Launching Pickle: AI agent browser where you can actually see what they're doing, with the option to run local on your machine
Pickle is a browser built for AI agents, where you can watch every action happen live and step in if something looks off, instead of the agent running headless in the cloud with no visibility.
It also runs local models out of the box, and no API key is needed to start. I'm currently working on the rough edges and would love honest feedback.
If you want to try it, you download it here: https://picklebrowser.com/
Curious what the Indie Hackers community thinks makes an agent browser trustworthy enough to hand it real work. Is it visibility into actions, tighter approval controls, or something else entirely?
The ability to actually watch the agent click around and step in feels like a much more interesting angle than just making another automated browser. The local model support is a nice touch too. Feels like Pickle could be getting a lot more organic traffic than it is right now, especially while people are still figuring out what they’re comfortable letting agents do. Curious what kind of tasks people are trusting it with first?
For me visibility is necessary but not sufficient — the trust primitives that matter are an auditable action log you can replay, and hard guardrails on irreversible actions (destructive file ops, payments, credential access). Local models + no API key is a real differentiator since it keeps the action history on your own machine. Does Pickle have a mid-flight kill-switch that persists session state so you can resume after stepping in?
The live visibility into agent actions is a clear part of the product. Curious which aspect of Pickle people focus on first when they try it.