06Documentation
Everyday tools
The surfaces you live in — and how each one feeds the evidence.
Write things down where the AI can read them
Pages is a full document editor. What you write there stays a document — until you tag it into an initiative, and it becomes evidence the AI can cite.
Most of what a PM knows never reaches the knowledge graph. It lives in a spec, a retro, a customer call summary. Vernais cannot cite what it cannot read. This page shows you how to write in Pages so that your own thinking can become a record the AI points at.
Start from a template, not a blank page
Pick the template that matches the job, then delete what you don't need.
Pages ships five starting points: Blank page, Product spec (PRD), Meeting notes, Retrospective, and Project tracker. Each one seeds real blocks, not placeholder text. The PRD arrives with Problem, Goals & non-goals, Success metrics, Requirements and Open questions headings. The Project tracker arrives with a working database in board view, already wired to Task, Status, Owner and Due columns.
- 1Open the tree
Go to Pages. The left rail lists your folders and pages.
- 2Add a page
Click
+in the rail header. PickNew folder, or one of the five templates underStart from a template. - 3Write
The page opens pre-filled and ready. Edits save themselves about a third of a second after you stop typing. The header shows
Saving…thenSaved.
Pages is single-editor. If a teammate saves a newer version while your save is still in flight, yours is refused and your editor reloads to theirs. Your unsaved text is lost. This is not live co-editing — one person at a time on a page.
Write with blocks
Type / and name what you want. That is the whole editor.
A page is a stack of blocks. Each block has a type — paragraph, heading, to-do, quote, callout, code, image, database. Any block can turn into any other type. That is why an imported Word file becomes editable content instead of a dead attachment.
The / menu carries the full set:
- Writing —
Text,Heading 1-3,Bulleted list,Numbered list,To-do,Toggle list,Quote,Callout,Code - Structure —
Sub-page,Table(a database),Simple grid,Table of contents,Divider - Media and math —
Equation,Image,Bookmark - Vernais blocks —
Initiative,Product,Initiative dashboard,Storage file,Send page to initiative
If you already type Markdown, keep typing it. At the start of a block these become the block instantly:
- Headings —
#,##,### - Lists —
-,*,+, or1. - To-do —
[]for an empty box,[x]for a ticked one - Quote and code —
>and ``` - Selection —
Cmd/Ctrl+Bbold,+Iitalic,+Einline code.TabandShift+Tabnest a list.
For a long doc, drop in a Table of contents block. It lists every Heading 1-3 on the page, indents by level, and updates itself as you write.
Track work in a database, not a grid
If you want status, owners or filters, reach for Table — never Simple grid.
Two blocks share the word table, and people pick wrong. Here is the split:
| Block | What it is | Pick it when |
|---|---|---|
Simple grid | Rows and columns of plain text. No types, no filters. | You want a small static table inside a doc. |
Table | A real database. Typed columns, four views, filters, sorts, formulas. | You want to track anything that has a state or an owner. |
A database gives each column a type, and the type decides how the cell behaves. There are 14: Text, Number, Select, Status, Multi-select, Date, Person, Checkbox, URL, Email, Phone, Formula, Created time, Last edited. Select and Status show colored chips. Person shows real teammate avatars. Formula, Created time and Last edited compute themselves and cannot be typed into.
The same records render four ways — Table, Board, Gallery, List. Switch with the chips at the top of the block. Switching a view never changes your data. From a column header you can Filter, Sort, Group, Calculate, Freeze up to column, or Hide. Filters and sorts save with the page, so the view you left is the view you return to.
Outgrew a Simple grid? Click Convert to database under it. That upgrade runs one way only — a database never converts back. The test: if you will ever want to filter it, start with the database.
Bring an existing document in
Import when the source of truth already exists somewhere else.
