Skip to content

Admin

Admin (/admin) manages who can use an instance and with which role, and the settings that govern FACE itself on that instance: its guard rules, whether anything may leave the instance, the standing rules and the decision log.

Open Settings at the foot of the rail. The Account tab has an Admin card with the Admin entry, shown only when the server reports that you are an admin of the selected instance. The server checks your role again on every call, so the page cannot be used by someone without it.

Roles are Admin, Editor and Viewer.

Members and roles

The list shows every person who holds a grant on the instance. You marks your own row. You cannot change your own access. A row marked Pending has been filed and applies when the platform reconciles.

  • Change role lets you pick a new role, then Save.
  • Remove asks Remove {email} from this instance? They lose access on their next request. Confirm with Remove access.

No person holds a grant on this instance means the list was read and is empty. Service accounts are counted but not listed.

Invite someone

Enter the address

Under Invite someone, type the person’s Email address.

Pick the role

Choose Admin, Editor or Viewer under Role. The default is Viewer.

Select Invite

Invited {email} as {role}. confirms the invitation. The person can then sign in and choose this instance.

If you have not chosen an instance yet, the page shows the instance picker first. Only an admin of this instance can manage its members. means your role on the selected instance is not Admin.

Grounding gate

GROUNDING GATE holds FACE’s own guard rules: the rules every agent call runs through, such as the OWASP and OpenBias rule sets. They govern the application, not your business, so they live here and not on Why. Your business rules are on Why.

  • Show the {count} rules lists the rule inventory with each rule’s CATEGORY, RULE and SEVERITY. Hide rules folds it away.
  • Type a term or hypothesis and select Check this text. It is checked against the same rules that guard the agent call. Nothing is sent and nothing is logged. The result reads No rule fired or Would be refused, with how many rules were evaluated. Not checked yet means nothing was checked; The check couldn’t run means it failed.

There is no “percent grounded”: a rule fires or it does not. Only an admin of the instance sees the gate, and every use is audited.

External web search

External web search is always on. Only a query a person types, or one of the confirmed public names an admin lists here (carriers, ports, competitors), is sent to public search engines; a query built from your data is made only from those names, and only when the text mentions them. Schema, table, column and row content never is. Each result says whether an external source was consulted, and which.

Public registries and market data

Public registries & market data are always on. The market and due-diligence agents contact public registries and statistics with the identifiers a person types, and nothing from your records; each result names every source, its publisher, licence and retrieval time. An operator can turn live public data off for a whole deployment, for example an air-gapped one; each source then says it was not checked, and why. Thresholds for this organisation sets the limits those agents apply. There are no built-in limits: a field neither your deployment nor your organisation has set reads No target set, and the agents show that figure as measured and flag nothing on it.

The same list holds the cycle-service level a site’s safety stock is sized for (for example 0.95). With none set, the site page computes no safety stock or reorder point, and says the level is not set.

Every change to the public names or the thresholds is audited with its author.

Standing rules

Standing rules lists the decisions made into policy: Decisions made into policy. Each takes effect only after a second, different admin approves the exact text; any admin can revoke it. A rule is proposed from an approved card on Actionable Twins.

  • Each rule shows its conditions (a later card must match all of them), its tier (Guarded — one named approver or Autonomous — runs with no approver), who proposed it and its state: Awaiting second approval, Active or Revoked.
  • Approve rule approves the exact text. The proposer cannot approve their own rule.
  • Revoke ends a rule. Any admin can revoke one.
  • No standing rule has been proposed. means the list is empty. Standing rules cannot be read, so none is in effect means the read failed, and then no standing rule applies.

A standing rule exists only as records on the hash-chained decision log. If the log cannot be read or its chain does not verify, no standing rule is in effect, and the instance policy alone decides. Where the instance policy and a standing rule both match, the stricter one wins.

Decision log

Decision log lists every approval, rejection, refusal and outcome, hash-chained and written before anything runs. Each record names the card, who decided, the inputs it cited and the model and prompt template involved. Chain verified ✓ ({count} records) confirms the chain is intact; Chain broken at record {seq} names where it is not. No decisions recorded yet. means the log is empty; a failed read says the log is unknown, not empty.

RPCs behind this page

  • runink.ui.access.v1.AccessService: ListMembers, InviteMember, SetMemberRole, RemoveMember.
  • MetasearchService.EvaluateBias, ListBiasRules, for the grounding gate.
  • TwinsService.ListStandingRules, ApproveStandingRule, RevokeStandingRule, for the standing rules.

See the API reference.