Following the Advisor's work

Activity record

While the Advisor works, its activity appears above the answer and grows as the work progresses. It is a visible record of what the Advisor does during the turn:

• Reasoning — rows with a short heading taken from the Advisor's reasoning. Selecting a row expands its text. Reasoning is a working note, not part of the answer, but it shows why the Advisor reached a conclusion.
• Tool chips — every search or action appears as a chip with the icon of its skill family and a label such as Searching Polish legal acts or Searching tenant corpus. A spinner marks a running call and ! a failed one. Consecutive calls are grouped under Using N tools… while they run and N tools used afterwards, with the number of failed calls when there are any.
• Web searches — while a web search runs, a chip shows the query. Pages that were found or read appear as chips with the site icon and title; they open in a new tab.
• Working notes — short progress notes written by the Advisor, in ordinary text and with citation chips where sources are cited.
• Research agents — buttons for the research agents the Advisor has started. See Research agents.

A note or plan marked Confidential draws on the organization's non-public material and should not be shared outside the organization.

When the turn ends, the activity folds into a single line, Worked for …, with the time the work took. Selecting the line expands the full record again; icons next to it identify the research agents of that turn, with their names shown on hover.

Output style

How much of the activity is visible while the Advisor works is set in Settings → Preferences → Advisor output style:

• Verbose — expands reasoning and activity as it happens.
• Compact — shows summaries and grouped tool activity. This is the default.
• Hidden — shows only Working… until the answer is ready.

Completed work stays collapsed in every style. See Preferences and notifications.

Plan

For tasks with several steps, the Advisor publishes a plan. The Plan control appears at the right edge of the screen (on narrow screens, a button with a list icon); a dot on it pulses while a step is in progress. The plan opens in place of the conversation list and shows:

• the question the plan answers;
• a status — In progress, Paused, Stopped or Not finished;
• the steps, each with its status — Next, In progress, Completed, Blocked or Skipped;
• Approach — the Advisor's explanation of how it works on the question, shown as Updated approach after the task changed.

A step that is still open when the turn ends is marked Unfinished or Stopped, and the panel states that not every step was finished. The plan is the Advisor's own account of its progress; a completed step is not a separate verification.

Research agents

Broad questions can be divided among research agents — separate researchers that each take on one part of the task. Each agent appears in the activity as a button with an icon for the area it mostly reads and the title it was given, and as an icon at the right edge of the screen.

Selecting an agent opens its agent session beside the conversation; the divider between them can be dragged to resize both. The session shows:

• the Assignment from the Advisor;
• the agent's own activity, including messages it received from the Advisor;
• the agent's report, with citation chips.

Close agent session closes the split view. An agent marked Earlier scope worked on the task before a note changed it. The final answer draws on the agents' findings and cites the sources.

Adding a note while the Advisor works

A running task can be steered without stopping it. While the Advisor works, the message field shows Add a note while the Advisor works... (Add a note... on narrow screens). Text sent with the send button (Send note to the running turn) goes to the running task instead of starting a new question.

Only the text of a note is sent. Attached files and context tags stay in the composer for the next message, and the composer says so.

Each note appears in the conversation with its delivery status:

Once a note is sent, the work done before it folds into Work before your update, the note appears with its status, and the Advisor's further work continues below it. A note that is too long stays in the field with a message suggesting to shorten it or to send it as a new message after the Advisor finishes. A note sent just after the task ended also returns to the field, with an explanation.

Clarifying questions

When a question is ambiguous — two bills with similar names, two MPs with the same surname, an unspecified Sejm term — the Advisor may ask before it continues. The activity shows Asking clarifying question, and the question appears in the conversation with its options: choice cards, checkboxes, a confirmation pair or a text field. In MCP terms, this is an elicitation request.

• Continue sends the answer, which then appears in the conversation as a reply.
• Skip declines the question; the conversation records You skipped this question.

Answer the question with its own controls. Text typed in the message field at this time is sent as a note to the running task, not as the answer to the question.

Stopping

The stop button (Stop research) next to the send button ends the running task. What the Advisor found so far stays in the conversation, with one of the notices:

• "Stopped. What the Advisor found so far is kept above."
• "Stopped before the Advisor produced an answer."

Plan steps that were still open are marked Stopped. If the stop request does not go through, a message asks to press stop again.

Interruptions and partial answers

• Partial answer — the Advisor could not cover the whole question in one turn. The notice begins with Partial answer. and suggests a narrower follow-up, for example limited to one sitting, document, person or period. The findings so far remain above the notice.
• Lost connection to the task — "The Advisor lost its connection to this task before finishing." What it had saved is kept; a new message continues the work.
• Dropped connection — "The connection to this task dropped." The Advisor may still be working, and nothing above is lost. Stop and notes stay available, but a new question cannot be sent until the turn is resolved. Reconnect reattaches to the task; Stop managing this turn releases the conversation for new questions without stopping the task.
• "Hit a temporary snag — retrying…" — a temporary problem that the Advisor retries on its own.
• "The Advisor hit a problem while finishing this answer. Please try again." — use Regenerate response below the answer or ask again.