Page toolbar → ⋯ → Import document. Pages accepts .pdf, .docx, .pptx, .xlsx, .csv, .html, .epub, .md, .txt and more. The file is converted to Markdown and appended as native, editable blocks under a heading named after the file. Headings and tables survive the trip. The file is also registered in Storage.
Import appends to the bottom of the page — it never replaces. Import the same file twice and you get two copies.
Add math, and argue in the margin
For anything with a formula, use / → Equation and type LaTeX. It renders as typeset math with a live preview. Press Escape or Cmd+Enter to commit, and click the rendered block to edit it again. A broken formula shows in red instead of breaking the page.
\text{conversion} = \frac{\text{purchases}}{\text{checkout starts}}For discussion, open ⋯ → Comments. A right panel shows the thread with the unresolved count. You can @mention teammates, resolve threads, and delete your own. The same panel lists Linked from — every other page that references this one through a sub-page block or a page mention.
Send a page into an initiative
Tagging a page is how your writing stops being a doc and starts being evidence.
One constraint shapes this. When Vernais answers from your connected data, every claim has to trace back to a record it can point at. It will not invent a cause. A question it cannot ground in your data returns nothing rather than a guess. Questions about the outside world, or about general knowledge, it answers like any good assistant — and tells you which world the answer came from. Your page is not one of those records — until you tag it. Then it is.
- 1Open the picker
Page toolbar →
⋯→Tag to initiative. Or type/→Send page to initiative. - 2Choose
Search and tick one or more initiatives, then hit
Send to N initiatives. - 3Read the receipt
The page saves first, then goes to each initiative. A toast tells you how many landed, and whether any classification will retry.
The whole page goes, not the part you selected. Each initiative treats it as a captured signal: the AI reads it, judges whether it is actually relevant to that initiative's problem, and routes it into correlations, hypotheses and Decide — the same path a tagged live-chat message takes. An irrelevant page gets filed as unrelated rather than forced into the story.
Do not tag a page that only holds your opinion. Evidence is a page carrying things the AI cannot get elsewhere — a customer's exact words, a decision and its reason, a number you measured by hand. The test: if a teammate asked 'how do you know that?', would this page answer them?
Watch an initiative's numbers from a page
The pinned Initiatives folder holds live dashboards, not documents you wrote.
At the bottom of the Pages rail sits a folder Vernais owns. Click + on it and pick an initiative, and a page appears showing that initiative's live dashboard: confidence, key metrics, the problem statement, signals by integration, outcome breakdown, themes and the primary hypothesis. You can also drop the same thing into any page with / → Initiative dashboard.
Here is the part that matters. These pages fetch aggregated counts only — never raw records. An initiative backed by millions of signals still renders, because the page is only ever asking 'how many', not 'show me each one'. So the numbers are read-only. They re-fetch each time you open them.
Want the actual records behind a count, not the count? Go to the Data section instead — that is where rows live. The dashboard page tells you the shape; Data tells you the contents.
The pinned Initiatives and Transcript folders cannot be deleted — the server refuses — and the rail will not let you drag them. If a Transcript date folder looks empty, that is expected — each transcript is restricted to that meeting's attendees.
Know the edges
Sub-page blocks.[Document truncated for this request] marker.Ask AI instead of the whole page.Two page states catch people out. Lock page makes a page read-only for everyone including you — editing, AI and import all stop until you unlock it, and the server enforces it. And on a public page, anyone in your workspace can edit the text, but only the creator can change visibility, lock, verify or delete it.
Related resources
Talk to your team without losing the thread
Channels, DMs, and group chats. Anything said in them can go straight into an initiative as evidence, in one keystroke.
Live chat sits inside your workspace. The workspace is the server: every channel, DM, and group lives in it, and everyone you can message is a member of it. Switch workspaces and you get a separate chat with separate people.
Pick the room that fits
Match the room to the audience: a channel for the team, a DM for one person, a group for a few.
