Skip to main content
Personal key administration over the API.
  • GET /api/v2/api-keys/admin/personal-keys lists every Personal key and host sign-in in the deployment with apikeys:read at platform scope. A row never names its owner and never carries a key’s secret or hash.
  • POST /api/v2/api-keys/admin/personal-keys/revoke revokes any person’s Personal key with apikeys:manage at platform scope and tells the owner, as the Console does. A missing or non-Personal key answers 404; an already revoked key answers 409.
  • A Personal key’s audit rows now record who decided the change: a revoke through the API records a platform key as a system actor and a Personal key as its owner. The Console’s own rows are unchanged. Both routes are reachable from MCP through forge_call_operation in the governance tool group.
My blocked tasks and workforce integrity findings over the API.
  • GET /api/v2/planning/admin/my-blockers and GET /api/v2/planning/admin/could-help run the Blueprint’s two person views for a Personal key, which acts as its owner; any other key is refused with 403 and personRequired: true. Each lists only ordinary project tasks whose project the key may read (projects:read at the project’s home), names no assignee, and says partial when a bound ran out. See My blocked tasks.
  • GET /api/v2/people/admin/integrity/findings lists a workforce integrity run’s findings, at most five a page, optionally of one classification or status, with workforce:read. A finding answers its invariant, classification, status, decision and the ids it affects; its key, evidence, resolved names, proposed repair and decision reason stay in the Console, since they can name a person. See People and members.
Spending limits and the budget tree over the API.
  • GET /api/v2/costs/admin/limits/all lists every scope’s spending limits a page at a time, with cost:read at platform scope. A limit never answers its scope id.
  • GET /api/v2/costs/admin/budgets/tree reads the budget hierarchy as a tree of up to 200 budgets, and refuses a larger hierarchy with 409 conflict rather than answer part of it. It answers each budget’s period spend, burn rate, days until exhausted and triggered alert count.
  • The Console’s emergency-stop switches still list every limit, and its budget analytics tree is refused past 200 budgets, as the route is. See Costs and budgets.
Chat attachment content over the API.
  • POST /api/v2/files/admin/chat-attachment-content answers the bytes (base64) of one of a chat channel’s attachments with chat:read at the channel’s home. It never answers a download URL. An attachment the channel attachments list leaves out answers 404, and a file of more than 4,000,000 bytes answers 409.
  • The read is recorded under the key, or a Personal key’s owner, before any byte is read. See Files.
Dashboard widget data permissions.
  • A Console dashboard widget now shows its data only to a viewer who holds, at the platform scope, the permission the Console’s own page for that data needs: cost:read for spend and tokens, workflows:read for active runs, audit:read for interventions, factories:read for active factories and agents:read for agent completions. The Morning Briefing needs all four of cost:read, workflows:read, audit:read and agents:read. A viewer without the permission sees the widget’s empty state. Before this change any signed-in person could read these deployment-wide figures by adding the widget to their own dashboard.
  • When a widget’s data source changes, the data fetched for the old source is cleared; the widget shows its empty state until the next scheduled refresh, within 15 minutes, or its refresh button. Display-only edits keep the data.
  • Active-run and agent-completion widgets now show run content masked, as the Runs and Delegated work pages do.
  • See Forge Console for the default widgets.
An External App's agents over the API.
  • GET /api/v2/external-apps/admin/owned-agents lists the agents an External App owns, newest first, with apps:read at platform scope. It answers only agents homed where the key also holds agents:read, and leaves the app’s users’ Personal Agents out, as on every agent route.
  • See External apps.
Uploads bound to a knowledge scope over the API.
  • POST /api/v2/files takes optional scopeType, scopeEntityId and source query parameters. A named scope binds the upload to that knowledge scope, the binding a Console upload makes, beside its binding to the credential’s org unit.
  • A named scope also needs files:write at the scope’s home (the deployment root for a scope with no org-unit, factory or project home); without it the upload answers 403 and nothing is stored. A scope naming a top-level org unit also makes the file readable in that org tree. An upload without scopeType is unchanged. See Binding an upload to a scope.
A paged model catalog with usage over the API.
