What to try first

Type anything — the agent answers, searches the web, and runs tools on its own. Try: “What time is it in Tokyo?”

Use the mode picker in the composer: Plan outlines steps first; Ask first confirms before every tool.

Type /research followed by a question. A multi-agent fleet searches dozens of sources, cross-verifies claims, and writes a cited report.

Research runs survive reloads — reopen the chat to watch them continue.

“Generate an image of a lighthouse at dusk”, or “create a PDF report about renewable energy”. Outputs land in the Files panel with versions.

Export any conversation from the header — Markdown or PDF.

Tap the mic in the composer to dictate. Every assistant message has a read-aloud button (copy row → speaker icon).

Pick a voice and speed in Settings → Personalization → Voice.

Say “create a skill that…” in any chat, or build one in Settings → Skills with a live preview of its SKILL.md source.

Skills activate automatically when their trigger words appear.

Settings → Connectors: Google Drive/Gmail/Calendar/Sheets, GitHub, Notion, Slack and more — every key is validated on save and testable.

More apps live in the Marketplace — connect them with your own OAuth app credentials.

Create a project, add custom instructions, upload knowledge files, and chats stay grounded in that context with cited retrieval.

Projects get a first-class section in the sidebar.

Settings → Tools sets what the agent may run on its own; the Activity panel shows every tool call with a clickable tool count.

Type / for the full command palette — every feature is reachable from it.