Your team and workspace

Two things decide what a teammate can do. Their seat sets the AI ceiling. Their permissions set everything else. You choose the seat when you invite them. You tune permissions after they join. There are no roles to pick — see Permissions and seats for the full model.

Pick the seat before you invite

The seat is a hard ceiling. Vernais takes the permission set you assigned and strips the three AI capabilities out of it unless the person is on a Max seat: chat.use, investigation.run and codebase.analyze. In the permission editor those three toggles sit greyed out on a Power seat, tagged ⚡ Max seat. No other grant buys AI back.

Invite someone by email

1. 1
   Open the invite panel

   Go to Team, then select Invite member. Email, the shared code and Super invite all live in this one panel.

2. 2
   Enter their address

   Type it into Email address. The invitation works for that exact address and no other.

3. 3
   Choose the seat

   Pick Max seat or Power seat in the Seat field. It is the only choice the invitation carries, and it defaults to Max.

4. 4
   Send it

   Select Send invite. They appear under Pending invitations until they accept.

The invitation is single-use and locked to that address. When they accept, they join with the seat you chose plus the default rights described below. To pull an invite back before it is used, select Revoke on its row — the link stops working at once.

Share one invite code instead

1. 1
   Set the joiner seat

   Team → Invite member → Share an invite link. Set New members join with this seat, then select Save.

2. 2
   Copy the code

   Select Copy. Anyone who enters that code joins with the seat you saved.

3. 3
   Read the line underneath

   It states exactly who the code admits right now, so you don't have to guess.

By default the code only admits people whose email domain matches the owner's company domain, and their email has to be verified. If the owner signed up with a free provider like Gmail, there is no company domain — the code then admits nobody, and email invites are the only way in.

To let a contractor or anyone outside your company use the code, turn on Super invite. The code then accepts any email address. If the code leaks, select Regenerate: the old code stops working for anyone who has not joined yet, and people already in are untouched.

Change what one person can do

1. 1
   Open their permission set

   Team → Members. Select the permissions link on their row — it reads N permissions · Manage, or Full access for the owner.

2. 2
   Toggle capabilities

   Each capability is a checkbox, grouped by section: AI Chat, Products, Initiatives, Data, Team, Integrations, Storage, Billing and more. Use Enable all or Disable all to move a whole section.

3. 3
   Read the tags

   A default tag marks an ordinary right that is on the day someone joins. A ⚡ Max seat tag marks an AI capability, greyed out on a Power seat. An always on tag marks a right nobody can revoke.

4. 4
   Save

   Select Save permissions. The change applies to their next request.

Capabilities come in three kinds, and knowing which is which explains most surprises.

• Ordinary rights — viewing the dashboard, data, initiatives, pages, the calendar. On the moment someone joins.
• AI rights — chat, investigations, codebase analysis. On by default too, but they survive only on a Max seat.
• Manage rights — connecting a tool, running the brain, editing any initiative, managing the team, workspace settings, billing. Off until you turn them on for that person.

You cannot edit the owner's permissions — the person who created the workspace always holds every one of them. A teammate you granted team management can edit other people but never themselves, and can only hand out permissions they already hold. Nobody can promote themselves that way.

Remove someone

1. 1
   Find them

   Team → Members. Search by name or email if the roster is long.

2. 2
   Remove

   Select the trash icon on their row and confirm.

3. 3
   Check what happened

   They lose access on their next request and need a fresh invite to come back. Their seat frees up for someone else.

Related resources

Vernais has no roles. There is no admin, no manager, no viewer. The workspace owner hands each person their own list of capabilities, one by one. On top of that sits a seat, which is a billing ceiling.

Read the two dials in this order. The seat says whether a person can touch the AI at all. The permissions say which sections and actions they get. An AI permission that the seat forbids does nothing.

Seats: the AI ceiling

Every member holds one of two seats. This is the hard billing line, so only the workspace owner can move someone between them. A person whose seat has never been set counts as Max.

The mechanism is short. Vernais takes the person's assigned permissions, then deletes the AI ones unless the seat is Max. In the code the seats are stored as chat (Max) and power.

You can see this in the permission editor. Every AI capability carries a ⚡ Max seat tag. On a Power seat those three checkboxes are also greyed out, so you cannot tick them at all. If they were ticked before the seat changed, they stay stored and stop working.

Permissions: assigned per person

Open the Team section, find the person's row, then click the permissions link on it — it reads N permissions · Manage. A modal lists all 37 capabilities, grouped by section. Tick what that person needs and save. The list you save becomes their whole permission set.

Each section is gated by one right. Switch that right off and the section vanishes from their sidebar — no error message, no greyed-out button. The other rights inside a section hide an action rather than the section: turning off Create initiatives leaves Initiatives visible and takes the button away.

The three tiers

Each capability carries a tier. The tier tells you what the capability is, not who gets it.

A new teammate you have not configured starts with baseline plus AI — 23 capabilities. On a Max seat they can chat on day one. On a Power seat the ceiling strips the AI three, leaving the 20 baseline rights.

Their real access is the set the owner assigned, minus the AI rights their seat forbids.

