Chat & agent features.

Approvals

• Reads are free, searching mail, listing events, checking balances run without prompts.
• Writes ask first, an approval sheet shows the exact tool and arguments before anything sends, posts, writes, or transacts.
• Always allow: tick it on the sheet to skip future prompts for that specific tool. Revoke anytime in Settings → Permissions. Money-moving tools (solana_send, solana_swap) and full memory overwrites can never be always-allowed.

Plans & the run timeline

For multi-step work the model declares a plan checklist and checks steps off live. Tool calls group into a timeline card, each expands to show its result. If the model repeats the exact same call, a loop-guard blocks it and tells it to change course.

Background runs

Runs continue when you navigate away, switch conversations, browse Settings, start another chat in parallel. Running conversations show a spinner in the sidebar, and you get a browser notification if a run finishes while the tab is hidden.

Artifacts

Substantial outputs (reports, code files, tables, pages) arrive as artifact cards, preview rendered Markdown / sandboxed HTML / CSV tables, copy, or download, instead of scrolling a giant chat bubble.

Composer commands

• /, opens the command menu (built-ins + your skills).
• /compact, summarizes older messages into one dense context block (keeps recent turns). Great for long chats and small on-device contexts.
• Attach images with the paperclip (vision-capable models), or dictate with the mic.

Memory

Xome keeps an editable memory.md (Settings → Memory) that's shown to the model each turn. It saves durable facts proactively, every write is gated by approval, and an Undo toast appears for a few seconds after each change.

Suggestions, forking, history

• After a turn, tappable follow-up chips may appear (cloud models only; best-effort).
• In History: pin, archive (hides from the sidebar), fork a conversation, search everything, and see per-conversation token usage.