Every workspace starts with a #general channel. Channels sit in categories down the left rail.
| Room | Who can read it | How it starts |
|---|---|---|
| Public channel | Everyone in the workspace | An admin creates it |
| Private channel | Only the people listed on it | An admin creates it and names the members |
| DM | You and one teammate | New conversation, then pick one person |
| Group DM | You and two or more teammates | New conversation, then pick two or more |
Opening a DM with the same person twice reuses the one you already have. You never end up with two.
A private channel and a group DM overlap. Ask for a private channel when the group is durable and wants a topic and pins. Start a group when it is a few people and one thread. Test: will this conversation outlive the week?
Forward a message
Forwarding moves a message to the people who need it, and builds the room on the way.
- 1Hover the message
A small toolbar appears. Click the forward icon.
- 2Tick the destinations
Pick any mix of people and channels. The button counts your picks and reads
Forward (N) → grouponce you tick two people. - 3Send it
Each ticked channel gets a copy. One person opens a DM. Two or more creates a brand-new group, seeded with that message.
React, reply, and tidy up
React when you have nothing to add. It closes the loop without a new message.
- React — hover a message and pick an emoji. Reactions stack with a count, hovering shows who reacted, and clicking your own removes it.
- Emoji — type
:plus two letters for inline autocomplete, or open the picker for categories, search, and skin tones. The picker stays open so you can add several. - GIFs — the GIF tab searches and posts a GIF as its own message.
- Stickers — built-in emoji-glyph stickers, no setup needed.
- Reply — quotes the message you answer, so the thread stays readable.
- Edit — you can edit your own messages only. Everyone sees the new text live.
- Delete — you can delete your own. An admin can delete anyone's.
- Pin — you can pin your own messages, and an admin can pin any. Pins open from the conversation header and everyone in the room can read them.
The GIF tab needs a Tenor or Giphy key in the server's environment, so whoever runs your deployment sets it. Without one the tab comes back empty with a note. Everything else in live chat works without any key.
Get someone's attention
Mention the person who has to act. Everything else is a badge they read later.
Type @ and pick a teammate, or use @everyone or @here. Type # to link a channel.
- A mention sends that person an inbox notification. It carries your name, the channel, and a snippet of what you wrote.
- @everyone and @here both notify every person in the conversation.
@hereis not the narrower one it looks like, so the cost scales with the room either way. - Unread badges count new messages per conversation. Opening it marks it read.
- Mention badges count only the messages that named you, so they survive a busy channel.
- Presence shows you as online, away, in a call, on Do Not Disturb, or invisible. Your status also gates your own notifications: on Do Not Disturb, none reach you.
- Typing shows under the message list while someone writes. It broadcasts at most once every three seconds.
- The member list opens from the conversation header, with avatars and presence grouped by status.
Find things and share files
The search box in the conversation header finds messages in that channel or DM. It searches the room you are in, not the whole workspace.
The paperclip attaches files. They stage as chips before you send, and everything shared in a room is also listed in the details rail. Paste a URL and it unfurls into a preview card.
Send a message into an initiative
Tag the message that carries the claim, not the one that agrees with it.
A teammate drops the exact detail your initiative needs. Capture it where it lands instead of retyping it later.
- 1Open the picker
Type
/at the very start of a message and chooseCapture as a signal. TheCapture as signalbutton in the composer toolbar does the same thing. - 2Pick the initiatives
Search by name and click one or more. They stage as removable chips above the input.
- 3Write the message and press Enter
It posts to the channel or DM as normal, and goes to each initiative you tagged.
- 4Check the initiative
Open the Initiatives section. Your message shows up in the evidence clusters, named
Chat — <your first words>, carrying who tagged it.
Behind the send, Vernais rewrites your message into a clean note and files it. It then judges whether the message truly bears on that initiative. Related evidence goes in the Related bucket, ranked by how strongly it relates. The rest lands in Unrelated, honestly, instead of padding the case. The judgement runs on your original words, not the rewrite.
