Interface and advisor workflow

Follow the interactive lessons to practice questions and context, projects and sharing, and plans and research.

Query composition area

The message input panel is located at the bottom of the screen:

• Suggestion lightbulb — in a new conversation, opens a list of recent topics. Focusing the empty text field also shows the list. An alert-related event appears first with the alert name in blue when one is available. Selecting a topic fills the draft and adds a source chip; review the draft before sending it.
• Text field — supports multi-line text input (Shift + Enter). The Enter key submits the query.
• + menu — appears in the expanded input controls. Repository opens the Repository and public corpus; @ opens the same search. Upload files opens the file upload window with an access selector and attaches successful uploads to the draft. Thinking depth sets Automatic, Surface, Balanced or Deep for the next message. Preferences opens the main Settings → Preferences page at the Advisor section.
• Space selector — next to +, selects a private or shared project for the conversation or creates one. A project groups conversations without automatically adding other conversations to the question's context.
• Send button — sends your question. While the Advisor works, the arrow sends your text as a note; the square button stops generation.

Automatic is the default when no thinking preference has been saved. Advisor chooses a level for each request based on the task and recent questions. A saved manual preference remains in effect; a choice in the message menu applies to the next message. After sending, the control returns to the saved preference. If automatic selection is unavailable, Advisor continues with its standard level, including deeper reasoning for broad research requests.

Analysis document and plan

The Advisor chooses research and visualizations to suit the question. For substantial analysis, it creates an Analysis document beside the conversation. The answer presents the main findings and an Open analysis document link. Short questions may receive an answer alone.

The document combines formatted text, tables, highlighted findings, sources, charts and research results. It develops while the Advisor works. A follow-up such as “add a comparison with last year” updates the same document. Bar and line charts, relationship graphs and sourced image galleries appear within its content. View data expands chart values; Sources expands links to the underlying material.

Chart sources can open the study used for the calculation, a source inside Persate, or an external page. Research links open the saved results in the analysis document. A request for a chart produces a rendered visualization; source code is not a substitute for an unavailable chart.

The document starts with the analysis title and, when useful, a short description. Its content updates as work continues. The view label, version selector, status and date are omitted from the document itself. The Analysis document tab and shortcuts use a document icon.

The saved document and research remain available after reopening the conversation, independently of the loaded chat messages. Load earlier research retrieves more studies. A link to an older chart reveals that chart while retaining the current document content. Stopping work preserves the content already produced; analysis limitations are described within the text.

The + menu retains Plan: it prepares an approach and waits for Execute plan in the Plan tab. Selecting the mode again or dismissing its label returns to an ordinary conversation. An ordinary conversation can also have a working plan, which the Advisor carries out without this separate approval step. Plan and Analysis document are available under Views in the additional Advisor workspace.

Research in the document

For custom criteria, Advisor chooses subdued colors that match each option’s meaning. Increasingly negative options can use progressively stronger shades of red. The palette is saved with the study and stays consistent across the legend, tiles and comparisons. Older studies retain their previous palette.

Include a total count in the request to limit a collection, for example: “Analyze the latest 1000 archived posts and identify which concern Ukraine.” The count limits the collected population across all pages. For this question, posts are selected before topic assessment, without first filtering for Ukraine. If fewer posts are available, the study shows the available count. Collection limit identifies the bound; reaching it does not mean the archive has been exhausted. Files and projects retain the full available text of the selected items.

Numeric legends span the width of the map. Yes/no criteria show 0%, 50% and 100%, with labeled endpoints. Hover over an endpoint for its definition, or open Method and coverage. These values express the model’s assessment of the criterion, not the share of a post devoted to a topic. Ordered scales show the score and definition of every level alongside its color.

Large maps have their own scrollable area with a limited height. A fade at the top or bottom indicates more results beyond the visible area. During collection, new rows stay in view when the map is at the bottom. Scrolling up pauses this following so earlier results can be read; returning to the bottom resumes it.

For a large collection, such as statements by a person or comparisons between groups, the Advisor can create a tile map in the analysis document. The default view shows the study title, a short introduction identifying what is being analyzed, and the grid. The text uses the same body size as the main chat. The color legend is a vertical list beneath the introduction, labeled with the current topic or criterion. All legend labels use the same font size. The Not assessed and Assessment unavailable color keys stay in place as results arrive. Counts are available on hover or in Study options. Light grey tiles have no assessment yet; burgundy tiles have no available assessment. Burgundy does not indicate a negative stance. Selecting a tile opens its details directly. The map updates as results arrive and remains available after reloading the conversation. Partial coverage remains identified, while generic collection-status and tile-unit captions are omitted from the document. Source links alone do not create an empty research section.

