Everything you need to know about using TaskFlow.
How do I run a task?
Go to the Dashboard, type your task in the input bar at the top, set a timeout, and click Run. The agent plans the task, opens a browser, and shows you everything live.
What is the timeout setting?
It controls the maximum time the agent can spend on your task. Set it higher for complex multi-step tasks. Default is 5 minutes.
What does Live / Idle mean?
Live means you are connected via WebSocket and receiving real-time updates. Idle means no task is currently streaming.
What sites can it access?
Any publicly accessible website — news sites (BBC, Reuters, HN), search engines (Google, DuckDuckGo), Wikipedia, GitHub, Reddit public pages, weather, sports, e-commerce product pages, and more.
Can it log into my accounts?
No. The agent cannot log into any accounts. It has no stored credentials and will not submit login forms.
What about paywalled sites?
It can reach them but cannot read the content behind the paywall. Sites like NYT, WSJ, and FT will only show teaser content.
Can it handle CAPTCHAs?
No. CAPTCHAs will stop the agent. If a site detects it as a bot, the task will fail with an error.
What is Summary Mode?
When on, the AI writes a 2-4 sentence plain-English summary of what it found after completing your task. When off, it returns just the raw extracted data.
What is the Live Browser toggle?
Shows or hides the live browser popup. Even when hidden, the agent is still running — screenshots are just not displayed.
What is Stream mode in chat?
When enabled, the AI response types out token by token in real time. When disabled, it waits for the full response before displaying.
Task failed with "element not found"
The page structure may have changed or the site uses bot detection. Try a different site or rephrase your task with a more specific URL.
Browser shows as Offline
The Playwright browser on the server failed to start. This usually fixes itself on the next server restart. Check the status indicator in the side panel.
Task stuck in Planning
The NVIDIA model is taking longer than usual. The planner model (49B) can take 10-20 seconds. Wait a bit before cancelling.
"Failed to fetch" error
The frontend lost connection to the backend. Refresh the page. If persistent, the backend may be restarting (Render free tier sleeps after inactivity).