A fresh, relevant signal can also nudge the initiative itself. It may move the confidence number, sharpen the problem statement, or add an insight. The move is bounded to 25 points either way, and most signals do not move it at all. Only genuinely new evidence does.
The tag chip on a sent message records what you picked. It does not confirm the filing succeeded. If a signal never appears in the initiative, re-tag it rather than assume it landed.
Deleting the chat message does not remove a signal already filed. They are separate records. Remove it from the initiative's evidence if you want it gone.
Chat is one of several doors into an initiative. See Tag evidence into an initiative for pages, files, and meeting transcripts.
Develop your intuition
The scoring reads what you actually typed. A message that carries the claim gets ranked; a message that only agrees with one has nothing to match on.
Turning on encryption for a DM costs you two things, because the server can no longer read the message: you cannot tag it into an initiative, and a mention in it raises no notification. Encrypted DMs carry a lock and an E2EE chip in the header. Use them for genuinely private talk, not for anything you want the AI to reason over later.
Related resources
Plan your week
Three views, drag-to-move for the events you organize, and suggested times worked out from your teammates' real calendars before you pick a slot.
Pick a view
The Calendar opens on Day view. Use Month to plan ahead, Day to run today.
The view switcher sits at the right of the toolbar, next to the bell. Today jumps back to now, and the arrows page back and forward one month, week, or day.
| View | What you see | Reach for it when |
|---|---|---|
| Month | A six-week grid. Each day shows up to three events — four when teammates are overlaid — then a +N more link that opens the whole day. | You want to spot a heavy stretch weeks out. |
| Week | Seven hour-by-hour columns, with a line across today marking the current time. | You are shaping the week. |
| Day | One hour-by-hour column. Add teammates and each person gets a column of their own. | You are running today, or comparing calendars. |
Week and Day scroll to about 7:30am on open, so your morning is on screen without a drag.
Put something on the calendar
The composer opens in the left panel, not a pop-up, so you can keep reading the grid while you set the time.
- 1Open the composer
Click
Createin the left panel, or click an empty slot in Day or Week. A clicked slot fills in its own time; a Month day cell starts you at 9am and switches to Day view. Your draft is drawn live on the grid as you adjust it. - 2Set the basics
Title, date, start, and end. Turn on
All dayfor a whole-day block. Set repeat to daily, weekly, monthly, or yearly, and every occurrence is drawn on the grid. - 3Open More options
This reveals location, color, busy or free, and description. It opens on its own once you add a guest, and it is always open when you edit an existing event.
- 4Invite people
Add workspace members under
Meet with. Their calendars overlay the grid in their own color, and they become guests of what you are scheduling. Adding one flips the draft to a meeting. - 5Save
Click
Save. The event lands on the grid straight away.
The caret beside Create opens four starting points: Event, Video meeting, Focus time, and Out of office. An out-of-office block does real work. A meeting invitation landing on top of it is auto-declined for you, and the organizer is told why. You can still open it and accept.
There is no separate meetings list. A meeting is a calendar event with Video meeting switched on, and the invitations and join link live on the event itself. See Run a meeting.
Move, edit, or delete an event
In Week or Day, drag a block you organize to a new time. It snaps to 15 minutes, saves on drop, and keeps every other field.
Drag only works on a timed event you organize. An all-day block, a repeating series, and anything someone else organizes stay put — use the composer, or propose a new time. In Week you can drag across to another day; in Day you are moving within the one day.
Moving a meeting re-arms its announcements. A new start re-arms the ready-to-start notice, and a new start or end re-arms the scheduled-time-is-up nudge. That is why a rescheduled meeting announces itself a second time.
To change anything else, click the event and use the edit and delete buttons on the card. Who can do what:
- Organizer — the person who created the event. Only they see the edit and delete buttons on the card, and only they can move it.