  • GET /api/v2/models/admin/catalog/list-with-usage lists catalog models a page at a time (agents:read), with the Console’s search, provider, status, group and sort filters. Each model carries agentCount and assignedAgents: only agents the key may read, never a personal or unhomed agent.
  • Each page takes limit (at most 100) and cursor, and answers meta.hasMore and meta.nextCursor. See Models.
Notification event types over the API.
  • GET /api/v2/notifications/admin/event-types lists every notification event type with its per-channel delivery and which channels have a provider. It needs all of audit:read, grants:read and policies:read at platform scope, as the Console’s event matrix does; a key missing any one gets 403 naming it. See Notifications.
  • An admin route’s 403 now names the permission the request was refused on. Before, a route that needs more than one permission always named its first.
Recent Slack sends over the API.
  • GET /api/v2/notifications/admin/slack-sends lists the most recent outbound Slack sends with audit:read at platform scope. A send never answers its channel or its error text.
  • The route is reachable from MCP through forge_call_operation in the governance tool group. See Notifications.
Your own Personal keys over the API.
  • GET /api/v2/api-keys/admin/own-keys lists your own Personal keys and host sign-ins, active and revoked.
  • POST /api/v2/api-keys/admin/own-keys/revoke revokes one of your keys; an already revoked key answers 409.
  • POST /api/v2/api-keys/admin/own-keys/delete deletes one of your revoked keys and disables its role grants; an active key answers 409.
  • They need no permission, only a Personal key whose own scopes reach apikeys:read (list) or apikeys:manage (revoke and delete); otherwise 403 names that key. They act only on your own keys: another person’s key, a missing key and a non-Personal key answer 404. Any credential but a Personal key gets 403 with personRequired: true. A revoke or delete is audited as you, through your Personal key. All three are reachable from MCP through forge_call_operation in the governance tool group.
A person too large to check in one request reads as out of reach.
  • Deciding whether a person is in a machine caller’s reach reads each of their worker seats. A person with more seats and assignments than one request can read used to fail the request on the platform’s read limit. Single reads and writes now answer that person exactly as one out of reach: the same 404, and nothing is written. A workforce repair applied through the API stops BLOCKED with REPAIR_OUT_OF_REACH.
  • The member, external tool identity and worker lists end a page before such a person and return them at the start of the next page, so nobody in reach is left out. Only a person still too large at the start of a page is omitted.
  • POST /api/v2/workforce/createPerson and POST /api/v2/people/admin/workers/create fail as before (500) when the email belongs to such a person, and write nothing.
External tool identities over the API.
  • GET /api/v2/connections/admin/external-tools/identity reads one external tool identity, and POST .../identity/update, .../identity/status and .../identity/verify rename it, change its scopes or status, and probe it against its provider, with platform:manage at platform scope. An identity never answers its credential.
  • An identity owned by a person a key may not reach (a person seated in or under a customer organization) answers 404, as a missing one does, and is never changed or probed. See Connections.
Member and external tool identity lists over the API.
  • GET /api/v2/people/admin/members lists members a page at a time (grants:read): each person’s seats, emails and whether their sign-in is linked.
  • GET /api/v2/connections/admin/external-tools/identities and .../identities/by-principal list external tool identities a page at a time (platform:manage at platform scope), without their credentials.
  • Each page takes limit (at most 50) and cursor, and answers meta.hasMore and meta.nextCursor. A person a key may not reach (seated in or under a customer organization), and an identity they own, is left out of the page, and a page also ends early when reading more people would exceed its read limits, so a page may be shorter than limit while meta.hasMore is true. See People and Connections.
Apply a workforce repair over the API.
  • POST /api/v2/people/admin/integrity/repairs/apply applies a previewed workforce repair run whose plan hash still matches (workforce:manage).
  • A run whose plan names a person a key may not reach (seated in or under a customer organization), or a seat, position or org unit placed in or under a customer organization, answers 404, as a missing run does, and is left unchanged.
  • The repair runs on the scheduler after the call returns. Before each step that writes, it checks reach again. If anything in its plan has left the key’s reach since it was applied, the run stops with status BLOCKED and error code REPAIR_OUT_OF_REACH, and that step writes nothing. See People.