The permission catalog

These are the 37 capabilities, grouped by section in the order the editor shows them.

One pairing is worth knowing. Any initiative capability quietly adds View initiatives when you save, because you cannot act on what you cannot see.

What the owner can always do

The owner is the person who created the workspace. That is permanent — there is no way to hand ownership to someone else.

• Hold every permission. All 37, always. The owner cannot be locked out.
• Stay uneditable. Nobody can change the owner's own permissions, including the owner.
• Change any seat. Moving a person between Max and Power is owner-only, because the seat is the billing ceiling.
• Grant anything. The owner is not bounded by their own set the way a delegate is.
• Delete the workspace. An owner cannot leave a workspace. Deleting it is the only exit.

Limits

When someone cannot see a section

Sections are hidden rather than disabled, so there is no error message to read. Check in this order — it is the same order the code checks.

• Seat first. If the whole Chat section is missing, look at their seat. A Power seat strips AI chat, investigations and codebase analysis together, and no permission will fix it.
• Permission second. Open their row in Team and click Manage. A section disappears when its gate right is off — View initiatives for Initiatives, Browse the KG data / graph for Data, View reports for Reports.
• Then the pane. Settings panes follow the same rule. Members & permissions and Invitations need team management, General and Workflows need workspace settings, Codebase needs the repo-choosing right, and Chat & AI needs AI chat.

Related resources

One constraint shapes most of what follows: 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.

So "what can the AI see?" has an exact answer. It sees the records in the workspace you are in right now. Not your account. Not the workspace you were in this morning. This page maps that line.

What a workspace actually is

When you create a workspace, Vernais creates two databases named after it. One holds your working data: chats, initiatives, files, activity. The other holds the knowledge graph. It copies the shape of the global setup — the table names and the columns — and copies none of the contents.

That matters more than it sounds. A second workspace is not a view, a filter, or a folder. It is a separate store. Data cannot leak across it, because a query run in one workspace never names the other database at all.

What stays inside one workspace

Memory deserves a second look, because it surprises people who use two workspaces. Tell the AI something useful in one workspace and it writes that to that workspace's database. Ask about it from another workspace and it has no idea. The write refuses to run at all if no workspace is active, rather than fall back to a shared store. See What it knows about you.

You connect each tool again in every workspace

This is the one that catches people, because the tool list looks shared. The catalog is global: every workspace shows the same set of supported tools, with the same logos. The connection is not. Each connection record carries the id of the workspace that made it, and every screen filters to your workspace before it draws a single card.

So a tool you connected last week in one workspace reads as not connected in the next one. The credentials do not travel, and the records they pulled do not travel either. A second workspace means connecting the tools again, syncing them, and running Corvex there. See Connect a tool and Build the graph.

Switching workspaces changes every answer

Switching re-points the whole app. Every section drops what it was holding and re-fetches from the new workspace's databases. The AI does the same. An investigation, a data lookup, or a chat question runs against the workspace that is active when you press send.

The honest consequence: the same question can give two different answers in two workspaces, and both are correct. One workspace has Stripe and Sentry synced. The other has nothing yet. The second one says it has no evidence, which is the right answer for the data in front of it. See When it finds no evidence.

What the line does not cover

Three things sit outside the workspace, and knowing which is which saves a lot of confusion.

• Your account — your name, avatar, password and signed-in devices belong to you, across every workspace you are in.
• The tool catalog — the list of supported tools is the same everywhere. Only the connections are scoped.
• General knowledge and the web — a workspace walls off your data, not the AI's ability to think. It will still explain a concept or search the web, and it tells you the answer came from there and not from your records.

Common questions

Related resources

What the activity log is

Vernais writes an entry each time someone changes the workspace. Creating an initiative. Connecting a tool. Running the Brain. Each entry holds the actor, the action, the time, and the thing it touched. The log is append-only. The code inserts entries and nothing else — there is no update path, no delete path, and no button that edits one. What you read is what happened.

What gets logged

Chat writes no entry at all. Asking the AI a question logs nothing, and no message text is stored here. Elsewhere the entry names the thing, not the work inside it. A page entry names the page, never its body. An integration entry names the tool, never the records it pulled. Two entries do carry your words. An investigation stores its question, and a scrape stores its query. Both sit in the detail line, capped at 600 characters.

Who sees what

Find an event

1. 1
   Search by person, action or target

   The box matches the entry title, the detail line, the actor's name and the target label. It waits 280ms after you stop typing, then runs. It does not match the action id or the meta chips.

2. 2
   Narrow by type

   The dropdown beside the search box lists only the types your workspace has, each with its own count.

3. 3
   Read the day and time

   The When column carries a day label — Today, Yesterday, a weekday, then a date — next to a relative time. Hover it for the full timestamp.

4. 4
   Open the row

   Click any row. A drawer shows the actor, action, detail, type, target id and the exact time. Meta values land here too, like a scrape's page count and intensity.

5. 5
   Load more

   The feed starts at 40 events and appends 40 at a time.

The three numbers at the top

Where the log stops

Related resources