- Guest — responds
Yes,No, orMaybe. On a timed, non-repeating meeting they also getPropose a new time, which the organizer accepts or declines. There is no edit button, by design. - Act task blocks — assigning an Act task mirrors its window onto the assignee's calendar as a busy block, titled with a
▸and markedAuto-synced from Act task. The task is the only writer, so change the task in Act.
See who is free before you pick a time
Add people under Meet with first. Everything else on this screen reads from who is on that list.
- 1Add the people
Search workspace members in the left panel and add up to eight. Each gets a color, an eye toggle to hide their calendar, and an ✕ to drop them.
- 2Read the grid
Their events overlay the grid in their color, and in Day view each person gets a column. The composer shows a one-line verdict on your chosen slot:
Everyone is free at this time, or2 conflicts at this time. - 3Take a suggested time
Under
Suggested times, Vernais offers up to three open slots, each labelled with a short reason. Click one to apply it to the event you are composing, orRefreshto ask again.
The suggestions are grounded, not guessed. Vernais works out the real free gaps first, deterministically, from everyone's actual events between 7am and 9pm. Only those gaps reach the model, so it can only pick a time when every attendee is free. It prefers mid-morning and early afternoon and steers around the lunch hour. If nothing fits, it says so rather than offering a slot that clashes.
Marking an event Free instead of Busy leaves that time open to schedulers — it never blocks a suggestion and never counts as a conflict. Use it for a soft hold. The test: if a teammate booking over this block would annoy you, leave it Busy.
Jump straight to an event
You rarely need to hunt for a date. Three shortcuts land you on the exact event:
- The bell in the Calendar header — your invitations for the next two weeks, answerable
Yes,No, orMayberight there. Unanswered ones come first and drive the badge. - Happening now, at the top of the bell — a meeting you organize, or said
YesorMaybeto, with aJoin nowbutton. It appears from five minutes before the start through thirty minutes after, or any time the meeting is live. - A meeting notification in the Inbox — its
TranscriptandRecordingbuttons switch to the Calendar and open that event's card, wherever it sits in time.
The Calendar is per workspace. Switching workspace shows a different, possibly empty, set of events. Nothing is shared across the line — see Workspaces and privacy.
Related resources
Look up a record yourself
Every record Vernais learned from sits in one table. Filter it, open any row, and read the original data the tool sent.
Vernais builds a knowledge graph out of your connected tools. The Data section is that graph, laid out as a table you can read. When Vernais answers from your connected data, every claim has to trace back to a record it can point at. This is where you go and check that record yourself.
What you are looking at
The Data section shows one row per thing in your workspace's knowledge graph.
Open it from the left nav. You land on the Nodes tab. The header counts what is there: N validated nodes · N sources. Three tabs sit in the left rail — Nodes, Validation graph, and Scrapes.
The table columns are Node, Type, Source, Stream, Lane, Connections, and Updated. Every row is one of a few kinds.
One record from one connected tool, after cleaning. A single Jira issue. One Stripe charge. Duplicates were removed before it became a signal.
A thing that many records talk about — a customer, a version, a bug. Stripe's Acme Corp and Zendesk's acme.com become one entity.
A group of related records that the maths found on its own, like promo-code failure at checkout. Nobody typed the list.
A parent topic over several close topics. Open one and it lists its subtopics.
Stream says which feed inside a tool the record came from. Stripe has charges. Jira has issues. One tool can send very different kinds of record.
The Scrapes tab lists pages Vernais collected from the public web, with the site, the kind, a detected sentiment and a rating. Targets in the left rail filters them by website. Once a scrape is stored, those pages are signals in this same graph, branded by the site they came from. See Scrape the web.
Narrow the list
Two filter groups live in the left rail: Node type and Source.
Node type narrows to Entity, Signal, Topic, Umbrella, or Metric. Source narrows to one tool, with that tool's real logo on the row. Both are multi-select. Only tools that really produced records appear.
