Starizon.ai

AI Agent for Auto Browse & Automation

🔥 Starizon is an AI assistant that lives in your browser sidebar. It reads the page you are on, and it can do the work on that page for you. Press Alt+Shift+S to open it (Command+Shift+S on Mac). 💪 Features 1️⃣ Chat & Summarize ✔️ AI chat: Ask anything from the sidebar. No extra tab, no copying text back and forth. ✔️ Page Q&A: Ask about the page in front of you. Answers come from what the page says, not from the model's memory. ✔️ Screenshot chat: Grab any part of the page and ask about the image. A chart you can't read, an error dialog that makes no sense, a layout that broke. Show it instead of describing it. ✔️ Multi-page chat: Select several tabs, or a whole tab group, and ask once. Compare prices, specs, or what three sources claim, side by side. ✔️ Document chat: Upload a PDF, Word or Excel file and ask about the clause or the number you need. ✔️ Page summary: Long articles, reports, manuals. Read the summary first, then decide whether the full text is worth your time. ✔️ Video summary: Get a timestamped outline of a YouTube video. Click any line to jump to that second. 2️⃣ Agent S6 web automation ✔️ Plain language is enough: Describe the task. No scripts to record, no selectors to write, no page structure to think about. It looks at the page and finds the button to click and the field to fill. ✔️ It uses the browser the way you do: It works from what it sees on screen. A long fiddly form, an entire signup or checkout flow, dozens of records to edit one by one in an admin panel. Hand it the tedious work. ✔️ Multi-step, across pages: One task can span several pages and tabs. It decides each next step as it goes instead of following a fixed script, so a redesigned page doesn't stop it cold. ✔️ It asks when it should: A login wall, a human verification check, a choice it isn't sure about. It stops and hands control back to you, then picks up where it left off. It won't click blindly or pretend it succeeded. ✔️ Change your mind mid-run: No need to wait for it to finish and redo the work. Stop it anytime, add a note with a new requirement, a correction, a constraint or a different direction, then continue. It resumes from that step rather than starting over. ✔️ It keeps you posted: Where it is, what it ran into, how it turned out. It tells you as it goes, and it can notify you when the run ends so you don't have to watch. ✔️ It looks things up: When the page can't give it what it needs, it searches rather than coming back empty-handed. ✔️ It reads pages in batches: Several pages in parallel, each turned into clean text, with the data you need pulled out along the way. ✔️ It uses what you've built: It calls the skills you saved when one fits, along with the apps and MCP tools you connected. 3️⃣ Replay ✔️ Get it right once, then one click: Save an Agent S6 run, a workflow you built, or an extraction you configured. Click to run it again without describing anything. ✔️ Works on a fresh batch of pages: It saves the steps, not hardcoded coordinates. The next page, another product, another customer. Same structure, same run. ✔️ Survives a redesign: If the starting page doesn't match or the structure changed, it repairs that step where it can and keeps going instead of failing outright. ✔️ Free reruns: Replaying on similarly structured pages costs you nothing. ✔️ Editable: Change the title and description, or open the flow and adjust it step by step. Save a copy to keep a variant and leave the original untouched. ✔️ Turn it into a command: Save it as a skill at the same time, then trigger it by typing /your-command, or let Agent S6 call it. ✔️ Put it on a schedule: Any saved replay can run daily or weekly on its own. ✔️ Stays manageable: Search, multi-select and bulk delete, so a large library doesn't turn into a mess. 4️⃣ Tab management ✔️ One sentence, several jobs: Type /tabs and say it plainly. Group, open, close, reorder, switch, ungroup. It follows more than one instruction in a single sentence. It works on the current window. ✔️ Automatic grouping: Dozens of tabs sorted by topic. Rename groups, recolor them, split one that got too crowded. ✔️ Ask without touching: How many are open, which are duplicates, which have sat untouched for a month. It answers and leaves your tabs alone. ✔️ Plan first, act second: You get a plan showing the grouping and anything it proposes to close. Nothing happens until you confirm, you can undo afterwards, and closed tabs come back with one click. ✔️ It doesn't take liberties: Duplicates and month-old tabs are not permission to close them. Unless you say so, nothing gets closed. Pinned tabs stay out of grouping, moving and closing entirely. ✔️ It can't read your pages: It sees tab