Kursübersicht
AcademyModul 7 · Controls and connections
The activity center: what ran, what's next
Das lernst du
- You can read the Activity panel and tell live jobs, history rows, and pending offers apart.
- You can act on lifecycle advisor offers, or politely defer them with "Later" and "Don't ask again".
- You can turn a task-suggestion badge above the composer into a finished deliverable with one click.
- You can follow the agent's in-chat activity rows and safely stop a background command or a goal.
You save a note and go make coffee. Consilience keeps working while you're away. It pulls the entities and connections out of the note, names an untitled document to fit its content, refreshes the topic map, and lines up a few suggestions. Where did all of that go? Into the Activity center, the bell in the title bar.
This lesson walks that whole trail and answers two questions: what just happened, and what should happen next.
The bell and the Activity panel
The bell sits in the title-bar toolbar at the top right, next to the Knowledge graph and Terminal buttons. Its tooltip reads "Activity". Click it and a 320px dropdown titled "Activity" opens.
The header has a "Clear" button (more on what it clears later), and before anything happens you see "No recent activity". The bell is hidden on the Welcome screen, since without a workspace there's no feed. It has no keyboard shortcut.
- Unread count badge: a blue pill with the number of rows you haven't read. Past nine it reads "9+".
- Live pulse dot: appears when you have no unread rows but something is still running in the background. A quiet "something's happening" signal.
- The panel can open itself in three cases: when you open a workspace with an offer waiting, when a knowledge-graph build starts, and when an agent action in your active tab hits a wall you can fix. The build case opens it so the progress bar is right there. Mid-session re-checks only update the bell.
What "In progress" shows
| Row | What you see |
|---|---|
| Agent turn | Document title plus "AI editing". Click it to jump to the tab where the agent is working. |
| Single-note extraction | File name plus "Extracting" and a spinner. One row per note being tagged. |
| Bulk extraction | "Extracting knowledge" with a done/total counter and a progress bar. One row stands in for the whole batch. |
| Credit-paused note | "Paused until your limit resets", with a still, dim ring instead of a spinner, so it reads as waiting. |
| Vocabulary tuning | "Teach the AI your field" plus "Tuning vocabulary" while the tune call runs. |
| Suggestion generation | "Preparing suggestions…" while the task-suggestion pool is built. |
Below "In progress" comes the history. Each record row has a colored kind-dot, a title, and a short relative time ("just now", then 5m / 2h / 3d). Some rows add a detail line: "Extracted", "Extracted {n} documents", "Document renamed", or "Topic map updated" with "{n} themes found across your notes."
Unread rows get an accent tint and a bold title, and clicking one marks it read and opens the note. For an auto-change it also opens the conversation with the diff card, so you can review or revert. Repeats of the same kind for the same file coalesce into one row that refreshes in place, so an "Extracted" on every save doesn't pile up.
The lifecycle advisor's offers
The lifecycle advisor watches your workspace and posts offers. The six kinds are stages of one pipeline, so the panel highlights just the most-upstream unmet stage as your recommended next step. Other pending offers fold under a "{n} more" toggle, so the panel never becomes a wall of competing buttons. Here they are in priority order.
| Offer | Accept button | Can you defer it? |
|---|---|---|
| Capability blocked, e.g. "Sign-in required", "Connection needed", "Web search is off" | "Sign in" / "Open Connections" / "Open Settings" | No controls. Session-only, gone after a restart. |
| "Extraction was interrupted" | "Resume" | No snooze. "Discard" is the only way out. |
| "Extraction needs a retry" | "Retry all" | "Later" / "Don't ask again" |
| "Build this workspace's knowledge graph?" | "Build" | "Later" / "Don't ask again" |
| "Notes changed outside the app" | "Re-extract" | "Later" / "Don't ask again" |
| "Tune domain vocabulary?" | "Tune" | "Later" / "Don't ask again" |
Kurzer Check
Which offer card cannot be snoozed with "Later"?
Task suggestions above the composer
Open an empty new chat and a badge strip appears above the composer, headed "Make from this workspace". Once the knowledge graph is analyzed, the app builds a pool of content-grounded suggestions and shows up to 5 badges. At most 3 are deliverables (a document, plan, comparison, itinerary, review, or summary); the rest are structural nudges.
Each badge carries a category icon, a title, a "why this" hint, and a "Make" arrow that auto-sends a ready-made prompt to the agent. The × dismisses that suggestion for good in this workspace. The "…" button regenerates the set, toggles auto-refill so the pool tops itself up as you work through it, or hides the strip for 7 days.
- "What could '{theme}' connect to?" for an isolated theme, with the hint "These notes sit apart from the rest of your workspace".
- "What runs through '{label}'?" for a hub concept, hinted as "It's the hinge joining several of your themes".
- "What connects '{a}' and '{b}'?" for two big themes that have no bridge between them yet.
- Isolate an entity on the empty-chat graph backdrop and the strip pivots to "Continue from '{label}'", offering "Everything you know about '{label}'": one page, every fact cited to its source note.
Kurzer Check
You open a chat that already has three messages in it. Where are the task-suggestion badges?
Activity inside the chat thread
Goal and tool activity rows live inside the conversation. They never surface in the bell panel. The standard system line shows the agent's tool calls and thinking with a live timer, so "Thinking for 12s" becomes "Thought for 12s" once the step lands.
Each row carries an inline args summary and a chevron that expands the output, diff, or reasoning. File-write rows arrive already expanded, so you can review the diff right there. A running background command carries a "Stop background command" button.
- A thread with an active
/goalpins a goal row at its tail. It reads "Goal: {condition}" plus a status: "Working… {n}m so far", "achieved", or a line saying it tried for a while and summarizing what got done. - Its action links are "Keep going" (resume after a hard stop) and "Dismiss". While a run is in flight, "Dismiss" turns into a two-step "Stop" then "Stop this?" confirm, so a stray click can't kill an autonomous run.
- When the agent resumes a turn on its own after a background command finishes, that turn opens with a thin status line saying the background task finished and it's continuing.
Housekeeping: how the feed keeps itself tidy
- The feed is stored per workspace, so switching workspaces gives you a clean, focused view of exactly the one you're in.
- The feed tidies itself: older rows retire after about a month, so the panel always reads as a snapshot of what's recent.
- A capability-blocked card lives for the current session, and the agent raises it again the instant it hits the same wall.
- The advisor is always watching. It re-checks what to offer on workspace open, on file changes, on a timer, on window focus, and after every extraction batch. File changes even include git pulls and edits made in other editors.
- activity
- A job running right now, shown live under "In progress". It's never saved to history.
- notification
- A persisted feed row: something that happened, or a question the app is asking.
- record row
- Read-only history. Clicking it opens the note (and any diff conversation) and marks it read.
- actionable offer
- A card with an accept button. It stays in the feed until you answer it, and it survives "Clear".
- lifecycle advisor
- The watcher that posts the six offer kinds. It highlights only the most-upstream unmet stage.
- coalescing
- Repeats of the same kind for the same file refresh one row instead of stacking.
Now that you can read the feed, the next lesson covers the plans and credits that fuel all this background work, including that "Paused until your limit resets" row.