Console Activity: switch Workloads/Storage/Threads to server-side list behavior + filter bars + updated org switching

[rowan-stein]

User request

Update Console UI so Activity lists (Workloads / Storage / Threads) use server-side sort/filter/pagination and reflect updated authorization + cluster-admin org switching behavior.

Source spec: https://github.com/agynio/architecture/blob/main/changes/2026-04-24-activity-server-side-lists.md

Specification (architecture/product)

From:

• changes/2026-04-24-activity-server-side-lists.md
• product/console/console.md (Workloads, Storage, Threads, Resource Lists, Real-Time Updates)

Key requirements:

• Replace client-side sorting/filtering/search with server-side list APIs.
• Add filter bars (and search for Storage) per spec.
• Deterministic cursor pagination (UI treats page_token as opaque; changing sort/filter/search resets paging).
• Real-time updates:
  • On workload.updated / volume.updated, if any sort/filter/search is active, refetch the current page from server.
• Auth / org switching:
  • Cluster admins can context switch across all orgs via Organizations.ListOrganizations.
  • Non-admin users should use Organizations.ListMyMemberships.

Work items

1) Update generated clients

• After agynio/api changes land, update buf.lock / regenerate so console has:
  • Runners list envelope fields + name/attachment enrichment fields
  • Threads ListOrganizationThreads
  • Identity nickname batch lookup (if console needs it for any view)

2) Workloads Activity tab

• Render the columns per spec (Agent name + link, Runner name + link, Thread ID, Status, Containers, Started, Duration).
• Replace useListControls client-side search/sort with:
  • server-side sort + filter + pagination via runnersClient.listWorkloads
• Implement the filter bar:
  • Agent multi-select (by name)
  • Runner multi-select (by name)
  • Status multi-select
  • Started date range
• Ensure changing sort/filter resets pagination and refetches from the first page.

3) Storage Activity tab

• Replace current client-side search/sort and per-volume attachment queries with server-side list behavior:
  • use runnersClient.listVolumes with filter/sort/page_token
  • use volume_name + attachments[] from the list response
• Implement:
  • Search: case-insensitive substring match on volume name
  • Filters: Status, Runner, Attached-to kind (agent/mcp/hook/unattached)
  • Sortable columns: Name, Size, Status, Created
• Render attachment column as: primary attachment name + “+N more”.

4) Threads Activity tab

• Implement the org threads list view using threadsClient.listOrganizationThreads:
  • columns: ID (truncated), participants (@NickNames), message_count, status, created date
  • filters: status, participant, created range
  • sorts: created, updated, message_count, status
  • cursor pagination

5) Real-time update refetch rules

• Update notifications subscription/invalidation logic so:
  • When any sort/filter/search is active on Workloads/Storage, workload.updated / volume.updated triggers a refetch of the current page (instead of patching local state).
  • With default (no active controls), keep existing in-place update behavior if still desired.

6) Cluster-admin org switching

• Update org context selection logic:
  • Cluster admins: use Organizations.ListOrganizations
  • Non-admin users: use Organizations.ListMyMemberships
• Ensure Activity pages (and any other org-scoped pages as needed) respect the selected org context.

Acceptance

• Workloads/Storage/Threads Activity lists are fully server-driven for sort/filter/pagination (client does not sort/filter across pages).
• UI does not display raw agent/runner/volume IDs when a server-provided name exists.
• Real-time update behavior matches the spec.
• Cluster admins can switch across all orgs; non-admins only see their memberships.

[rowan-stein]

Additional note re Notifications (docs vs current code):

• Console currently does notificationsClient.subscribe({}) (empty rooms) and filters only by envelope.event. Gateway/Notifications currently behave like a firehose.
• The Notifications issue will likely enforce room-filtered + room-authorized subscriptions, making SubscribeRequest.rooms semantically required.

Console-side implications:

• Update useNotifications to accept rooms: string[] and call:
  • notificationsClient.subscribe({ rooms })
• Activity list pages should subscribe to org-scoped room:
  • rooms: ["organization:${organizationId}"]
• For Workloads list invalidation, prefer listening to workload.updated (published to org room) rather than workload.status_changed (currently workload-room only).
• Storage list should listen for volume.updated once Runners publishes it to the org room.

[rowan-stein]

Additional user-reported UI gaps (Workloads list at organizations/:id/activity/workloads):

• Agent ID is displayed instead of agent name.
• Runner ID is displayed instead of runner name.
• Missing a clear "View workload" action/button.

These must be fixed in the existing PR #90 (no legacy fallbacks).

[rowan-stein]

Implemented and merged via #90