Numa is not a chat window you paste questions into. It is a program that lives on your desktop, sees your screen when you ask it to work, moves your mouse and keyboard when needed, and remembers how you do things if you show it once. Think of it as a junior coworker who never sleeps, never gets bored of repetitive tasks, and gets faster the more you teach them.
Why it's different
Every other AI tool sits inside a window and waits. You copy-paste the output into your real apps. You still do the clicking. The gap between "it told me what to do" and "it is actually done" is still yours to close.
Numa closes that gap. It reads your screen the way a human would — no API, no integration catalog, no special plugin for each app. It sees whatever is on your screen, decides the next action, moves your mouse, types, and keeps going until the task is done.
Your unusual internal tool, your company's custom dashboard, the app no integration catalog covers — Numa works in all of them because it works the same way you do: by looking and doing.
A concrete example: you could ask ChatGPT how to reorganise a folder of invoices. Numa can actually open the folder, read the file names, rename them the way you showed it last month, and message you when it is done — while you are in a meeting.
What Numa is not trying to be
Numa is not a rigid automation tool like n8n or Zapier where you wire up "if this email arrives, send that reply" in a diagram. Those tools are excellent for high-volume, simple triggers. Numa is for messier work — the kind where you use five different apps in a specific order that no integration catalog covers, and where the steps change slightly every time depending on context.
Classic automation tools are like programmable machines. Numa is like a person. You do not program it with logic gates — you show it. You do the task once while explaining out loud. Numa watches and listens. Next time you say a short phrase, it runs the steps exactly as you showed.
Who it's for
Numa is built for everyone — not just developers. We deliberately avoided hardcoded paths because not everyone uses Slack, Stripe, and Gmail. Numa adapts to what you actually use.
It will serve you well if
You repeat multi-step tasks on your computer and wish someone else could handle the boring parts. Your tools are spread across apps that do not talk to each other natively. You want something that adapts to how you work, not how a product designer imagined a workflow. You are willing to teach it once — record a workflow — to get reliable results later.
It's probably not for you if
You only need simple automations — "when email arrives, send template reply" or "when purchase happens, send receipt." Dedicated tools do that better and more reliably. You need guaranteed, production-grade pipelines with zero margin for error. Or you are not comfortable giving an app broad access to your computer — Numa needs that access to function.
What you need
macOS or Windows for the full experience. iPhone (App Store) as an optional companion for remote control — not a replacement for desktop. No Linux, no Google Play. An always-on internet connection — Numa's brain runs on our servers. An account, signed in with Google or email. A microphone only when recording workflows or using voice mode — otherwise unused.
Desktop = Numa acts on your computer, sees your screen, does your work. Phone = check in, trigger things remotely, quick access. Think pocket remote, not the main stage.
Hardware
Numa runs on your machine. While thinking happens on our servers, your computer still does the real work: reading the screen, moving the mouse, typing, opening apps, running code, and handling screenshots as tasks unfold. Heavier use — long multi-step jobs, several apps at once, large files — asks more of your hardware.
We do not hard-block installs below these numbers. Older or tighter machines can still run Numa. Things may simply feel slower, especially on demanding tasks.
Minimum: 8 GB RAM, ~2 GB free disk, a modern quad-core CPU (roughly 2018 or newer), macOS 11 or later or Windows 10/11 64-bit. GPU not required.
Comfortable: 16 GB RAM, 5 GB or more free disk, 6+ CPU cores. macOS: Apple Silicon (M1 or newer), 16 GB unified memory. Windows: a recent dedicated GPU (2 GB+ VRAM) helps for smoother screen-heavy work — optional, not required.
macOS notes
Apple Silicon (M1/M2/M3/M4) is the comfortable tier for everyday agent work. Intel Macs are fine for lighter tasks; long or complex runs may feel slower if RAM is tight or many apps are open.
Windows notes
A dedicated GPU (e.g. NVIDIA) is optional — not required to install or use Numa, but can help when the screen updates constantly during a task. CPU only or integrated graphics works for most jobs; expect more lag on long, screen-heavy runs at minimum RAM.
What actually needs the horsepower?
Short tasks and light agent runs: minimum spec is usually fine. Chat and folders without heavy runs: yes. Long multi-step tasks across many apps, large files, many screenshots per task, or code-heavy runs: comfortable spec recommended.