HUD dashboard · read-only · speaks Portuguese

Your robots are at work. One screen shows everything.

And says good morning: what ran since yesterday, what failed, what’s waiting for you, and the first thing to do.

Banner: your robots on screen, a dashboard that shows everything and says good morning
In short

jarvis-painel is a sci-fi-style screen (a “HUD,” the information panel you see in movies) that brings together, in one place, everything running automatically on your computer: Linux timers, queues, and log files. It’s for people who have bots, routines, and queues running but don’t know what each one has done or what’s waiting for them. It shows what finished successfully, what failed, and your to-do list, and reads a summary aloud in Portuguese. You only need Python 3 and a browser; everything runs on your machine, with no key or paid service, and it only reads data: the buttons don’t run anything.

Video

Watch: the panel in 2 minutes

The video (in Portuguese) shows the screen at work: the bots, what failed, what is waiting for you and the spoken good morning.

Open on YouTube → · 9:16

What it is

A screen for the robots you can’t see

People who automate things end up with dozens of routines running in the background. The dashboard reads them all and shows them in three sections: Systems, the core that speaks, and "Esperando você" (Waiting for you).

Banner with six features: Systems, Waiting for you, Good morning by voice, Focus mode, Read only, and Mobile phone

Actually read on an INEMA machine on Nov 8, 2026, at 08:18 (the numbers will be different on your machine and change every hour):

62systems read (53 timers, 6 services without timers, 3 queues, and logs)
40finished successfully
11failed (the same count in the core and the list)
42waiting for you (11 failures, 30 pending items, 1 queue)
The dashboard on a computer: Systems on the left, the JARVIS core with rings and a wave in the middle, Waiting for you on the right, and a shortcut bar at the bottom
The dashboard saying good morning: the wave around the core follows the speech, and the text appears as if typed below. The pending items were replaced with "Pendência N" (Pending item N) in --anonimo mode for this screenshot.

🛰️ Everything in one place

Linux timers with the latest result from each one, failed services, JSON queues, logs, and your markdown file of pending items.

🔊 Talks to you

"Bom dia" (Good morning), "Status" (Status), "O que espera" (What’s waiting), and "Modo foco" (Focus mode). Voice in Portuguese from the browser itself, with no paid service.

🔒 Read only

"Aprovar" (Approve) and "Depois" (Later) only mark items on the screen. Nothing is started, stopped, deleted, or sent. Anything it couldn’t read appears as "exemplo" (example).

How it works

Two scripts and a page

coletar.py reads the machine and writes to estado.json. bom_dia.py writes the summary. The page reads both and draws the screen.

config.json→ coletar.py→ estado.json→ bom_dia.py→ index.html→ you

Central core

Three rings spinning at different speeds, pulsing light, the name JARVIS, and a status line. Startup takes about 3 seconds: "Inicializando… Sistemas no ar." (Initializing… Systems online.) "Bom dia" (Good morning), Nei.

Systems (left)

Each timer, service, and queue has a dot and a status word: ok, failed, running, or waiting. Below, see how long ago it ran and what it does.

"Esperando você" (Waiting for you) (right)

First, anything that failed, then the pending items in your file and items waiting in queues. "Aprovar" (Approve) and "Depois" (Later) buttons, which only mark items on the screen.

Top

Clock, date, and the "Voz: ligada" (Voice: on) badge. Click the badge to turn off the voice.

Command bar

"Bom dia" (Good morning) · "Status" (Status) · "O que espera" (What’s waiting) · "Modo foco" (Focus mode), also on keys 1, 2, 3, and 4. Focus dims the panels and leaves only the core.

Colors

Ice cyan and white on navy; amber only for alerts. Rings at 60 frames per second and no horizontal scrolling on mobile.

Prerequisites

What you need beforehand

Nothing to install besides Python. Ollama is optional.