titles, addresses and group names. What is written on the page stays out of reach. ✔️ Ask a whole group: Type @ in chat to point at an entire tab group and question or summarize all of it at once. 5️⃣ Knowledge Base ✔️ Local-first: The full text and the search index stay on your machine. Pages behind a login save too, the ones cloud clippers never reach. ✔️ Privacy-first: Nothing is uploaded to a server to be indexed. Search runs on your device, and only the passages needed to answer reach the model. With your own API key, requests go from your browser straight to the model provider and never touch our servers. ✔️ Save as you go: Right-click to keep a whole page, or highlight and keep only that part. Save first, sort later. ✔️ Bring it in bulk: Years of bookmarks (whole folders at a time), the row of tabs open in this window, a list of links you pasted, files from your computer. Pick them and import in one pass. Files can be PDF, Word, Excel, web pages, plain text and code. ✔️ Keep what the chat produced: Conversations and summaries have a save button. When an answer is worth keeping, one click puts it in your Knowledge Base and it stays searchable. ✔️ Collections: Group by project or topic, then use @ to aim a question at one collection so nothing irrelevant muddies the answer. ✔️ Answers with sources: It runs its own searches, opens what is actually relevant, and tells you which saved item each claim came from. 6️⃣ Scheduled tasks & web monitoring ✔️ Scheduled tasks: Use /schedule for a one-off reminder, or have an agent run on your rhythm. A digest of the sites you follow, every morning. ✔️ Schedule what you already have: Saved replays and skills go on a schedule without describing them again. ✔️ Web monitoring: Use /monitor and describe the change you care about in your own words. A price drop, changed terms, a spot opening up. It watches and tells you. ✔️ Watch one area: Point it at a region of the page so changes elsewhere don't bother you. ✔️ When they run: Scheduled tasks and monitors run while the browser is open. 7️⃣ Bookmark search ✔️ Local-first: Your bookmarks stay in your own browser. Nothing is uploaded or synced to any server. ✔️ Privacy-first: Full addresses stay on your machine, so a token inside a link never goes out. Read-only as well: nothing is added, renamed, moved or deleted. ✔️ Search by what you remember: Use /bookmarks and give the clues you still have. Roughly what it was about, a word in the title, the folder it went into, when you saved it. ✔️ No folder digging: Find one link among thousands without clicking through a tree. ✔️ A clear division: Bookmark search doesn't read page contents. To ask what a page actually says, save it to your Knowledge Base and ask with @. 8️⃣ Extract web data ✔️ Say what you need: Point at a list page and name the fields. You get a structured table. ✔️ Export it: XLSX and JSON, ready to use. ✔️ Save it as a template: Extractions save like any other action and rerun free on pages with the same structure. 9️⃣ Skills, Apps & MCP ✔️ Skills: Save the flows and prompts you reuse. Trigger one yourself, or let Agent S6 call it. ✔️ Slash commands: Type /summarize, /extract and the rest straight into the input box. ✔️ Apps: Connect the external tools and services you already use so it can finish work across sites. ✔️ MCP: Install preset MCP servers, or add your own with JSON or a URL. Agent S6 only calls the tools you enabled. 🚀 Getting started 🔸 Click "Add to Chrome" and pin it to your toolbar. 🔸 Sign in, or enter your own API key. 🔸 Press Alt+Shift+S (Command+Shift+S on Mac). 🔸 Start using it. ❓ FAQ 📌 Do I need a ChatGPT or OpenAI account? No. Create a free account on our website. 📌 ChatGPT isn't available where I live. Does this work? Yes, we're available everywhere. 📌 Is it free? New users get 200,000 tokens. Bring your own API key and most features work too. Consider Pro when your usage grows. 📌 Does what I save to the Knowledge Base get uploaded? No. The full text and the search index stay on your machine, and nothing is uploaded to a server to be indexed. Search runs locally, and only the passages needed to answer reach the model. With your own API key, requests go from your browser straight to the model provider and never touch our servers. 📌 Does bookmark search expose my private data? Your bookmarks stay in your own browser, with nothing uploaded or synced. Full addresses stay on your machine, so a token inside a link never goes out. It is read-only as well: it never adds, renames, moves or deletes a bookmark. 📌 Will it mess with my tabs? No. You see the plan before anything happens, it acts only after you confirm, you can undo it, and closed tabs come back. 📪 Contact us Questions or suggestions: 💌 support@starizon.ai