Browser Nodes
A Browser Node is a real webpage on the Architect canvas — not a screenshot, not a search snippet. Each node hosts a live WebKit page you can scroll, click, and resize. Put several on one board, wire their results into the rest of the workflow, and reopen the whole desk later from Finder.
Canoryn is not trying to replace Safari as your everyday browser. It makes the web a live surface inside a workflow so you and the agent work on the same pages.
What you can do on the node
| You | The workflow / agent |
|---|---|
| Open several Browser Nodes side by side | Load URLs into specific nodes |
| Resize nodes to match how you read | Extract text / HTML from the rendered page |
| Hover the page to scroll and click (like a normal browser) | Read the page map, then click / fill fields |
| Switch browser profile (Default, Work, Personal, Incognito) | Act in that profile’s session (cookies stay on this Mac) |
| Use canvas Pan tool to move around the board without fighting the page | Pass live outputs into AI, documents, email, and other steps |
| Open in Safari from the node chrome when you want a full browser | Snapshot the page when you need a still |
Pan vs page control
- Hover the webpage — scroll and click go to the site; drag the chrome (title / address bar) to move the node.
- Canvas Pan tool — move the whole board; page hit-testing pauses so gestures don’t fight.
- Address bar, back / forward, reload — normal navigation on that node’s page.
Why this beats “generic search”
Search APIs and raw HTTP fetches miss what you actually use:
- JS-rendered docs and dashboards
- Pages you’re logged into (with a named profile)
- Several sources open at once in a layout you chose
A research or reading board can keep GitHub + an RFC + a design site (or whatever you live in) open together. The agent works on those sessions — not a disposable crawl of the open web.
Profiles (sessions stay on your Mac)
Each Browser Node can use a profile:
- Default
- Work
- Personal
- Incognito
Cookies and login state live in that profile on your Mac. They are not written into the .cryn workflow file, so sharing a workflow does not leak your session.
Multiple live browsers on one board
Typical patterns:
- Dev reading — API docs, GitHub, issue tracker, local dashboard
- Design research — competitor sites, galleries, your Figma-adjacent tabs as live pages
- Ongoing research — official product pages side by side with a markdown report node
Wire browser outputs (title, URL, extracted text) into AI or document nodes so the board isn’t just tiles — it’s a runnable pipeline. See Research boards.
Reopen the whole session from Finder
Workflows save as .cryn files. Double-click one from Finder (or open it in Canoryn) to bring back the canvas layout — including Browser Nodes and their URLs — so your desk returns without rebuilding from chat history.
Expand collapsed groups if a page looks blank after restore; the live surface reattaches when the node is visible again.
What the agent can do (short)
Against a Browser Node (or a shared agent session), workflows and chat can:
| Action | Purpose |
|---|---|
browser.load | Navigate and wait for the page |
browser.read_page | Map headings / interactive elements (refs) |
browser.click / browser.fill | Act on refs or selectors |
browser.extract_text / extract_html / extract_structured | Pull from the rendered DOM |
browser.snapshot | Capture a PNG |
browser.navigate | Back / forward / reload |
browser.list_sessions | See open sessions |
browser.run_js | Advanced scripting (consent-gated) |
Prefer refs from browser.read_page over guessed CSS. Consent still applies for sensitive steps (fill, run JS, etc.).
Full catalog: Web & Information.
Tips
- Start from Architect Copilot or Chat: ask for a board with N browser nodes for the sites you care about.
- Keep official / logged-in sources on the board so synthesis stays grounded.
- Use Work vs Personal profiles so agents don’t mix contexts.
- Canoryn may snapshot or pause off-screen pages to keep the canvas smooth — hover or activate a node to bring the live view back.