The count on each filter row is a corpus total, not a count of your current view. Picking Jira does not change the Topic badge. That is deliberate — the numbers stay a stable reference instead of shifting under you.
The search box sits at the top of the page. Results update after a short pause. Searching sends you back to page one. Under the table, Prev and Next walk 50 rows at a time.
The Sort dropdown sits next to the search box.
| Sort | What it orders by | Watch for |
|---|---|---|
| Name (A–Z) | The record's name. | The default. |
| Most recent | When the record was created. | Newest first. |
| Most connected | How many source records mention the thing — the drawer's Mentions number. | This is not the Connections column. That column counts graph links. Two close numbers with confusingly similar names. |
Browsing is for checking one record. If you want a number across many records, ask the chat — it counts your whole corpus, not a 5,000-row pool. The test: are you reading a row, or counting rows?
Open a record and read the receipt
Click any row and a drawer slides in with everything Vernais stored about that record.
- 1Click the row
The drawer opens on the right. Its top line names the kind — the tool and stream for a signal, the type for an entity.
- 2Read the meta grid
Type, Integration, Stream, Lane, Entity type, Connections, Mentions, Members, External ID.
- 3Read the signal text
The cleaned text Vernais actually reasons from.
- 4Check the tool's real fields
The named fields from the source — Jira priority, status and assignee; Stripe metadata.
- 5Expand Raw payload
The untouched original record, exactly as the tool returned it. Nothing is summarised away.
That last step is the whole point. Take a claim, find its record, open Raw payload, and compare it against the source tool. If Vernais stored something, it can show you where it came from.
The Lane column says how a record was processed. S is mostly fields and numbers, like a Stripe charge. T is mostly prose, like a Slack message. H is both, like a Jira ticket with fields on top and a written description below. It explains why some records are rich with text and others are bare.
Follow the connections
The Connections list in the drawer is clickable — each row jumps to that record.
Each connection shows a direction arrow, the relationship in plain words, and the other record's name. Click one and the drawer reopens on it. You can keep chasing the thread across tools.
Connections are earned, not guessed. Three kinds exist:
- Mentions — this record talks about this thing. A Jira ticket points at the promo-code entity.
- Shared entity — these two records are about the same thing. A Jira issue and a GitHub commit that both name
v2.2.1get linked to each other. - Relationship — a fact pulled from the record's own text, kept only after a check confirmed the text supports it.
Open an entity and the drawer shows Resolved from N mentions. Those are the different wordings across tools that Vernais decided are the same thing. That merge is how one tool's data reaches another's.
Open a topic and you can audit the grouping yourself. Keywords are phrases lifted from the members' own text. Members are the records in the group. Context links · not verified members is a separate list — those records only look close. They are there to navigate to, and they never count as members.
A very connected entity does not link every pair of its records. Past a limit, Vernais links one representative record per tool instead. So two Jira issues sharing a busy entity may have no direct line. They still connect through the entity itself.
See it as a map
The Validation graph tab draws the same records as dots and their connections as lines.
It follows the table. Your Node type and Source filters flow straight into it, so the scope you built stays.
- Signals are coloured by the tool they came from, and carry that tool's logo.
- Entities are coloured by what kind of thing they are — bug is red, version is purple, feature is green.
- Topics are amber.
- Lines are coloured by the kind of connection.
The legend sits at the bottom left. Types expands to all 13 entity types with their colours. The sources button expands to the tool logos present.
- 1Click a node
Its neighbours stay bright and everything else fades. A panel opens on the right with the same detail as the table drawer, and the graph stays visible and usable behind it.
- 2Click the background
The focus clears.
- 3Search the graph
Type a name in the toolbar box. The view re-seeds around matches, then pulls in their neighbours so you get a connected picture instead of scattered dots.
- 4Set Max
Choose how many nodes to draw. The list steps up to your workspace total, marked
(full). - 5Press Fit
Zooms and pans so the whole graph fits the window again.
