Defined append-only local capture architecture for Office Window

2 min read

Defined append-only local capture architecture for Office Window

  • Day: 2026-06-21
  • Time: 12:00 to 12:10
  • Project: Dev
  • Workspace: WP 2: Operational
  • Status: Completed
  • Priority: HIGH
  • Assignee: Matías Nehuen Iglesias
  • Tags: Nextjs, Systemd, Append-Only, Event-Sourcing, Capture-Pipeline, Local-Service

Description

Session Goal

Clarify the deployment and data-flow strategy for Office Window so the scaffold can be closed into a stable, local-only operational state without risking mutation of canonical office data.

Key Activities

  • Reviewed guidance to lock value by committing the current work, ensuring a reproducible build, and running the app as a persistent local service on 127.0.0.1 instead of publishing it publicly.
  • Consolidated the deployment closure pattern around Next.js + systemd user service: build first, then run locally with next start, and only activate the service once the production build is valid.
  • Reframed the UI/automation architecture from passive observability to controlled input:
    • the UI should not edit canonical state directly;
    • it should emit append-only events that can be safely reingested later.
  • Defined a safe write path for human feedback via a separate append-only inbox (/capture and POST /[[api]]/capture) storing JSONL events rather than mutating office state.
  • Extended the capture design to row-linked voice/text capture, where audio or text is associated with a specific queue row but remains an immutable record pending later reingestion.
  • Identified a pipeline risk: the capture inbox can decay if raw intake, AI processing, and reingest outcomes are not separated into distinct stages.

Achievements

  • Established a coherent local-only deployment closure for Office Window with reproducible build, bookmarkable access, and persistent service operation.
  • Clarified the event-sourcing style boundary: append-only capture for human feedback, with canonical state protected from direct UI mutation.
  • Mapped the capture workflow into a multi-stage lifecycle (raw inbox → transcription/processing → routed artifact candidate → reingest), which reduces ambiguity and supports safer automation.

Pending Tasks

  • Implement or verify the systemd user service and fixed-port local startup flow.
  • Add the append-only capture endpoint/UI and JSONL inbox storage.
  • Build the reingestion worker/stages so captured items do not remain stranded in the inbox.
  • Validate the end-to-end capture loop with tests and commit the finalized scaffold.

Evidence

  • source_file=2026-06-21.sessions.jsonl, line_number=3, event_count=0, session_id=5e1e2f50c5bff3aa63bb3ac6ce944a65fd2ceac8470580afe582adc3c959bbce
  • event_ids: []

Graph View

Backlinks