Skip to content
Botscape

All features

Sixteen features. Six stages. Six seats.

In the order you meet them. Each stage is hosted by the seat that does the work there. Each feature shows what it does, how you can tell, and where it currently stops.

11 shipped · 5 partial · 16 in all

Stage 1 · Prime hosts

Arrive and set up

Onboarding, settings, templates, connections

Prime says: Three steps. Then I’m standing and answering.

Onboarding. Three roles, three checklists: operator, owner and member each see only their own next step. Progress is marked as you go, so a half-finished setup picks up where it stopped. The last step is a real reply from Prime, not a tour. Status: shipped. Onboarding covers the first workspace. Adding a second workspace repeats the owner steps.

Onboarding

Shipped

A new workspace goes from empty to “Prime is standing and the AI is answering” without guessing what’s next.

How you can tell

  1. Three roles, three checklists: operator, owner and member each see only their own next step.
  2. Progress is marked as you go, so a half-finished setup picks up where it stopped.
  3. The last step is a real reply from Prime, not a tour.

LimitOnboarding covers the first workspace. Adding a second workspace repeats the owner steps.

Workspace settings. Timezone, approval posture (Always Ask) and catch-up behavior live on one page. The workspace clock previews its offset before you apply it. A stale save asks you to reload. Settings carry a record of who changed what and when. Settings apply per workspace. There is no organization-wide default yet, so each workspace is configured on its own. Status: shipped. Settings apply per workspace. There is no organization-wide default yet, so each workspace is configured on its own.

Workspace settings

Shipped

One page where your rules are set once, previewed before they apply, and never silently overwritten.

  1. Timezone, approval posture (Always Ask) and catch-up behavior live on one page.
  2. The workspace clock previews its offset before you apply it. A stale save asks you to reload.
  3. Settings carry a record of who changed what and when.

LimitSettings apply per workspace. There is no organization-wide default yet, so each workspace is configured on its own.

Templates. Each template lists its seats, ranks and reporting lines before you create anything. Every agent is created paused; you switch seats on one at a time. A seat starts working only after an explicit activation. Collections you can preview include Development team and Ticket Fixer / Prime loop. New seats start paused. Templates are applied once. Later changes to a template do not flow into teams already created from it. Status: shipped. Templates are applied once. Later changes to a template do not flow into teams already created from it.

Templates

Shipped

Stand up a whole team in minutes, previewed first, nobody switched on until you say so.

  1. Each template lists its seats, ranks and reporting lines before you create anything.
  2. Every agent is created paused; you switch seats on one at a time.
  3. Collections you can preview include Development team and Ticket Fixer / Prime loop. New seats start paused.

LimitTemplates are applied once. Later changes to a template do not flow into teams already created from it.

Connections and keys. The key ledger shows present, absent and deploy-only per key, with values always masked. Access needs three things to agree: the provider’s capability, an explicit grant, and the workspace’s write switch. Prompts only learn that a secret exists, never its value. The ledger and grants are live. MongoDB, Oracle and SQLite sit in the catalog beside GitHub and the rest, and those tiles still show as not yet wired. The public data route stays cannot-act until a saved connection, a grant, and dispatch. SQLite runs on a paired machine, not as a file this site opens. Oracle runs in thin mode only. Status: partial. The ledger and grants are live. MongoDB, Oracle and SQLite sit in the catalog beside GitHub and the rest, and those tiles still show as not yet wired. The public data route stays cannot-act until a saved connection, a grant, and dispatch. SQLite runs on a paired machine, not as a file this site opens. Oracle runs in thin mode only.

Connections and keys

Partial

Connect an account once, then decide exactly who may use it, without pasting a secret into chat.

  1. The key ledger shows present, absent and deploy-only per key, with values always masked.
  2. Access needs three things to agree: the provider’s capability, an explicit grant, and the workspace’s write switch.
  3. Prompts only learn that a secret exists, never its value.

LimitThe ledger and grants are live. MongoDB, Oracle and SQLite sit in the catalog beside GitHub and the rest, and those tiles still show as not yet wired. The public data route stays cannot-act until a saved connection, a grant, and dispatch. SQLite runs on a paired machine, not as a file this site opens. Oracle runs in thin mode only.

Stage 2 · Scout hosts

Do the work

Rooms, durable runs, planning, asides

Scout says: Send it and close the laptop. I keep going.

Rooms

Shipped

You decide how a conversation behaves, and waiting work is one glance away.

How you can tell

  1. Each room uses one speech mode: Floor, Addressed, Council, or Observe.
  2. A working line shows which agent is busy and on what, while the thread stays readable.
  3. Decision cards surface in the room where the question came up, with counts in the sidebar.
  4. Prime publishes a canvas2d diagram, comparison, or short animation with the reply. illustration_preview checks the document, illustration_lookup returns the saved message, revision, and hash, and illustration_read opens that revision.

LimitVoice and video are not part of a room. A preview is not saved until the reply publishes it. Private and aside transcripts are refused. Chat draws canvas2d only, and the illustration is not a workspace file. A missing messages:read grant on the read route is cannot-act.

