Open-source AI agent for Chrome. Reads and acts on the page you're on using any model you choose — OpenRouter, OpenAI, Claude, Gemini, Groq, or a local model. Keys stay in your browser. MIT, no account, no telemetry.
No reviews yetBe the first to leave a review for OpenSidekick
Maker
📌
Hey PH. I spent the last few weeks building this because I wanted the "AI drives my browser" thing without handing my browsing to one AI company.
So: bring your own model (any OpenAI-compatible endpoint or the Anthropic API, including local Ollama / LM Studio), keys stay in your browser, and the whole thing is MIT so you can read exactly what it does. It reads the page and acts on it, with a plan/ask/auto toggle so you decide how much rope it gets.
Two things I'm proud of that sound small: a Test button that checks whether a model can actually do tool calls + vision before you waste time on it, and the agent always hands logins and captchas back to you instead of trying to be clever.
It's early. If you tell me which model you ran it with and where it fell over, that's worth more to me than an upvote.
Report
Love that keys never leave the browser, that's the right call. One thing I'd love to see is a per-site allowlist or blocklist so I can lock it down to only run on domains I trust, with a quick toggle for new sites. Would make it feel safer to leave running in the background across my workflow.
Report
Maker
@n_yalc26947 that's a great idea, thank you. Will add it to the roadmap for future releases.
Report
Maker
@n_yalc26947 Shipped in 0.1.6: Settings → Site permissions → 'Only work on sites I allow', plus a trust toggle on the site chip under the chat box. In that mode it won't even read a page on a domain you haven't trusted.
Report
Keys living in the browser with zero telemetry is a really thoughtful default, and swapping between local and hosted models per task was smoother than I expected.
Love that keys never leave the browser, that's the right call. One thing I'd love to see is a per-site allowlist or blocklist so I can lock it down to only run on domains I trust, with a quick toggle for new sites. Would make it feel safer to leave running in the background across my workflow.
@n_yalc26947 that's a great idea, thank you. Will add it to the roadmap for future releases.
@n_yalc26947 Shipped in 0.1.6: Settings → Site permissions → 'Only work on sites I allow', plus a trust toggle on the site chip under the chat box. In that mode it won't even read a page on a domain you haven't trusted.
Keys living in the browser with zero telemetry is a really thoughtful default, and swapping between local and hosted models per task was smoother than I expected.
@talha132854 so glad to hear that. Thank you!