The Temple Archives is the telemetry side of the fleet — the long-running listeners that subscribe to MQTT topics, persist what they hear, and project it through a Farsight API that other systems can query.

It is not a real-time visualization surface. That's the Galaxy Map and HoloVids. The Archives are for the slow, accountable, queryable view: "what was happening on this topic at this time, and where did the data come from?"

What's in the Archives

Config — archives-config.yaml

farsight:
  baseUrl: "http://farsight.temple-archives:8080"
  retention:
    default: "30d"
    overrides:
      - { topic: "fleet/critical/+", retention: "365d" }

feeds:
  - id: nodered-fleet-status
    topic: "fleet/+/status"
    publisher: "nr-prod-01..05"
    shape: { instance: string, status: string, ts: number }
    cadence: "5s"
    persistence: rolling-30d

  - id: restreamer-events
    topic: "restreamer/events/#"
    publisher: "datarhei/core"
    shape: { event: string, processId: string, ts: number, payload: object }
    cadence: "burst"
    persistence: rolling-7d

The file is editable in Settings → Temple Archives and committed to git. Adding a new listener means adding an entry here, then deploying — there's no "add a listener from the UI without committing" path on purpose. Persistent telemetry is a slow, deliberate change.

The HoloNet relationship

HoloNet (on the Force page) is the browse surface for MQTT. The Archives are the persist + query surface. They share the same broker (TBMQ) and the same topic map, but they're independent — HoloNet doesn't need the Archives to be online, and the Archives don't need HoloNet to exist.

If an operator wants to debug a single topic in real-time, HoloNet is the right tool. If an operator wants to ask "what did this topic look like last Tuesday at 3am", that's the Archives.

Routes

• GET /api/republic covers the Republic side.
• The Archives plane does not yet have its own dashboard-side route — Farsight is proxied at the cluster level and queried directly by the client. This is a deliberate optimization (Farsight responses can be large) and a candidate for a server-routed caching layer later.

Operational notes

• TBMQ + redis-cluster. If the host node carrying a redis-cluster pod restarts, the nodes.conf file in that pod's PV can be corrupted, which surfaces as the Archives losing subscriptions. Recovery is FORGET + MEET + REPLICATE against the broken peer — the runbook lives in the team's bd knowledge base.
• Snapshot ingest path is separate. Camera-feed snapshots go through POST /api/restreaming/snapshots and live in ReductStore — they are not in the Archives. The two collections have different retention rules and different access patterns.

Built on