Automate everything. Use Codex, like an open source computer, with several agents that can work for you autonomously.
This is the 4th launch from Ara. View more
Ara
Launching today
Automate everything. Bring your own LLM provider and use Codex app like Computer use for free, with several agents that can work for you autonomously.
Free Options
Launch Team / Built With
Ara
Hey Agent Operators! 👋
We built Ara for people who want an AI that actually does things on their computer — not just types back in a chat window.
Instead of juggling ChatGPT, Slack, your browser, your IDE, and 14 other tabs, Ara lives in your notch. Hit a hotkey, give it a task, and watch it execute on your Mac.
Things you can text into existence:
🖥️ "Pull this week's Stripe revenue and drop it in our standup doc"
📧 "Reply to the last 5 emails in my inbox"
🔍 "Find every YC P26 AI agent company and put them in a Notion doc"
📅 "Book a 30-min with John and Sarah next week, send invites"
⚡ "Open my repo, pull main, run the tests"
No tab switching. No copy-paste. No babysitting. Just hit the notch, give the task, walk away.
Wispr Flow on the surface. Jarvis underneath.
If you try it, drop the wildest thing you got Ara to do — we ship daily based on what breaks 🚀
Ara
@adi_singh5 hyped for this 🙌
Ara
Let me know how you use it! Free to try with Bring your own AI!
ClawSecure
@adi_singh5 The notch-based UX is a clean entry point, but the interesting part is what's happening underneath. A Mac-native agent that operates across Stripe, email, browser, IDE, and calendar simultaneously means you're handling cross-application context in a way that browser-based agents can't touch.
We run complex multi-agent workflows daily at my company, and the hardest problem is always the orchestration between different application contexts. An agent that can pull data from Stripe, write to a Google Doc, and send a Slack message in one task is crossing multiple authentication boundaries, API surfaces, and data formats in a single execution chain.
A few technical questions I'd love to understand:
How are you handling authentication and credential management across all these integrations? If Ara has access to Stripe, email, calendar, and IDE simultaneously, that's a significant permission surface. Is there a scoped permission model where users can restrict which apps Ara can touch per task, or is it all-or-nothing once connected?
On the execution model: when Ara operates on the Mac desktop directly, is it using accessibility APIs and screen-based interaction, or native API integrations with each application? The reliability difference between those two approaches is massive, especially for tasks that require precise data extraction versus just clicking buttons.
How do you handle failure states mid-task? If Ara is five steps into a multi-app workflow and step six fails (API timeout, auth expired, unexpected UI state), does it roll back, retry, or surface the partial result? Multi-step cross-application tasks are where most agent workflows break in practice.
And on security: with an agent that has desktop-level access to everything on a Mac, how are you thinking about sandboxing and audit trails? Can users see exactly what Ara accessed and modified after a task completes?
Congrats on the launch! The "hit the notch and walk away" UX is compelling, but the trust layer underneath is what determines whether power users actually walk away or keep watching over its shoulder.
Ara
@adi_singh5 @jdsalbego Great breakdown! TLDR: we navigate apps via accessibility AX trees plus vision for computer use, sensitive apps blocked by default, every action visible live in the notch, stoppable anytime. Failures surface partial state.
ClawSecure
@adi_singh5 @smyhre Great to know. thats the right way to design this. great job
Product Hunt
Ara
@curiouskitty So we see if a task is completely lit or if the action is usable for other tasks and thus they're saved to memory. Long-term memory is like an Obsidian, a graph structure, 3D actually, to save these personal action items. Ephemeral task states are way more technical regarding the actual state of the product and not personal memory. We do, however, communicate with the CLI agent of the user to get the largest project that they are currently working on.
TypeBoost
That’s a really interesting approach. I’m excited to try this.
I imagine it probably takes a while to get used to handing off things that used to be manual work, but once that shift happens, I can see this becoming really powerful.
One thing I’m curious about though: is there some kind of security layer or approval system to prevent accidental file deletions or unintended emails or stuff like that?
Ara
@bennyqp Great to hear, yes security is important so it uses the same priveleges as your agent cli that you adapt in. If you log in with Claude Code (yes its still allowed) those permissions will follow.
We also have dedicated cursor views with live update so you can monitor what is acceptable and what is not :)
Ara
Good question! Ara has built-in guardrails. Sensitive apps like password managers are blocked by default, you can view every action and stop it anytime, and you can set up repeatable workflows for tasks you run often.
Ara
Hey Builders & Automators!! 👋
We built Ara for people who want AI to actually do the work, not just chat about it.
Think Wispr Flow, but with hands. Just say it. It's done.
Ara lives in your notch and drives your Mac like you would, moving the cursor, clicking apps, filling forms, navigating the browser. You watch it work.
🖱️ State-of-the-art computer use and ACP running right in your notch
🤖 Bring your own LLM and use Codex like Computer Use, free
🧠 Smart, auto-routed models (Claude, GPT, Gemini + 180 others)
🔗 1,000+ integrations (Gmail, Slack, Notion, GitHub, and more)
No prompting gymnastics. No babysitting. Just say it.
Try it and tell us what you automated first 👇
Ara
@smyhre this is cool!
Imagine being productive during your coffee chat or while you take a lap around the office. This is sick and will be experimenting with what I can get it to do for my startup without breaking :)
Ara
@joe_setpoint Exactly! The future is background agents handling most of the work while you're off doing literally anything else. Keep us posted on how it goes!