Python 3.9 or newer and git

Python comes with Ubuntu and Mac. Git (which downloads the toolkit) doesn’t always: on Ubuntu, install it with sudo apt install git. Check both:

python3 --version
git --version

Linux with systemd (for real data)

A timer is a task that runs on its own at a set time (Linux’s “scheduler,” called systemd). Timers come from systemctl --user. Without it (on Mac, for example), systems appear as “example”; pending items and queues remain real.

systemctl --user list-timers

Browser and, if you like, Ollama

Chrome, Edge, or Firefox. The pt-BR voice depends on your system’s voices. Ollama is a free program that runs an AI model on your own computer (installation at ollama.com); here, it just makes the summary more natural. If you have it, this command lists the models:

ollama list
Usage guide · step by step

From scratch to "Bom dia" (Good morning) in 5 minutes

Every command below was run on a real machine. Paste it into the terminal, one block at a time.

1

Download the kit

Creates the jarvis-painel folder with the scripts, page, prompts, and tests.

git clone https://github.com/inematds/jarvis-painel.git
cd jarvis-painel
2

Say who you are and where your pending tasks are

The config.json is set up for INEMA. Open it in the terminal editor (save with Ctrl+O, Enter, then exit with Ctrl+X):

nano config.json

The simplest option is to delete everything and paste this, changing the name and path to your pending tasks file (if you don’t have one, leave it as is: the panel shows an "exemplo" (example) in its place):

{
  "nome": "Marta",
  "pendencias": { "arquivo": "~/pendencias.md", "parar_em": "Resolvidas" },
  "fontes_extras": [],
  "ignorar": ["snap.*", "run-u*", "*.scope"]
}

The INEMA version included in the kit also reads two task queues and a log:

{
  "nome": "Nei",
  "pendencias": { "arquivo": "~/projetos/wifi/pendencias.md", "parar_em": "Resolvidas" },
  "fontes_extras": [
    { "nome": "fila-espanhol", "tipo": "fila_json", "caminho": "~/projetos/wifi/fila-espanhol/fila.json" },
    { "nome": "log-fila-espanhol", "tipo": "log", "caminho": "~/projetos/wifi/fila-espanhol/fila.log" }
  ],
  "ignorar": ["snap.*", "run-u*", "*.scope"]
}

In the pending tasks file, each line that starts with - **title** or ### title becomes an item, up to the ## Resolvidas section.

Markdown is a plain text file (.md) where **word** marks bold text; create yours with nano ~/pendencias.md and write one pending task per line, like this: - **Pay the bill**. A JSON queue is a list of tasks that a robot writes to a file; a log is the file where it records what it did.

3

Check the system

Read-only: timers, the latest result from each service, failures, pending tasks, and queues. The final line summarizes what it found.

python3 coletar.py

On the INEMA machine, the terminal showed (the script writes in Portuguese):

estado.json: 62 sistemas (53 timers) · 40 ok · 11 falharam · 42 esperando você · 0 exemplos

If there’s a typo in config.json, the script gives the line and column where the error occurred.

4

Write the morning summary

Based on fixed rules, with no AI: what the robots did since yesterday, what you’re waiting on, and the first thing to do. Saves to bom_dia.txt.

python3 bom_dia.py

With Ollama running, the local model makes the text sound more natural without changing the numbers (the model is set in config.json; if Ollama doesn’t respond, the rule-based text remains). On the INEMA machine, it took 17,5 seconds with qwen3.8:27b.

python3 bom_dia.py --ollama
5

Open the page

The page needs a local server to read estado.json. Then open http://localhost:8000 in your browser. It rereads the state every 60 seconds. Leave this terminal open: if you close it, the page stops updating (to shut it down, press Ctrl+C).

python3 -m http.server 8000

For the next steps, open another terminal and go to the folder with cd ~/jarvis-painel.

6

Use the shortcuts

Click or press keys 1 to 4: "Bom dia" (Good morning), "Status", "O que espera" (What you’re waiting on), and "Modo foco" (Focus mode). Under "Esperando você" (Waiting for you), "Aprovar" (Approve) and "Depois" (Later) only mark items on the screen and are saved in this browser; "Desfazer" (Undo) removes the mark. Nothing is run.

