Using the Knowledge Browser

The Knowledge Browser is your window into everything Upstream has learned from your documents, conversations, and data. It gives you three ways to explore your knowledge: a high-level overview, a searchable list, and a visual relationship map.

Three views

Each view is its own page under Knowledge: Overview (/knowledge), Insights (/knowledge/browse) and Graph (/knowledge/graph). Switch between them with the section navigation above the content, which also links to Sources, Review and Wiki. The same items appear under Knowledge in the sidebar.

Your search, type filters and sort order are saved in the page's web address, so refreshing the page or sharing a link brings you (or a teammate) back to the same filtered view. Older links that used ?tab= still open the right page.

Overview

Overview and Dashboard use the same approved-knowledge summary, separate from the recent-item preview. Knowledge items counts approved knowledge, Knowledge types counts its categories, and Added this week counts items created in the last seven days. Older records without a review status are included; pending, rejected and deleted items are not.

When the summary is complete, counts include older knowledge beyond the preview. A Partial summary labels observed counts At least; an unobserved count is Unknown, not zero. The most-referenced value, category shares and trend are hidden while coverage is incomplete. Summary unavailable means the read failed: use Retry summary. Recent items can still load independently.

Knowledge added shows creation dates for currently approved knowledge across 30 UTC calendar days, including today. It is not approval history or a cumulative total. Open View daily counts for the table; Tab and Enter work too. Most referenced among recent items ranks only the recent preview, not the whole workspace. The separate review count may be limited; a failed review read shows unavailable, not “all caught up.”

Select a Recent additions card to open that item's full detail page. You can also Tab to a card and press Enter.

On Dashboard, each Recent activity preview row opens its knowledge item or uploaded document. Browse knowledge opens Overview; summary cards and Dashboard's By type panel are informational.

In Overview's By type panel, select a row to open Insights with that type filter. Insights and Graph show a limited set of items, so an older-only category can have a summary count but no matching items in that preview. Their counts describe items shown, not workspace totals. Search and filters carry between those views without changing the summary.

Insights

The Insights page shows a limited set of recent knowledge items. From here you can search by keyword and filter by entity type to find exactly what you are looking for.

Each entity card shows trust information at a glance:

Status pill -- Entities that are awaiting review show a Pending pill, and rejected entities show a Rejected pill. Approved entities show no pill.
Confidence percentage -- Colored by confidence level: green for 80% and above, amber for 60-79%, and red below 60% -- the same levels used in the review queue.
View source chip -- If an entity was extracted from a document, a View source chip appears on the card. Click it to open the original uploaded document.

Click anywhere on a card to open the entity's detail page. Hover over a card to reveal the Edit and Delete actions.

Graph

The Graph page displays your workspace as an interactive relationship map. Nodes are not only knowledge insights: your sources, wiki pages, artifacts, canvases, epics, sprints and tasks appear as their own kinds, each with its own shape and icon, and lines represent the connections between them. This view is especially useful for spotting patterns -- like which objections cluster around a specific competitor, which battle card cites a pain point, or which wiki page curates a group of insights.

Lines come only from links the objects already store: relationships between insights and where an insight was extracted from, an artifact's citations, sources and parent, a wiki page's embedded insights, artifacts and source excerpts, its [[links]] and its place in the page tree, a canvas's indexed source, and a task's sprint and epic. Nothing is guessed, and an object with no links is still shown on its own. When a link points to something that is not loaded in this view (deleted, outside the loaded window, or not visible to you), the details panel says how many such links there are instead of drawing them.

The chips above the graph show one entry per kind with the number of nodes shown. Select a chip to hide or show that kind; a dashed chip is hidden. A + after a count means more of that kind exist beyond what was loaded, and the note under the chips lists exactly what loaded for this view. The kind you choose is part of the page's web address, so sharing or refreshing the link keeps it; at least one kind always stays shown. Hiding a kind never changes the type filters in the search bar: those narrow insights by category, which is a different thing from a kind.

Searching matches insights the same way as the Insights page and matches other kinds by their title, summary and tags within what is loaded. A category filter in the search bar narrows insights; other kinds then appear only as the muted neighbours of a matching insight.

