ™‹ Auto-approve doesn't seem to be working? β‘  Check whether you're logged in with detector-agent whoami, and β‘‘ if it still doesn't work, just tell the AI "turn on smon." (You'll almost never type smon yourself β€” that's the AI's job.)

Web dashboard β€” everything at a glance in your browser πŸ‘€β€‹

You can see what your AIs are doing right now, live in a web browser.

πŸ‘‰ https://ism.microwiseai.com

πŸ”‘ Important: log in with the same Google account you used for detector-agent login. (If you log in with a different account, your sessions won't show up.)

What you can see here:

  • 🟒 Live session status (working / waiting / idle / error)
  • πŸ’° Token usage and USD cost (see at a glance how much it's costing)
  • 🌳 The manager–worker (TDA) tree (who's assigning work to whom)
  • πŸ“Š Rankings Β· stats Β· devices tabs

πŸ–₯️ Think of it as "a control-tower screen just for your AI team." You can skim everything from the browser without watching the terminal.

8. Wrapping up Β· cleaning up πŸ§Ήβ€‹

When the work's done, close the rooms to tidy up.

Close just one room

isesh stop mywork

Clean up a whole manager team at once (manager + workers together)

# Closing the manager cleans up its attached workers too (the default behavior)
itda stop session-name

# If you just want a preview of what would actually close
itda stop session-name --dry-run

# Keep the manager and clean up only the workers
itda stop session-name --workers-only

🧹 One good habit: when you're all done, check what rooms are left with itda list -a and tidy up β€” keeps things clean.

9. Stuck? πŸ†˜β€‹

  • Docs site πŸ‘‰ https://docs.ist.microwiseai.com (Type a command name into the search box and most things come up. No pressure β€” just pick out what you need.)
  • Just ask πŸ‘‰ Screenshot the screen that's not working and send it to me (the friend who set this up for you). That's the fastest way πŸ˜„

One common issue:

# If a room isn't showing up or you're wondering about the list (everything, including stopped ones)
itda list -a

10. If you want to learn more πŸ“šβ€‹

The above is plenty for now, but if you want to dig deeper:

  • Slowly read through the Quick Start / Guides sections of the docs site above.
  • And printing out the "cheat sheet" (ONE-PAGER.md) below and sticking it next to your monitor is handy.

If you want to get more hands-on practice, install the 'tutorial kit' and you can follow along one step at a time, interactively. Just type the one line below:

skit install @marnex/tutorial-kit

Once you install it and open a tutorial session, a "so what should we try now?" guide pops up, so just follow along 😊

πŸ’‘ This kit is being released soon, so if it doesn't install yet, let me (the friend who set this up) know!

And the real power of this tool is tying several AIs into a team so they assign work to each other and collaborate (TDA) (section 6!). You don't need to know it now β€” run a few rooms and you'll come across it naturally πŸš€

Have fun with it! At first, just remember start (open a room) β†’ list (see the list) β†’ attach (go in), and Ctrl-b d (slip out) πŸŽ‰