Guide
Agent panel
The agent panel is where you build. Open it with Ctrl+. (or ⌘+.on macOS), describe what you want, and watch the agent create projects, write site code, set up payments, and wire workflows — every action shows up live so you can follow what's happening.
Chronological timeline
As the agent works it streams a chronological mix of text and tool calls. You see "I'll start by creating three subscription tiers" first, then the three polar_create_product cards, then "All three created, now let me build the pricing page", then the write_filecard, and so on. The order matches the agent's actual decision flow — no "tools at the top, text at the bottom" reshuffling.
A short row of dots underneath the active assistant bubble tells you the agent is still thinking between actions. When the dots disappear the run is finished.
Conversation history & persistence
Each conversation is saved to your workspace. Refreshing the page or closing the browser doesn't lose the chat — when you reopen the panel, your last conversation is right where you left it.
Tap the history clock icon in the panel header to see every past conversation. Click any one to jump in, rename it with the pencil icon, or export it as Markdown for sharing. The + icon starts a fresh conversation when you want to switch topics.
Slash commands
Type these in the input to control the session quickly:
/help— show available commands/clear— start a new conversation/context— show what the agent currently knows about your project/export— download this conversation as Markdown
Attaching files
Click the paperclip 📎 to attach any text-based file — HTML, CSS, JS, Markdown, CSV, JSON, YAML, etc. — and it gets inlined into your message as a code block. Useful when you want the agent to reference an existing template, import a spreadsheet of products, or paste in a long copy deck without losing formatting.
Monthly quota meter
At the bottom of the panel you'll see Agent messages this month: X / Y. That's how many model conversations you've had against your tier's monthly limit (Free: 50, Starter: 200, Pro: 750, Business: 3,000, Enterprise: unlimited). The counter ticks down after each completed message so you always know where you stand.
Quota resets the 1st of each calendar month (UTC). If you run out mid-month, upgrade your tier from Pricing and the new limit applies immediately.
What the agent can do for you
The agent has direct, action-taking tools (not just text generation). A non-exhaustive list:
- Site building —
create_project,write_file,read_file,set_project_status(publish/unpublish) - Database —
provision_database,create_table,add_column,insert_row,list_tables - Payments —
polar_create_product,polar_list_products,polar_archive_product(see the payments guide) - Site assistant —
create_assistant,update_assistant,add_assistant_knowledge - Integrations —
connect_integration,list_project_integrations,test_integration - Workflows —
create_workflow,update_workflow,get_workflow,delete_workflow. The agent builds the same automations you build on the visual canvas — a trigger (cron, webhook, form_submit, db_change, analytics_threshold, inbound_email) plus action and condition nodes, including branching (if/else) graphs. Anything it builds opens and runs identically in the Workflows canvas. - Research —
web_search(live web search) andfetch_url(read a public page). The agent uses these to research content for your site, read a URL you paste, or look up an unfamiliar provider error before retrying. Only public web addresses can be fetched — internal and private hosts are blocked.
You don't call any of these directly. Just describe what you want in words and the agent picks the right tools.
How the agent handles errors
When a tool call fails (a provider rejects a value, a resource conflicts, a service is briefly down), the agent doesn't stop silently. It reads the error, and then either fixes its input and retries, inspects current state with a read-only tool, researches the error message on the web, or asks you one specific question with concrete options. Every failure ends with a plain-language explanation in the chat — if you ever see a failed tool call in the timeline, the agent's next message tells you what happened and what it plans to do about it.