Browsie — Chrome Web Store creatives (internal)

How to capture each tile:
1. Open DevTools (F12 or Ctrl/Cmd + Shift + I).
2. Right-click the tile in the page → "Inspect".
3. In Elements, find the <div class="tile tile-shot"> (or tile-small / tile-marquee) node.
4. Right-click that node → "Capture node screenshot".
5. PNG drops at the exact required dimensions. Upload to your Chrome Web Store listing.

Required minimum to publish: 128×128 store icon (already in /icons/icon128.png) + at least one screenshot at 1280×800. The promo tiles below are optional — used for featured placement and category browsing.

Screenshot 1280 × 800 required · ≥1 · max 5

[ BROWSIE · v0.1.0 ]

live preview

An autonomous browser agent

Set the goal.
We browse.

Type a goal. Browsie reads the page, plans subgoals, and navigates, fills, clicks, and extracts until it's done. Runs against your local LLM. Local-only. No telemetry. Free.

PLAN·FILL·CLICK·EXTRACT·SCHEDULE

local-first · no telemetry · free

v0.1.0 · MV3 side panel · designed for power users

To capture this: DevTools → Inspect → right-click the <div class="tile tile-shot"> node → "Capture node screenshot". PNG drops at exactly 1280×800. Re-style the embedded mock popup to make 5 distinct screenshots (Goal / Plan / Tasks / Library / Run log).

// Chrome Web Store listing copy

Name max 50

Category pick one

Short description max 132 chars

Autonomous browser agent. Type a goal — Browsie navigates, fills, clicks, and extracts. Runs on your local LLM. Local-only. Free.

Detailed description max 16,000 chars

Browsie is an autonomous browser agent that turns a one-line goal into actual browser work. Type "find the cheapest non-stop flight to Lisbon in October and copy the link" — Browsie reads the page, plans subgoals, and navigates, fills, clicks, and extracts until the goal is done.

Built for power users who want a focused cockpit, not a chat window.

▼ THE LOOP — three small roles, one goal
1. Director reads your goal once and decomposes it into 1–6 ordered subgoals, each with a clear "done when" predicate.
2. Actor takes one tool action per turn — fill, press_key, click, navigate, extract, scroll, wait_for, set_focus, ask_human, or done — based on a compressed accessibility-tree snapshot of the active tab.
3. Reflector checks whether the current subgoal is satisfied after every action, advances the checklist, and ends the run when everything's ticked.

▼ ELEVEN VERBS — that's the entire surface
Browsie's actor only emits one of: navigate, fill, press_key, click, scroll, extract, extract_records, wait_for, set_focus, ask_human, done. No code execution. No file write. No shell. The agent works through the page like a human does.

▼ VISIBLE PLAN — see what's about to happen
The subgoal checklist sits in the side panel. As Browsie works, items tick green. When it's stuck, it tells you which subgoal — not just "I failed."

▼ LIVE STEP LOG — every action with a one-line thought
Color-coded by outcome (action / ok / error / done / paused). Scroll back through what it did, why, and what came back from the page. Each run is saved to a local history.

▼ TYPED EXTRACTION — strings, lists, CSV
Ask for one field, get a string. Ask for a list of records, get a typed table. Copy-as-CSV in one click. The accessibility-tree snapshot keeps the agent honest about what's actually rendered.

▼ SAVED TASKS + SCHEDULING
Promote any goal to a reusable task. Schedule it to run on an interval, or once at a future time, via Chrome's alarms API. Re-run by name. Morning routines write themselves.

▼ THE LIBRARY — your real data, sent on demand
A local profile (name, email, address, freeform preferences) saved to chrome.storage.local. Browsie sends those values to the LLM only when the form actually needs them. Forms get filled with your real data, not placeholders.

▼ SAFETY RAILS — built in, not bolted on
- Hard line: refuses payments, deletes, and credential changes by default via the system prompt. Reversible actions (chat sends, comment posts, search submits) go through, with a graceful ask_human pause when the agent lands on a binding button.
- Click-loop, no-effect-streak, and step-cap detectors auto-pause when the agent is spinning.
- Single-tab. Rides your existing session — never initiates logins.
- ask_human verb: when ambiguous, the agent stops and asks.

▼ LOCAL LLM BY DEFAULT
Configured for LMStudio, Ollama, or any OpenAI-compatible endpoint. The default points at localhost. Page snapshots travel exactly as far as your inference server. Bring your own model.

▼ PRIVACY POSTURE
- All settings, library, tasks, and run history live in chrome.storage.local. Never synced.
- No analytics, no telemetry, no third-party scripts, no contact-server endpoints.
- Zero outbound calls except (a) your configured LLM endpoint and (b) the page the agent is currently working on.

▼ PERMISSIONS
- storage: save settings, library, tasks, run history locally
- sidePanel: render the cockpit in Chrome's side panel
- scripting: run on the active tab to read the AX tree and execute fill / click / scroll
- tabs / activeTab: read the current tab URL/title for plan grounding
- alarms: run scheduled saved tasks
- host: <all_urls> (you choose the page; Browsie only acts when you click Execute)

Browsie is an independent open-source utility. Free. v0.1.0. Read the FAQ and full Privacy Policy at the project website.

Homepage URL

Suggested search keywords internal hint, not a Web Store field

browser agent, autonomous browser, ai browser extension, chrome agent, llm browser automation, local llm chrome, ollama chrome extension, lmstudio chrome, ai web automation, gpt browser agent, side panel agent, browser copilot, autofill agent, scrape agent, web automation chrome

// BROWSIE · Chrome Web Store creatives

Set the goal.
We browse.

Browsie.

// Chrome Web Store listing copy

// BROWSIE · Chrome Web Store creatives

Set the goal.We browse.

Browsie.

// Chrome Web Store listing copy

Set the goal.
We browse.