The icon beside the study title opens Study options, including filters, List, Comparison, and Method and coverage. A selector appears when several studies are available. The document can also contain separate sections for specific studies. Changing the view uses the same saved results; selecting a legend category filters them even while options are closed.

Bulk analysis can also cover archived Media posts and available original documents attached to legislative projects. Specify the period, people or organizations and the criterion, such as relevance to a named client's activities. Materials are collected progressively and assessed under the same criteria without attaching each record manually. Posts include available reply and quotation context. After selecting a tile, Source context shows the accompanying information used in its assessment.

Follow-up questions can read the original passages for several selected results together, using the saved study and its filters. For example: “Explain the 50 results with the strongest negative assessment and quote their sources.” Advisor reuses the collected originals and identifies them as historical snapshots. A new assessment uses a separate study; reading examples does not change existing scores.

Advisor can also open several known platform sources together across resource types, including posts, legislation, recordings and permitted repository materials. Reads are divided into pages limited by the number of resources and the size of the response. Long sources continue in further fragments. An unavailable source remains an explicit gap while other readable sources can still be used. Existing permissions and additional verification for repository files continue to apply.

The map covers archived materials available and permitted for that analysis. A missing document, passage or context does not establish the absence of a mention or a negative assessment. Incomplete coverage is identified in the study. Assessment maps currently support public materials. Access to private repository files alone does not make them available for this analysis.

For recordings, the Advisor collects and assesses successive transcript pages as one study. A request can name speakers and topics to compare. Speaker names come from the source; unrecognized names and missing attribution are reported. Topic assessments distinguish positive, negative, neutral, mixed and unclear positions from passages that are unrelated to the topic.

Choose an assessment criterion or topic when the study includes several. Speaker, year, group and status filters narrow the displayed passages when those fields are available. Text search finds matching words in the saved passages; it does not run a new semantic analysis. Comparison shows the distribution of assessments for each selected speaker or group. Select a tile or a row in List to read the original excerpt, inspect its scores and use Open source. Changing these controls uses the saved results without repeating the assessment. Older studies without speaker labels need a new analysis before speakers can be compared.

Counts describe collected passages, not necessarily every statement or person in the requested population. Missing or failed assessments remain separate from negative judgments. The Advisor reports incomplete source coverage. Assessments depend on organization settings and are unavailable in a conversation containing confidential context; a separate conversation using public sources is required in that case.

Analysis document appears on the right-hand rail when the conversation has results. The first result can open the document automatically; an open source tab stays selected. Later updates do not reopen a hidden workspace. Select the document icon to open its tab; select the active icon again to hide the additional workspace.

The split-view button opens an additional workspace beside the conversation. Persate pages lists pages available in the organization's package, using the same access rules as the main navigation. This includes Editor when it is enabled. Selecting a page opens it in a tab; the conversation stays open beside it. If access to a page is withdrawn, its open tabs are removed from the workspace. The Plan and Analysis document views require Advisor access.

The search field in New tab filters views, agents and Persate pages. Choosing a page or view replaces that empty tab. Use + to open another new tab. Choosing an already open page switches to its existing tab without creating a duplicate.

In these tabs, contextual sidebars start closed to leave room for the page content. Use Filters, Details or the page's corresponding button in the lower-right corner to open a sliding panel within the tab. Close it to return to the full content area.

Selecting a research agent opens its session in a separate workspace tab. The session shows the assignment, working notes, tool activity and the answer with source links, using the same presentation as the main conversation. Completed activity is collapsed and can be expanded above the answer.

Websites and web search

In the updated desktop app, opening an external source from an answer displays the website in the Advisor pane beside the conversation. Its address bar accepts website addresses and search queries and provides back, forward and reload controls. Open in browser explicitly opens the current page in the system browser.

When working in an owned conversation, select Add to conversation below the address bar, or drag the page title into the conversation, to attach its link and title to the draft. Then enter a question and send the message.

In the web version, external sources open in a separate browser tab. In the desktop app, a source that cannot open inside the app shows a message; it does not automatically launch the system browser. Embedded website browsing requires an updated desktop app. Closing the additional workspace ends its website sessions.