Nodes are filled by entity type; larger nodes have more connections in the displayed graph, and highly connected hubs glow. Curved arrows show relationship direction. Zoom in to reveal relationship labels, or hover/select a node to show its incident labels at overview. Hovering another node also reveals its labels without clearing your selection or highlighted path. Highlighted paths also reveal their labels. Light and dark themes follow your app preference. With reduced motion enabled in your device settings, the initial layout settles without animated physics; dragging and zoom controls remain available.

The note above the graph says how much of each kind is loaded. Artifacts, sources, epics, sprints and tasks load a page at a time, most recent first; choose Load more … on a kind to add its older objects. Insights show the most recent 100, and search reaches insights, sources and wiki pages beyond what is loaded (other kinds are searched within the loaded pages). Wiki pages and canvases are always loaded whole.

You can Tab to graph nodes and press Enter or Space to select them. On the Graph page, selection opens the existing details panel. Every node's name starts with its kind or, for an insight, its category, so a screen reader announces "Wiki page: …" or "Pain point: …"; the same nodes are also available as a plain list of links for assistive technology.

Selecting a source, wiki page, artifact, canvas, epic, sprint or task shows its title, status, type or dates, when it was last updated, and its Connections with the relationship in each direction. Choose View … on a connection to move the selection to that node, or use the Open … button at the bottom to go to the object's own page: the source's detail page, the wiki page, the artifact, the canvas, or the Sprint Planner opened on that epic, sprint or task (an epic opens in the planner's Tree view; if the planner's lists fail to load, the link is kept with a Retry). Selecting an insight shows the familiar insight details, plus a Workspace links list of the artifacts, pages, sources and other objects that cite, embed or sourced it.

Creating, publishing, editing or deleting an object anywhere in the app updates the graph the next time you open it, with no page reload: a wiki page you just published is a node as soon as you navigate to Graph.

Edge labels show the specific relationship, such as part of, addresses, or switches from, rather than its explanation. Underscores display as spaces. Long custom relationship names are shortened with an ellipsis on the canvas only; select either connected node to read the full name in Connections.

Each connection separates the relationship from its Explanation (when one is recorded) and the connected insight preview. Outgoing means the selected insight points to the connected insight; Incoming means the connected insight points to the selected insight. The relationship keeps its original wording in either direction. Choose View connected insight by click, tap, or Tab then Enter/Space to read that insight in the panel; keyboard focus moves to its content. The preview uses up to two lines, while full relationship names and explanations wrap. These controls also appear in the document graph's Entity details modal. Viewing connections does not change the stored content or relationships.

Graph filters keep matching entities prominent and retain their available direct neighbors as context. Context nodes have an unfilled, dashed outline; their real connecting edges are dashed too. They remain selectable, with readable labels and the same detail/source actions. The graph count separates matches from context; Insights still lists only matches. For example, selecting Customer Feedback can keep a connected Pain Point visible without including unrelated pain points or recursively expanding the whole graph. Context comes only from the currently loaded, authorized records and may be incomplete—missing neighbors are not fabricated or fetched by this view.

On a document’s detail page, selecting a node in Entity Graph opens an Entity details modal with its content, source, tags, connections, and View Full Detail link. Follow a connection to inspect another entity in the modal. Press Escape, use Close, or click outside to dismiss it and return focus to the opening graph node if it is still available, including after entity data refreshes. You can open the same node again.

After the layout settles, the graph fits its nodes and labels inside the available space. This also applies to the entity graph on a document's detail page. Use the Fit graph control (the expand icon beside the zoom buttons) to bring the whole graph back into view.

While the graph is fitted, resizing the window or opening a sidebar adjusts the fit, including with reduced motion enabled. Clicking a node does not turn off automatic fitting. Panning, zooming, or dragging a node keeps your chosen view until you use Fit graph again. Fitting the view preserves selected nodes and highlighted paths. Filtering or refreshing an already-positioned graph keeps its camera when some entities remain in common, although physics can still reposition nodes. A wholly different graph starts a fresh layout and fit. On a narrow screen, use zoom to read labels in a dense overview.

