MOP-launch-film-inline.mp4

MOP — top, for the age of AI agents.

MOP is modeio's all-in-one product for seeing what AI agents do on a computer. It brings the familiarity of top, htop and btop together with AI security review: inspect processes, follow their activity, and open the evidence behind a decision.

The film introduces the three terminal views, the guard model slot, and the vLLM-Jev engine behind them—with specialized guards for privacy, permissions, content and compliance in the product roadmap.

Install · Usage · Live demos · Models · Docs

Install

Homebrew on Linux

brew install mode-io/tap/mop

Debian / Ubuntu — add the signed repository and install:

curl -fsSL https://github.com/mode-io/mop/releases/download/apt/setup-apt.sh | sudo bash

After that, use sudo apt install mop and sudo apt upgrade as usual. Run mop, hmop or bmop from any directory. No GitHub account is required.

RPM, Arch, portable Linux, uv and pipx →

For AI review, start the local OpenJev service. For background output and API-body collection, enable full capture.

Use it like top

Command	Familiar style	Best for
mop	top	A compact process list
hmop	htop	Meters, process trees and inspection
bmop	btop	Resource graphs and a security dashboard

P CPU sort · M memory sort · ↑ ↓ select · Enter inspect · m models · ? help · q quit

mop                          # watch your system
mop -p 12345                 # focus on one process
mop run -- python app.py      # review a command and its output

For API requests, after enabling full capture:

mop-network exec -- python app.py

To keep reviewing after you close the terminal:

mop daemon start

All three views connect to the same background observer.

See it in action

Recorded from a running Linux host. The clips show live process views, CPU/memory sorting, row selection and navigation through the model catalog.

View	Recording
MOP	Watch video
H-MOP	Watch video
B-MOP	Watch video
Models	Watch video

How it works

Observe → Review → Inspect

MOP collects system activity and captured content, groups related evidence, and splits long inputs into small review units. A shared guard policy classifies them as normal, suspicious or high_risk. Open a finding to inspect its evidence. unknown means no usable judgment was returned.

Reviews run in the background through vLLM-Jev. You can sort, inspect and manage processes from the same terminal. Customize categories and the shared prompt →

Models

OpenJev powers the current core slot while we build the MOP model family. Press m, or run mop models, to explore the catalog.

Model	Focus	Status
jev-core	General activity review	OpenJev placeholder available; dedicated model coming soon
guard-privacy	Personal data and secret exposure	Coming soon
guard-permission	Permissions and authorization	Coming soon
guard-content	Harmful content and scams	Coming soon
guard-compliance	Policy and compliance review	Coming soon

Model slots and custom rules →

---

Installation · Controls · Capture coverage · Architecture · Performance

Built by modeio. Inspired by top, htop and btop.