Media
5 items
Description
doeverything — an AI agent that drives your browser. Chat in the side panel; it navigates, fills forms, and runs tools for you. doeverything is an AI agent that lives in your browser sidebar and does exactly what the name says — everything. Tell it what you want. It navigates to the right pages, clicks buttons, fills out forms, pulls data, grabs screenshots, and works through multi-step workflows — all on its own, on any website. Open source — read the code, file issues, or contribute at https://github.com/tufanYavas/doeverything ━━━━━━━━━━━━━━━━━━━━━━━━━━━ WHAT IT CAN DO ON ANY WEBSITE ━━━━━━━━━━━━━━━━━━━━━━━━━━━ - Navigate — opens URLs, follows links, and handles multi-page flows - Click and interact — presses buttons, toggles checkboxes, and picks from dropdowns - Fill forms — types, completes fields, and submits, including on modern single-page apps - Read pages — sees visible text, links, and interactive elements without dumping raw HTML - Take screenshots — full-page or region captures mid-task for verification - Run JavaScript — injects custom scripts for advanced extraction or automation - Inspect network requests — reads and replays HTTP requests to see which APIs a site uses - Remember across sessions — saves site-specific learnings and recalls them next time ━━━━━━━━━━━━━━━━━━━━━━━━━━━ YOUR API KEY, YOUR MODEL ━━━━━━━━━━━━━━━━━━━━━━━━━━━ Bring your own API key from whichever model provider you already use — no subscription required. Prompts and replies travel between your browser and that provider. Compatible endpoints let you plug in self-hosted models or private deployments. ━━━━━━━━━━━━━━━━━━━━━━━━━━━ AUTOMATE ON A SCHEDULE ━━━━━━━━━━━━━━━━━━━━━━━━━━━ Set tasks to run on their own — daily, weekly, monthly, or every N minutes: - "Every morning at 8am, check my GitHub notifications and flag what needs attention" - "Every Monday, pull this week's analytics from the dashboard into my notes" - "Every 30 minutes, check if the product is back in stock and ping me" Scheduled runs happen in a background window, retry on failure, and send a desktop notification when done so you can review the results. ━━━━━━━━━━━━━━━━━━━━━━━━━━━ SKILLS & SLASH COMMANDS ━━━━━━━━━━━━━━━━━━━━━━━━━━━ Save your best prompts as reusable skills and fire them off with a slash command: /summarize — summarize the current page in bullet points /research $company — run competitive research on a company /scrape — turn what's on screen into a structured table /download — save what you're reading to a file Skills support argument templates ($company, $url), tool restrictions, domain-specific activation, and model overrides. ━━━━━━━━━━━━━━━━━━━━━━━━━━━ USE IT FROM OUTSIDE THE BROWSER ━━━━━━━━━━━━━━━━━━━━━━━━━━━ doeverything also runs as an MCP server. Paste the relay URL into an MCP-compatible desktop assistant, and it can browse through your Chrome — searching, reading, reporting back. Or install the local bridge, so the coding agent in your editor can read docs and click through the app you're building. ━━━━━━━━━━━━━━━━━━━━━━━━━━━ YOU STAY IN CONTROL ━━━━━━━━━━━━━━━━━━━━━━━━━━━ Four permission modes — pick whatever fits the task: 🔍 Ask every time — approve each action before it runs 📋 Follow a plan — review the agent's plan once, then let it go ✅ Allow for site — one approval per website per session ⚡ Hands-off — full automation; the agent still refuses payments, account deletion, and mass-sends unless you explicitly tell it to It remembers what you've approved on each site, so you aren't asked twice. ━━━━━━━━━━━━━━━━━━━━━━━━━━━ PERSISTENT MEMORY ━━━━━━━━━━━━━━━━━━━━━━━━━━━ doeverything keeps what it learns. If a store's checkout needs an extra click, it writes that down and reuses the note next visit — less rediscovery, fewer wasted tokens. - RAM memory — fast, conversation-scoped, clears on a new chat - Persistent memory — survives restarts, stored per domain, no size limit - Browse, export, import, and wipe memory from the settings panel ━━━━━━━━━━━━━━━━━━━━━━━━━━━ BUILT FOR REPETITIVE WORK ━━━━━━━━━━━━━━━━━━━━━━━━━━━ If you pull data from a dozen sites a week, fill the same forms all day, test user flows by hand, or copy dashboard numbers into spreadsheets, doeverything takes over the repetition — and gives you back the hours it used to eat. ━━━━━━━━━━━━━━━━━━━━━━━━━━━ PRIVACY ━━━━━━━━━━━━━━━━━━━━━━━━━━━ Your API keys are encrypted in chrome.storage.local and sent only to the model provider you pick. doeverything does not collect, store, sell, or log your conversations or browsing data — no account system, no analytics. When MCP relay mode is on, browser tool results (page snapshots, screenshots) pass through a relay server in real time and are thrown away immediately — nothing is kept or processed by doeverything's servers. ━━━━━━━━━━━━━━━━━━━━━━━━━━━ Add doeverything to Chrome. Tell it what you need. Watch it happen.