Laying out the graph… and a progress bar before anything appears.(full) shows your real corpus total. If that total is over 5,000, you get a capped view.Capped chip. Filter by source or node type to see a complete slice instead of a partial whole.When the page is empty
Data always shows the ACTIVE workspace, and a new workspace starts genuinely empty.
Every workspace has its own private knowledge graph. Nothing leaks in from another one. Until you fill this workspace's graph, the table reads No knowledge-graph data yet and the graph reads Nothing to visualize. Three things have to happen, in order.
- 1Connect your tools
Left nav → Integrations. See Connect a tool.
- 2Sync the data
Connecting alone is not enough. The raw records have to actually arrive.
- 3Run Corvex
Left nav → Integrations → Run Corvex. It is not in the Data section, which is where people look first. It runs 15 stages over a few minutes, and each one reports its real numbers as it goes.
Running Corvex is an elevated permission. Browsing Data is on by default for everyone. So a teammate can read the empty state, be told to run the Brain, and not be allowed to. Ask your workspace owner. See Permissions and seats.
Re-running the Brain rebuilds the topic layer from scratch. Topic names and ids can change between runs. Your signals and entities survive.
Related resources
Keep files where the AI can use them
Storage lists every file your workspace has touched, whatever made it. It is where a file lands after the AI reads it — and where you go to find it again.
Storage is an index, not a drawer. Every file your workspace touches gets a row here: a chat attachment, a meeting transcript, a report you exported, a page import, a scraped dataset, a file on an initiative. The original stays where it was made. Storage points at it.
Where your files come from
You do not have to file anything. When a file enters the workspace, Storage records it and drops it in a folder that matches its source. Open the section from Setup → Storage in the sidebar.
| Folder | What lands in it |
|---|---|
| Chat attachments | A file you attach to a chat message |
| Initiatives | A file you add to an initiative's Files tab |
| Recordings | A meeting transcript |
| Reports | A report PDF you export, and reports the AI writes for you |
| Pages | A document you import into a page |
| Web scrapes | The dataset from a finished scrape |
| Uploads | A file you upload in Storage yourself |
Saving the same file again updates its existing row instead of adding a second one.
Give a file to the AI
Vernais reads a file the moment you hand it over. It does not read the Storage list on its own. There are two ways to hand one over.
- Attach it in chat — the text is pulled out of the file and sent with your question, so the model can quote it.
- Add it to an initiative — upload it on the initiative's Files tab. It is read there as evidence, and it can update that initiative.
Attach a file in a chat thread that already produced an initiative and you get both at once. The file is read for your answer and filed on that initiative.
When Vernais answers from your connected data, every claim has to trace back to a record it can point at. A file you attach becomes one of those records for that question. It will not invent a cause from a file it never saw.
If you want the file kept, not read, skip the attachment and upload it in Storage. The test: do you want an answer, or a copy?
Upload a file straight to Storage
- 1Open where you want it
A file lands in the folder you have open. From
Home,All files,RecentorStarredit lands inUploadsinstead. - 2Drop it in
Drag files onto the page, or click
Upload. Several at once is fine. - 3Read the toast
It confirms how many files were added, and tells you when one was too large to preview.
Find a file again
The left rail is the search surface. Home shows loose files that sit in no folder. All files shows everything you are allowed to see. Recent and Starred do what they say. Your folder tree sits below them.
- Search — matches the file name and its short snippet. It does not read inside the full document.
- Sort — recent or oldest first, name A–Z or Z–A, largest or smallest first.
- Star — click the star on a row to pin it to
Starred.
Know where a file came from? Open that source's folder first, then search inside it.
Open and preview a file
Click a file for its details, or open the viewer to read it in place. Images, PDFs, video, audio and text render right there. An Office file or an unknown type shows its snippet and a download button. A file with no stored contents points you back to its source section, where the full document lives.
Put a stored file in a page
Add a Storage file block to any page and search your storage. The block shows the file with View, Download, and a link back to the Storage section.