Wiki page hierarchy

Open Wiki to browse its page tree. While reading, the path above the page title links to its parent and other ancestors; Child pages lists the pages directly below it. If you opened a page directly or reloaded it before the tree was cached, choose Load page hierarchy to fetch the shared page list. Reading the page alone does not trigger that organization-wide scan. All wiki pages returns to the full tree. These are ordinary links: Tab and Enter, or open in a new tab, work as usual. What links here is separate: it lists pages that cite this one, not its children.

Use More page actions → New page under this one to create a child, or Move in the tree… to change a page's parent. A topic can be a parent page; cross-topic category filtering is not available yet.

Knowledge Chat

Click Chat at the right of the section navigation to open Knowledge Chat, a panel for asking questions about the loaded knowledge base without leaving the browser. It is available on every Knowledge page, including Sources and Review. Answers cite the knowledge items they draw on; select a citation to open that item's detail page. On wide screens the panel sits beside the current view. On phones and smaller tablets it appears below the view and scrolls on its own, so the page never scrolls sideways. Close it with the Close Knowledge Chat button or by clicking Chat again; closing stops any answer that is still being written. The conversation stays in the panel while you move between Knowledge pages and until you leave the section.

On wide screens, conversation history uses the available panel height with the message box below it. In a very short window, the panel scrolls with the page so multiline drafts and Send remain reachable.

Searching and filtering entities

Use the search bar at the top of the Overview, Insights and Graph pages to find entities by keyword. You can also filter by any of the ten entity types:

Pain Points -- Problems your prospects and customers describe
Objections -- Reasons people give for not buying or adopting
Feature Requests -- Capabilities people ask for
Content Ideas -- Topics and angles worth creating content about
Competitor Mentions -- References to competing products or approaches
Deal Context -- Details about specific opportunities and accounts
Market Insights -- Trends, shifts, and industry observations
Messaging Themes -- Language and positioning that resonates
Customer Feedback -- Reactions, praise, and criticism from users
Process Gaps -- Workflow breakdowns and inefficiencies

Combine search terms with type filters to narrow results quickly. For example, filter to "Objections" and search for "pricing" to find every pricing-related objection in your knowledge base.

Viewing entity details

Click any entity card to open its detail view. You will find:

Content -- The full extracted text or description of the entity
Confidence score -- A percentage indicating how confident the extraction is (higher means more reliable)
Relationships -- Other entities this one connects to, with labeled relationship types
Source -- Recorded source details, including a source link when available. Chat Proposal identifies an item proposed in chat; it is not a link to the original conversation.
Creation metadata -- When the entity was created and how it entered your knowledge base. For items you created, Created by shows your signed-in profile name, email, or You. Other creator IDs show Unknown user with the recorded ID, not a guessed teammate name.

On phones and smaller tablets, the full entity detail page shows its title, content, source, and related entities first. Actions, Metadata, and Tags follow below; scroll down to edit or delete the entity. On wider screens, those panels stay in the sidebar.

If the entity was extracted from a document, the Metadata panel includes a View source chip. Click it to open the original uploaded document the entity came from.

Navigating between connected entities

From any entity's detail view, click on a related entity to jump directly to it. This lets you follow threads through your knowledge -- start at a competitor mention, follow it to the objections associated with that competitor, then see what messaging themes counter those objections. Each click takes you deeper into the web of connections.

Adding knowledge items manually

Not all knowledge comes from a document. You can record a customer need, objection or other useful information directly:

On Insights or Graph, click Add item to open Add knowledge item.
Choose the type.
Enter the content. Optionally add a Source label and comma-separated tags. The label describes where the information came from; it does not upload or attach a document.
Click Add item.

Use Edit on an item to update it, then Save Changes. A failed save keeps the dialog open so you can retry without retyping.

Deleting an item

Delete opens Delete knowledge item? before any change. Delete item removes it from Knowledge and new search results, but keeps its source document and other knowledge items. There is no in-app undo. Use Cancel to keep the item.

Reviewing suggested knowledge -- Compare full statements and sources, then approve or reject suggestions
Asking Questions with Chat -- Query your knowledge base conversationally
Creating Artifacts -- Generate deliverables from your entities

Still need help?

Email us at hello@getupstream.ai -- we read every message.