OpenSidekick — Bring-Your-Own-LLM Browser Agent

About

OpenSidekick puts an AI agent in your browser's side panel — one that can read the page you're on and act on it: click, type, fill forms, navigate, and work across tabs.

Unlike closed assistants, you bring your own model. Connect a hosted model provider or run a local model on your own machine — your API keys stay in your browser and are sent only to the provider you choose. There is no OpenSidekick account, no middleman, and no telemetry. Not sure a model will work? A built-in Test button checks it for tool-calling and vision in one click.

WHAT IT CAN DO
• Summarize or answer questions about the current page
• Extract information such as emails, prices, tables, or links
• Fill in forms and click through multi-step flows
• See the page when it needs to — vision is on by default for models that support it
• Search, navigate, and work across multiple tabs to finish a task
• Save prompts, schedule recurring tasks, and record and replay workflows
• Connect Model Context Protocol (MCP) tool servers to extend what the agent can do

PRIVATE BY DESIGN
• Your keys and settings stay on your device
• Requests go straight to the model provider you choose
• No analytics, no tracking, no accounts
• 100% open source (MIT) — read or fork the code on GitHub

SAFE BY DEFAULT
• One-click approval control: plan first, ask before acting, or auto
• New sites ask before the agent acts; sensitive sites (banking, payments) always confirm
• The agent won't bypass logins or CAPTCHAs — it hands those back to you

Open source: https://github.com/esterhuizen/opensidekick