Skip to content
Sadie

Access

Access is where you set the rules. Not the soft rules Sadie infers from your behavior; the hard ones you want enforced every time.

The surface sits on two tabs: Collaborators and Policies. They live together because both answer the same question: what is allowed to happen with Sadie’s outputs?

Collaborators

Section titled “Collaborators”

Studio workspaces are single-tenant by default. Every document lives inside a workspace that only you can see. To share one, you add a collaborator grant: an email address plus a scoped role.

Three roles:

  • Viewer can read documents in the workspace.
  • Commenter can read and add text-anchored comments.
  • Editor can read, comment, and edit.

Grants are local-only in v1. There is no SSO, no email invitation flow, no external identity provider. The person you grant to needs their own session on this deployment. Treat it as a trust-network primitive, not a hosted collaboration product.

You can revoke a grant at any time. Revocation is immediate and severs any active presence on the workspace.

Policies

Section titled “Policies”

Policies are deterministic rules Sadie cannot break. They complement your Soul’s allergies, which are soft and probabilistic. Allergies nudge the LLM; policies catch the output.

Every policy has a kind, an action, and a set of scopes:

  • Kind is phrase (literal, case-insensitive), regex (global), or semantic (evaluated by an LLM judge against a plain-English rule).
  • Action is rewrite (substitute the match), block (suppress the whole output), or warn (log and leave the text).
  • Scopes is any subset of chat, studio, and brief.

A policy scoped only to chat does not run in Studio or briefs. This lets you say “never apologize in chat” without affecting a Studio draft that legitimately includes the word “sorry.”

Studio’s policy enforcement is always record-only regardless of action: matches are logged, your text is never mutated. See Studio for why.

The view includes a Quick add section with curated presets: no em-dashes, no en-dashes, no negation framing, no AI openers, no servile openers. Each preset is one click to add and can be edited or disabled afterward.

Below the policy list, the Recent violations panel shows the last 25 matches across all surfaces so you can see what has been firing.

Section titled “Related”