Skip to content

Guardrails built into the platform.

Bots work in real systems, so the protections live in the platform itself rather than in the model's judgment. Here is what is in place.

Many workspaces, sealed apart.

Teams and agencies split their work into workspaces. Nothing in one is visible from another.

Your organizationWorkspaces never share data
  • Head office

    • Bots
    • Memory
    • Files
    • Browsers
    • Sign-ins
    • Connections
    • Rules
    • Spending limits
  • Client A

    • Bots
    • Memory
    • Files
    • Browsers
    • Sign-ins
    • Connections
    • Rules
    • Spending limits
  • Client B

    • Bots
    • Memory
    • Files
    • Browsers
    • Sign-ins
    • Connections
    • Rules
    • Spending limits

Security details

Each workspace is sealed off

  • A workspace holds its own bots, memory, files, browsers, sign-ins, connections, rules, and spending limits.
  • The database checks access on every row, for people and for the background workers that run your bots.
  • Each bot gets its own browser, and a browser is never shared across workspaces.
  • The live view of a bot's browser opens only for signed-in members of its workspace.

Approvals cover the exact action

  • An approval is tied to the specific tool, details, and target you saw. Change any of them and it no longer applies.
  • Each approval works once and expires. Editing an action creates a new approval for the edited version.
  • Approve-and-remember is limited to one kind of action, a maximum count, and an expiry date.
  • Only people in your workspace can approve. Text in an email, web page, or file cannot approve anything.

A vault the model never reads

  • Saved sign-ins are stored encrypted in a secrets vault. Once saved, the password is never shown again, even to you.
  • Bots see a sign-in's label, not its value. The password is typed into the bot's browser outside the model.
  • Codes and verification steps are handed to a person, and saving one-time-code seeds is optional per sign-in.
  • Changing passwords or security settings is always left to a person.

Masked on screen

  • Password fields are masked in every screenshot and live-view frame that is sent to a model.
  • Rules are enforced by the platform, so nothing a bot reads in a page, file, or email can change them.

An audit log of who did what

  • Approval decisions, takeovers and handbacks, saved sign-ins, and workspace changes are recorded with who did them and when.
  • Every step a bot takes is logged on its run, so you can see exactly what it did.

Templates copy setup, not data

  • Installing a team template copies its bots, instructions, skills, routines, and rules.
  • Sign-ins, conversations, and browser sessions never come with it, and memories only do when they were marked safe to share.

Sensible defaults on day one.

Every bot starts with these rules. Admins can adjust them per workspace or per bot, and a rule locked higher up can only be made stricter.

What a bot does on its own

Default rules
  • Read informationDoes it
  • Write draftsDoes it
  • Update tasks, notes, and files hereDoes it
  • Change records in your appsAsks you first
  • Send emails, texts, and postsAsks you first
  • Delete data in your appsAsks you first
  • Share files or data elsewhereAsks you first
  • Spend moneyAsks you first
  • Install apps and skillsAsks an admin
  • Change passwords or security settingsAlways yours
Your first job ticket1Forgebot

See the guardrails for yourself.

Pick a plan, name your first bot, and hand it a job. Prices are in Canadian dollars, billed monthly.