7

Update automatically (optional)

Open the scheduler with crontab -e and paste the line below at the end of the file. It works if you downloaded the kit to your home folder (change the path if you downloaded it somewhere else). Every 5 minutes it rereads the system and rewrites the summary; coletar.py finds your timers even when run by cron.

*/5 * * * * cd $HOME/jarvis-painel && python3 coletar.py && python3 bom_dia.py
8

Take a screenshot without showing your pending tasks

Anonymous mode replaces each title with "Pendência N (data)" (Pending task N (date)). Timer names remain visible.

python3 coletar.py --anonimo
9

Make sure everything is working

The tests run offline and check data collection with fake systemctl output, an unreadable item that becomes "exemplo" (example), and the summary without Ollama. This command works without installing anything:

python3 -m unittest discover tests

If you have pytest, python3 -m pytest -q works too.

On screen

The startup, the phone, “Aprovar” (“Approve”), and focus

Screenshots from the same machine, with tasks in anonymous mode. The interface and the voice are in Portuguese.

Startup: the core appears with the rings and the text “Inicializando… Sistemas no ar.” (“Initializing… Systems online.”) (62 read)
Startup takes about 3 seconds: “Inicializando… Sistemas no ar.” (“Initializing… Systems online.”), then “Bom dia” (“Good morning”), Nei.
“Esperando você” (“Waiting for you”) panel with an item marked “Aprovado (só na tela)” (“Approved (on screen only)”) and the “Desfazer” (“Undo”) button
“Aprovar” (“Approve”) only marks the item on screen: it dims with “Aprovado (só na tela)” (“Approved (on screen only)”) and the “Desfazer” (“Undo”) button.
“Modo foco” (“Focus mode”): side panels dimmed and only the core lit up
“Modo foco” (“Focus mode”): the panels dim and the core is left alone.
The panel on a phone: core at the top, counters, lists, and the shortcut bar fixed at the bottom
On a phone (390 px): no horizontal scrolling, with the shortcut bar fixed at the bottom.
Ready-made prompts

Three prompts to copy and paste

They’re in the prompts/ folder (with English and Spanish versions in prompts/en/ and prompts/es/).

Jarvis’s personality

Paste this at the start of a conversation with a model: English, short sentences, only use the numbers from the dashboard, say "I don’t know" when data is missing, and never do anything on its own. File prompts/en/jarvis-personalidade.md.

Customer research, read-only

For Claude Code with a browser. Change the industry and city. File prompts/en/pesquisa-clientes.md.

Search for <industry> in <city>. Open the first 3 sites that appear.
Make a table with: name, website, phone number, and one thing the site does poorly.
Rules: only read. Don’t contact anyone or fill out any forms.

Fix the dashboard

If the dashboard opens to a blank screen or with an error, open Claude Code or Codex in the folder, paste the request and the console error (F12 key). File prompts/en/conserta-tela.md.

Want to talk by voice with a model running on your GPU? The dashboard works on its own, and the sibling kit jarvis-local adds hearing and speech. Use prompts/en/jarvis-personalidade.md as its instructions.
Limits and next steps

What it does today and what it doesn’t

So you know what to expect before using it day to day.

Today
Version 1.0Reads the user’s systemd timers and services, pending items in markdown, JSON queues, and logs. Speaks using the browser’s voice. Summaries are generated by rules and, optionally, by Ollama.
Doesn’t do
No actions on the machineIt doesn’t start, stop, or restart services; "Aprovar" (Approve) and "Depois" (Later) go nowhere beyond the browser. System timers (root) and cron jobs don’t appear.
Voice
Depends on the browserSome browsers only speak after the first click on the page. If no pt-BR voice is installed, the text still appears on screen.
Idea
More sourcesRead cron and system timers, and Docker containers. Not available yet.