What the Advisor sees

When you send a message or a note during a task, the Advisor receives context from the active workspace tab: its address and title, a bounded excerpt of visible text, selection and filters. For supported viewers this also identifies the open file, legislative attachment, recording and playback position, PDF page or Editor draft. You can ask “Summarize the document I have open” without attaching it again. Full source reads still use your account's permissions. PDF/image pixels and inaccessible embedded frames are not transcribed automatically.

The Advisor also receives the names and addresses of the other open tabs. It can refresh this list, select a relevant tab or open a new Persate page, view or desktop website using its workspace tool. It receives fresh context after the interface confirms the action. Background page contents are read only after selecting that tab. Tab navigation respects the organization's package; the tool does not click page controls or submit forms. Password fields, authentication screens, ordinary form drafts, cookies and browser storage are not included.

File tagging (@)

Entering the @ symbol in the text field opens the document search engine covering the Repository and the public corpus:

• Search by name — a quick mode for files with known titles.
• Hybrid search — a toggle on the right side of the search field enables searching documents based on their content.
• Context chips — appear above the text field for selected files and sources. Multiple items can be attached simultaneously; a source chip links to its material when a direct link is available.

Attach a file directly when it contains the sought answer. If the information location is unknown, allow the advisor to search the resources independently.

Dragging materials into a conversation

Open a Persate page in the additional workspace and drag a material onto the conversation: its messages, empty space or message field. The conversation highlights the drop area. Repository files, legislative projects and documents, recordings, Sejm and Senate voting rows, people, committee rows, and Media stories and events support this action. In Media, drag a story's title or an individual event's title or timeline bar to attach that specific source. Links to attachments and other materials can also be dragged from Persate pages.

The material appears as a chip with its title. Dropping it does not send a message: add a question, such as “Summarize these materials”, and send the draft. Each chip identifies a specific source for the Advisor to read using the account's access permissions. Active-tab context is added separately when sending the message.

After refreshing or reopening the conversation, the question and its source chips remain separate. Regenerating the answer uses the same attached sources.

Materials dropped while the Advisor is working stay in the draft for the next message. They are not added to a running turn through a text note. Read-only shared conversations do not accept drops.

Message timeline

Every turn produces a message timeline streamed before the response content. It constitutes a visible record of the advisor's actions during operation.

Progress notes

Collapsible rows summarize the Advisor's work while the answer is being prepared. Expand a row to read it. Completed work folds automatically; the record can be reopened later.

These notes describe progress. Check the final answer and its cited sources before relying on a conclusion.

Tool call badges

Searches and other actions appear as badges with an icon, an operation label and a status. Completed actions can show their duration. Consecutive actions form a collapsible group with the number of tools used.

Clarifying questions

When the Advisor needs more information, a question and its answer controls appear in the conversation. Select an answer and choose Continue, or choose Skip. Use the question's own controls; text sent through the composer during work is a note to the running task.

Links in responses

Response content may include source chips providing quick access to specific materials:

• Votes in the Voting ledger.
• Stakeholder profiles and alert details.
• Original posts on X.
• Specific files or their fragments in the Repository.
• Recordings or specific moments, labeled with the recording title.

If a saved citation has no available source link, a small ? appears in its place. Hover over it or focus it with the keyboard to read the explanation. The same indication appears in the analysis document.

Under + → Preferences → Advisor, Open Advisor sources defaults to Advisor right pane. In the desktop app, the other choice is Open in platform: it navigates to the selected Persate page in the same app window. External websites stay in the Advisor pane under either choice. In the web version, New browser tab opens the source in a separate browser tab. The setting applies to source chips, document links, research sources and attached references. A recording link keeps its selected moment, and a file link keeps its document target.

Per-message buttons

Icons are available below every completed advisor message:

• Copy — copies the response content to the clipboard. Citation badges are replaced by their textual representation.
• Restart arrow — resubmits the previous query and replaces the current response with a new one.

Plan beside the conversation

The Open Advisor view button in the upper-right corner of the conversation opens an additional workspace. Choose Plan under Views to open the current answer's steps in a tab. The steps and their status update while the Advisor works. Before a plan is available, the tab shows an empty state.

The Plan shortcut on the right edge opens the same tab once the current answer has a plan. On a phone, use the round list button. Closing the plan tab returns to another open tab; the conversation history keeps its current visibility. The + button opens the view selector again.