Rooms. Each room uses one speech mode: Floor, Addressed, Council, or Observe. A working line shows which agent is busy and on what, while the thread stays readable. Decision cards surface in the room where the question came up, with counts in the sidebar. Status: shipped. Voice and video are not part of a room. A preview is not saved until the reply publishes it. Private and aside transcripts are refused. Chat draws canvas2d only, and the illustration is not a workspace file. A missing messages:read grant on the read route is cannot-act.

Durable runs. A message is stored before any agent starts on it, so a closed tab never loses the request. Work continues on the server; the room notes that you closed the tab instead of stopping. The answer is saved to the room and waits for you, with the steps that produced it. A run that needs your decision pauses at that point. It does not guess and continue. Status: shipped. A run that needs your decision pauses at that point. It does not guess and continue.

Durable runs

Shipped

Send the message, close the laptop, come back to the answer.

  1. A message is stored before any agent starts on it, so a closed tab never loses the request.
  2. Work continues on the server; the room notes that you closed the tab instead of stopping.
  3. The answer is saved to the room and waits for you, with the steps that produced it.

LimitA run that needs your decision pauses at that point. It does not guess and continue.

Planning and clarification. Clarifying questions come before the plan, so the plan answers the question you meant. Planning produces a saved file in the workspace, not a chat message that scrolls away. While planning, the agent may write to that one file and nothing else. Each save carries a receipt you can verify against the file. Plans are documents, not schedules. Turning a plan into scheduled work is a separate step in Routines. Status: shipped. Plans are documents, not schedules. Turning a plan into scheduled work is a separate step in Routines.

Planning and clarification

Shipped

Ask for a plan and get a real saved document, with writes held to just that file.

  1. Planning produces a saved file in the workspace, not a chat message that scrolls away.
  2. While planning, the agent may write to that one file and nothing else.
  3. Each save carries a receipt you can verify against the file.
  4. Clarifying questions come before the plan, so the plan answers the question you meant.

LimitPlans are documents, not schedules. Turning a plan into scheduled work is a separate step in Routines.

Asides. Side conversations between agents collapse to one line: who talked and how many messages. You can open any aside and read the full exchange. The main thread holds only what was decided or asked of you. Asides are between agents in the same room. Cross-room coordination goes through Command. Status: shipped. Asides are between agents in the same room. Cross-room coordination goes through Command.

Asides

Shipped

Agents coordinate without hogging the floor or burying you in side chatter.

  1. Side conversations between agents collapse to one line: who talked and how many messages.
  2. You can open any aside and read the full exchange.
  3. The main thread holds only what was decided or asked of you.

LimitAsides are between agents in the same room. Cross-room coordination goes through Command.

Stage 3 · Research Lead hosts

Build compounding memory

Memory and context, knowledge, skills

Research Lead says: I read what the workspace decided before I answer.

Knowledge

Shipped

What the workspace decided stays decided: versioned, auditable and visibly in use.

How you can tell

  1. Knowledge is published in numbered versions; a candidate waits until someone publishes it.
  2. “Why did this bot use v4?” has an answer: every reply records the version it read.
  3. Older versions stay readable, so a decision can be traced back to its wording at the time.

LimitA candidate stays unpublished until an owner or a commander with knowledge:publish publishes it. Any other seat can propose. A missing grant is cannot-act.

Knowledge. Knowledge is published in numbered versions; a candidate waits until someone publishes it. “Why did this bot use v4?” has an answer: every reply records the version it read. Status: shipped. A candidate stays unpublished until an owner or a commander with knowledge:publish publishes it. Any other seat can propose. A missing grant is cannot-act.

Memory and context

Partial

Stop re-explaining; every turn starts with the room, the open work and what is already decided.

  1. Each turn starts from the room’s chat history, its open tasks, the published knowledge that applies to that agent, and recalled earlier context.
  2. A panel between the composer and the chat history shows what context a reply included.
  3. Standing policy in published knowledge is trusted. Other knowledge cannot override it.

LimitEarlier context can include another room you can enter now, where that agent still has a seat. A private room also needs every participant’s consent. A paraphrase is not a match.

Memory and context. Each turn starts from the room’s chat history, its open tasks, the published knowledge that applies to that agent, and recalled earlier context. A panel between the composer and the chat history shows what context a reply included. Standing policy in published knowledge is trusted. Other knowledge cannot override it. Earlier context can include another room you can enter now, where that agent still has a seat. A private room also needs every participant’s consent. A paraphrase is not a match. Status: partial. Earlier context can include another room you can enter now, where that agent still has a seat. A private room also needs every participant’s consent. A paraphrase is not a match.

Skills

Partial

When a workflow works, make it repeatable; when a capability is missing, make the gap visible.

  1. A skill is a document with six required headings, so every skill reads the same way.
  2. A missing capability becomes a proposed skill, visible in the Inbox as a skill gap.
  3. Skills are reviewed before an agent may use them.

LimitSkills can be written, proposed and reviewed. Measuring how often a skill is used is not yet available.

