Show HN: Output – An AI Agent That Controls Your Computer (Mac and Windows)

15 points by sudhar172 1 month ago | 14 comments
Hey HN,

After months of building, we're excited to share Output — the first AI agent that can truly control your computer.

You can speak or type instructions like:

“Fill out this job application using the info in this PDF” “Edit this video in DaVinci Resolve” “Schedule a Zoom call and send the invite”

Output understands your intent and directly clicks, types, and navigates apps across your system — without relying on scripts, plugins, or integrations. It's like giving your computer a brain.

Works on both Mac and Windows

No setup or scripting required

Handles real apps like Chrome, Excel, Notion, Zoom, etc.

Designed for general-purpose use — students, founders, creatives, you name it

We're still early and improving every day, but we'd love your feedback. Live demo + early access: https://theoutput.co/ You can follow us on X: https://x.com/theoutputai

Happy to answer any questions and go deep on how it works!

  • cess11 1 month ago
    According to the video examples it's very slow.

    Where can I look at the privacy policy?

    • sudhar172 1 month ago
      We’ve already optimized it for speed — most actions complete in 1 to 2 seconds in real usage

      We’re still finalizing the privacy policy and will share it publicly soon. For now, no data is stored long-term, and we’re exploring both cloud and local processing options with user control in mind.

      • cess11 1 month ago
        Which seems slow, and you'd also need to add the time it takes to write the prompt.
    • cadamsdotcom 1 month ago
      “It’s like giving your computer a brain”

      Agree with the sentiment but trust in software is at an all time low and you have to work against that.

      An example success is the developer of Superwhisper. Crazily responsive online, always polite, explains features with blog posts and videos, has a fan base of creators on youtube who fill in gaps in the docs and talk about creative & cool uses for the tool, and where possible the software provides local only ways to get work done.

      If going closed source (which is fine) you could completely copy their strategy.

      • sudhar172 1 month ago
        You're right, trust is the hardest part, especially with software that can control your computer. We're focused on giving users transparency, control, and a clear view of what the agent is doing. Local mode and permission controls are on our roadmap. Really appreciate you highlighting this.
      • trumbitta2 1 month ago
        How can I trust something like this?
        • sudhar172 1 month ago
          That's a very reasonable concern. Trust has to be earned, especially with software that can take actions on your behalf.

          We’re starting with clear boundaries — no hidden actions, full visibility into what the agent is doing, and confirmations for anything sensitive. A local-only mode is in development for users who want full control.

          This kind of tool only works if people feel safe using it, so building that trust is a core part of what we’re focused on, not just a nice-to-have.

        • daveguy 1 month ago
          I'm having trouble finding the source code.
          • sudhar172 1 month ago
            We’re not open source yet, which is why the code isn’t available right now. We’re focused on getting the core product stable and secure first, but we’re definitely open to sharing parts of the stack over time.
          • yahoozoo 1 month ago
            Is this a proprietary model?
            • sudhar172 1 month ago
              We use an open-source vision-language model as the base and fine-tune it heavily for agent behavior. The intelligence is in how it reasons, acts, and adapts — not just in the raw model.
              • yahoozoo 1 month ago
                Cool! Thanks for the reply. I’ll try to keep an eye on this. Mind if I ask what open source base model is?
            • 1 month ago
              • devincognito-in 1 month ago
                [dead]
                • ohadron 1 month ago
                  [flagged]
                  • 1 month ago
                  • revskill 1 month ago
                    So we can tell robot to obey us? What if they really have a brain ?
                    • sudhar172 1 month ago
                      That is the goal. We want to build something that truly understands and can take initiative. But we are designing it with strong guardrails, clear permissions, and full user control from the start. Intelligence without boundaries is neither useful nor safe.