Research results and follow-up questions

The result unit follows the question: a project, a source such as a post or file, a person, a group, or an explicitly requested passage. For project research, one tile represents one project. Its documents and assessed passages remain available under Evidence after selecting the result.

Method and coverage explains the assessed scope and how passage judgments are combined. Relevance uses the strongest supported relationship; an incomplete collection cannot establish that no relationship exists. Other conflicting categories remain mixed, and numerical results are passage means. These judgments describe collected material, not unread documents.

Project results appear independently once their documents have been read and the collected material assessed. Other projects can still be processing. Incomplete material means coverage is insufficient to rule out a relationship; Unclear means the content itself received an ambiguous assessment. Waiting for documents is not an unclear assessment.

In a result's details, Why is the material incomplete? distinguishes unavailable text, failed or partial reads, collection limits, shortened context and missing assessments. Source counts are not multiplied by the number of passages. A supported relationship can still be shown when other documents are missing. Older analyses may not have recorded the detailed reasons for their gaps.

After the study finishes, apply filters and choose Ask about these results. Enter a further question and choose Ask Advisor. The new study uses the saved source material for those results, including all collected passages belonging to a selected project. The earlier study remains available and the new result identifies its parent. This does not refresh the sources. A changing or unavailable selection must be refreshed or resolved before continuing.

Conversation history

The right interface column contains a list of all conversations on the account:

• New conversation — clears the workspace and initiates a new session.
• A new conversation starts with a temporary title and automatically receives a short subject title, independently of whether research finishes. Older prompt-based titles are updated when the conversation opens. A manually chosen title is preserved. The … menu offers Rename conversation and Move to project.
• Selecting a conversation opens it with the full record — progress notes, action badges, and citations are restored.
• Delete conversation in the … menu asks for confirmation before permanently deleting the conversation and its history.

Projects contains expandable folders; Recent lists conversations without a project. Hover over the Projects heading to reveal + for creating a project. The heading and each folder can be collapsed. Drag a conversation onto a folder or use … → Move to project; its row is then indented underneath.

Each project loads its conversations independently of the recent conversation list. Load more conversations inside a folder shows more of its history. The list refreshes as other members share conversations.

Drag a conversation above or below another conversation in the same section to change their order. The … menu also offers Move conversation up and Move conversation down; when a conversation has keyboard focus, Alt + Up/Down performs the same action. The order is saved for the current account, including shared conversations, and is retained after refresh and when loading more history. New conversations appear first. After a manual reorder, the Recent heading becomes Conversations.

Moving a conversation into a shared project opens a confirmation showing the conversation and project names. All members of that project will be able to read the entire conversation. Choose Share and move to continue or Cancel to keep its current location and access. This also applies when choosing a shared project beside the message field. Private projects do not require this confirmation.

Projects may be private or shared with a group. Membership provides permanent group projects and an organization-wide project, which cannot be deleted. Conversations stay private until their author places them in a shared project. Shared rows show the author's profile picture, with initials when the picture is unavailable.

The author can withdraw sharing with … → Make private. The conversation returns to the unfiled list, and other members lose access, including an open live view. Move to project → No project or moving to a private project has the same effect.

Other members have view-only access. They also see the answer and progress as they arrive, along with later messages from the author. They cannot send messages or stop the author’s work. When a saved analysis is available, it opens automatically beside the conversation. Show conversation closes the analysis pane and returns to the saved chat. Conversations without an analysis open directly in the chat. Readers can inspect the document, charts, saved research and sources, and copy answers. Only the author can send messages, ask follow-up questions, regenerate answers, update the analysis, rename, move or delete the conversation. Access to linked source materials is checked separately. Live agent sessions remain available to their owner.

On narrow screens, the history column is accessible via the History button in the top right corner.

Refresh and resume

An active turn is not lost upon leaving the view, refreshing the browser, or experiencing a temporary connection loss. The advisor executes the turn as a detached task and transmits events to clients connected at the time of their triggering. When the author reopens the conversation, it reconnects to the live stream and restores the progress to date. Other members can read saved results and watch the current answer arrive in view-only mode.

Empty view and error states

Stopping a response displays Response generation was stopped. as plain text in the answer. Any text already generated remains above it.

• "No conversation selected" — the workspace is empty; select an existing session or create a new one.
• Error block with a code — appears below a partial response when a turn fails. Restarting the turn is the recommended first action.