Skills. A skill is a document with six required headings, so every skill reads the same way. A missing capability becomes a proposed skill, visible in the Inbox as a skill gap. Skills are reviewed before an agent may use them. Skills can be written, proposed and reviewed. Measuring how often a skill is used is not yet available. Status: partial. Skills can be written, proposed and reviewed. Measuring how often a skill is used is not yet available.

Stage 4 · Ticket Fixer hosts

Stay in control

Command, inbox, approvals

Ticket Fixer says: I file what I find. Merges follow your approval policy.

  1. CommandWho is on the desk, and what each seat may do.
  2. InboxWhat was filed: bug, wishlist, task, or skill gap.
  3. ApprovalsWhat waits for you, tied to the revision you saw.

Command. The org chart shows every seat, its rank and its reporting line. A report goes to the next rank up in the same workspace: no skipped rank, no cycle. Rank is not permission: a higher seat does not inherit the grants of a lower one. An action needs the scope on the seat and on the key it presents. Pausing a seat stops its work at once and is recorded. A paused seat cannot send, and a revoked seat or key is refused outright. Command shows one workspace at a time. There is no cross-workspace view. Status: shipped. Command shows one workspace at a time. There is no cross-workspace view.

Command

Shipped

Always know who is on the desk, what each seat may do, and who it reports to.

  1. The org chart shows every seat, its rank and its reporting line. A report goes to the next rank up in the same workspace: no skipped rank, no cycle.
  2. Rank is not permission: a higher seat does not inherit the grants of a lower one. An action needs the scope on the seat and on the key it presents.
  3. Pausing a seat stops its work at once and is recorded. A paused seat cannot send, and a revoked seat or key is refused outright.

LimitCommand shows one workspace at a time. There is no cross-workspace view.

Inbox. Four ticket kinds: bug, wishlist, task, and skill gap. A bug closes as verified only after the fix is checked, not when the fix is claimed. A blocked ticket says what it is waiting on. The Inbox tracks work inside the workspace. It does not sync with external trackers. Status: shipped. The Inbox tracks work inside the workspace. It does not sync with external trackers.

Inbox

Shipped

File a bug, wishlist or skill gap once, and “done” actually means done.

  1. Four ticket kinds: bug, wishlist, task, and skill gap.
  2. A bug closes as verified only after the fix is checked, not when the fix is claimed.
  3. A blocked ticket says what it is waiting on.

LimitThe Inbox tracks work inside the workspace. It does not sync with external trackers.

Approvals. A decision card names the action and the revision it applies to, such as “Merge revision 7f3a?” The decision is waiting on that exact revision. Status: partial. Approvals cover merges and external actions. Approval for spend limits is not yet available.

Approvals

Partial

The consequential calls wait for you, tied to the exact revision you looked at.

How you can tell

  1. A decision card names the action and the revision it applies to, such as “Merge revision 7f3a?”
  2. Approve and reject are both recorded with who decided and when.
  3. If the revision changes after you looked, the approval does not carry over.

LimitApprovals cover merges and external actions. Approval for spend limits is not yet available.

Stage 5 · Bridge hosts

Put it on autopilot

Routines

Bridge says: 06:00, every day. I leave a record.

Routines

Shipped

Work you would have to remember to start starts itself, on time, and leaves a record.

06:00Every day, in the workspace timezone. One record per run.

LimitA clock routine is daily or weekly in the workspace timezone, or every two hours from a UTC anchor. An event-only routine starts from its trigger URL or the desk Run button.

Routines. A routine runs on a schedule in the workspace timezone, such as 06:00 daily. Every run leaves a record: when it started, what it did and whether it finished. Status: shipped. A clock routine is daily or weekly in the workspace timezone, or every two hours from a UTC anchor. An event-only routine starts from its trigger URL or the desk Run button.

How you can tell

  1. A routine runs on a schedule in the workspace timezone, such as 06:00 daily.
  2. Every run leaves a record: when it started, what it did and whether it finished.
  3. A routine that needs a decision pauses and asks, like any other run.

Stage 6 · Major hosts

Reach out into the world

Major for Chrome

Major says: Only the tabs you hand me. I refuse fields I can identify as sensitive. Don’t give me credentials.

  • Release roomDelegated
  • docs.exampleDelegated
  • bank.exampleNot delegated

botscape.app/features/chrome-extension

Major for Chrome. Major works only in tabs you delegate; other tabs stay invisible to it. The extension refuses a field it can identify as a password, payment field, or one-time code, from the input type and selected attributes. Do not give it credentials. Every action in a delegated tab is logged to the room that asked for it. Status: partial. Major is available for Chrome on desktop. Other browsers are not yet supported.

Major for Chrome

Partial

An agent that can work in a real browser, but only in the tabs you hand it.

LimitMajor is available for Chrome on desktop. Other browsers are not yet supported.

How you can tell

  1. Major works only in tabs you delegate; other tabs stay invisible to it.
  2. The extension refuses a field it can identify as a password, payment field, or one-time code, from the input type and selected attributes. Do not give it credentials.
  3. Every action in a delegated tab is logged to the room that asked for it.

Leave the work. Keep the command.

Pick what your first team should do. Every agent starts paused, and nothing reaches the outside world until you say so.

Set up your deskRead the security model