Folders and who sees what
The source folders build themselves. You add your own on top.
| What you can do | What you need |
|---|---|
| See your own uploads and your usage | Nothing extra — this one is on by default |
| See everyone's files and who uploaded them | The See all members' files + who uploaded permission |
| Create, rename or delete a folder | The same permission, because folders are shared by the whole workspace |
Files on an initiative are shared with everyone, whatever your permissions. Deleting a folder does not delete files: they move to All files, and a source folder rebuilds itself the next time that kind of file arrives.
Remove a file
Removing a row takes the file out of the index. The original stays where it was created, so a chat attachment or an exported report is still in its own section.
A file you uploaded straight to Storage has no other copy. Removing it is permanent — there is no trash and no undo.
Related resources
See the state of play
Five numbers and six panels. Here is what each one counts, where it reads that count from, and which ones judge you.
Open Dashboard from the sidebar. Everyone on your team sees it by default, though a workspace owner can switch it off for a specific person — see Permissions and seats. The page refreshes itself every 30 seconds while you are looking at the tab, so you can leave it open.
Read the five numbers
Every card is a button. Click a number to land in the section it was counted from.
| Card | What it counts | Where the number comes from |
|---|---|---|
| Signals | One cleaned record from a connected tool: a Jira issue, a Slack message, a Stripe charge. | Processed signals in your workspace graph. If the graph is empty, it counts raw synced records instead. |
| Connected integrations | Tools you can pull from, over the size of the whole catalog. | This workspace's connections only, and only those that actually store credentials. A connection created but never given credentials never reads as connected. |
| Confidence | A 0.00 to 1.00 average of how sure the AI is across your initiatives. | Only initiatives that have a score. Blanks are left out, so they cannot drag the average down. The caption tells you how many were counted. |
| Escalations | Initiatives that are drifting. Turns red above zero. | Two rules, either one fires. See below. |
| Decision-ready | Initiatives you can make a call on. | Three rules, any one of them qualifies. See below. |
Read the caption under Signals before you compare it to anything. grounded in the knowledge graph means the records are processed and connected. synced records across your sources means they are still raw, and the number changed meaning. Two workspaces showing the same Signals number are not always in the same shape.
The two rules that decide red and green
These two cards judge your portfolio, so it is worth knowing exactly what trips them.
- Escalations fires when an initiative is high or critical impact and is not yet decision-ready, or when nobody has updated it for 14 days or more. Staleness alone is enough. A low-impact initiative can escalate purely by sitting still. Fourteen days is the default and a workspace can be configured to a different number.
- Decision-ready counts an initiative if it has reached the Decide, Act or Measure stage, or its confidence is 70% or higher, or the AI rated it high confidence. Any one of the three.
Decision-ready is an OR, not an AND. An early-stage initiative with one strong score counts, even though nobody has planned anything. If you want the ones with a plan behind them, open Initiatives and read the stage instead. The test: do you want a verdict, or work you can hand out?
Read the panels below
They appear in this order down the page.
- Initiative timeline plots when initiatives were created, across the full span of your data, bucketed by day, week or month depending on how wide that span is.
- My tasks shows up to 6 of your open tasks, active first, then by priority, then by due date.
View all tasksopens your full queue. - Inbox shows your latest 8 notifications, with a dot on the unread ones and a filter box for the list.
- Decision-ready initiatives lists the top 4 by confidence. Click one to open it.
- Source sync health shows up to 6 connected tools with a health state: error, syncing, healthy or idle. Broken ones sort to the top. Check this first when a number looks wrong.
- Where time goes shows the share of active initiatives sitting in each stage. It counts initiatives, not hours logged. Nobody is tracking your time.
What the range toggle does not do
A workspace can show connected sources and zero signals at the same time. The tool is connected, but nothing has been synced and processed into this workspace's graph yet. See Build the graph.