The inbox is now your Agents’ work feed
The workspace home reads like a feed of what your Agents produced — every unreviewed document across the workspace, newest first, cleared by reviewing.
Feature-level changes as they ship. One entry per feature — what it does, why it exists, and how to use it.
This first batch is a snapshot: everything inside the new Superself today, for readers coming back after the rebuild around SharedContext.
The workspace home reads like a feed of what your Agents produced — every unreviewed document across the workspace, newest first, cleared by reviewing.
Instead of confirming a document, leave feedback on it — the Agent addresses it in its next session and the document returns to your review queue.
A new /use-cases hub with seven articles, each starting from a real pain — lost context, repeated explanations, unreviewed agent output — and showing the working setup.
Every documentation surface — docs, drops, updates, use cases — now serves in Korean under /ko, with language-aware links throughout.
The document panel gets back/forward navigation with its own local history, projects rename inline from the breadcrumb, and the panel header reads cleaner.
The /install page leads with a copyable setup prompt for agents that cannot fetch URLs, the Claude Code plugin installs as fxy, and connected Agents can survey the whole workspace with workspace_map.
The workspace home is now a status inbox — approval requests, unconfirmed documents, and active scopes surface in one place.
A desktop-style shell — folder-first navigation, a resizable document panel, and ⌘K search across your whole workspace.
index.md is a folder’s first screen — rules, decisions, and roadmap as prose, with the document list generated automatically.
Every document shows whether you have reviewed it — Agents treat unconfirmed documents as proposals, never as settled decisions.
Every write is a ledger commit — browse a document’s timeline, open it as it was at any point, and revert when a change is wrong.
A live session map shows what each Agent session is doing; session logs and briefs let the next session start where the last one stopped.
HTML visualization documents and image attachments are first-class knowledge — moodboards, canvases, and charts live next to your markdown.
Archive finished projects out of the way, reactivate them when they come back, and restore anything from trash — nothing is lost by tidying up.
ChatGPT, Claude, Gemini, Grok — any client that speaks authenticated remote MCP connects through a standard OAuth consent you control.
Connect from ChatGPT, Claude, Grok, or any AI that can use MCP tools.
Get connected