Skip to content

Sidebar

Item What it does
New chat Starts a Quick Chat in the current workspace
Marketplace Discover (Synoec, MCP servers from public catalogs, Composio) and Installed; see Marketplace
Local AI Ollama and Bonsai 2 on your own hardware; shown only when there is something to manage. See Local AI
Context The project knowledge filesystem; see Context filesystem
Usage Spend, limits and budgets; see Usage and pricing
Search threads Finds chats by name across all projects, including archived ones
Pinned Pinned chats, shown above Projects
Projects Folders you opened with Open project folder. Eight show before Show more; the project in use, or one with work running, stays visible past the fold
Chats Quick Chats with no project folder, eight before Show more; New Quick Chat starts one with the same welcome as a new project chat

Chats are ordered by their latest activity. A cold start holds each list’s places with skeleton rows until the first read lands. A failed read of the chat list never replaces it: chats already shown stay, and the read repeats on its own, with a new support host if the last one went away. A host that ends without being asked is recorded in host-exits.log in the app’s data folder, with the reason its guardian saw.

Each chat row has a status dot: it pulses in the accent color while the chat works, turns yellow when it needs your attention, and shows green or red for the last turn’s outcome until you open the chat. A new chat is named from your first message, then from a short generated title; a name you set with Rename… is kept.

Hover a row, or use Conversation actions in the chat header, for Rename…, Pin chat / Unpin chat and Delete…. The header menu also has Read complete saved history. Delete offers Archive (kept in search; Restore chat brings it back) or Purge…, which permanently deletes the chat’s history after a second confirmation.

Right-click a project for Repository (see Repository) or Remove project…. Removing a project keeps its folder and memory on disk.

A chat can hand a separate task to a subthread: an ordinary chat in the same project, listed under the chat that started it. Ask for one in your own words (“do the docs in a separate thread”), or the agent may suggest one.

  1. A card in the turn asks New thread or Do it here.
  2. For a new thread it then asks which model to use: the suggested one, or another from Other. Several subthreads are asked on one card, page by page.
  3. The subthread appears under its parent, joined by a thin rail. The parent’s dot carries the status of its subthreads, and the rail pulses down to the one that is working. The parent’s chevron (Hide subthreads / Show subthreads) folds them.

A subthread answers its parent plainly, without the Persona’s voice. When it finishes, the parent gets one short notice and carries on only with work you already asked for. Notices and messages between threads show as labelled notes (Specialist update, Message from another thread, Brief from its parent thread), never as your words, and never count as your approval.

A subthread that changes files works in its own Git worktree on a new branch with the Branch prefix from Settings → Git; hover its row to see the branch. Your checkout and branch are never touched, and nothing is merged or pushed for you. Threads that edit files in Settings → Git can instead Ask each time (the writer waits for Start thread) or Share this checkout (writers run one at a time); a folder that is not a Git repository always shares. To review a writer, hover its row and choose Compare, merge or archive its branch: the dialog shows its changes against your checkout, then Merge… merges its branch into your checked-out branch (only when both checkouts are clean) and Archive… archives it.