Mailboxes, integration tokens, and message content are organization
assets. Keying their DEK by the connecting user meant offboarding that
user made every ciphertext they created unreadable.
Migration 000039 drops user_encrypted_keys and creates
organization_encrypted_keys keyed by organization_id. The new table has
no FK on purpose: platform secrets live under the zero UUID (no
organizations row), and DEK rows must never cascade-delete because a
lost DEK is unrecoverable.
Pre-production, so there is no data migration; ciphertexts sealed under
the old per-user DEKs are abandoned with the table.
Adds the persisted notification feed, user preferences, realtime notification events, and advanced-service notification hooks for replies and deliverability health events.
Replace the client-side, 100-row-capped deals board as the default view
with a server-driven cross-pipeline table so totals stop lying at scale.
- POST /crm/deals/search: faceted filter body (query, status, pipeline,
stage, owner, campaign, value range, close-date range), whitelisted
sort, offset pagination, and Contact/Stage/campaign hydrated via JOIN.
- POST /crm/deals/summary: COUNT + SUM(value) per status and per stage
over the same filter, with a mixed-currency guard, so header and board
column totals are true server aggregates.
- DealsTable: cross-pipeline default view (infinite scroll, server
filters/sort, honest summary stats) with a Table|Board toggle; the
kanban stays as the single-pipeline working view.
- Deal attribution: campaign_id + source_mailbox_id columns + indexes
(migration 000022) as groundwork for revenue lineage.
Add the conversation-label join table and expose thread label fields on the Unibox response models so inbox rows and overview data can carry category metadata.
Add an organization credit ledger with idempotent consumption, plan credit grants, Anthropic/OpenAI writing providers, and a gated generation endpoint for campaign writing assistance.
Scope campaign A/B variants to individual sequence steps and select step variants deterministically per contact and step. Keep campaign-level variants on the existing assignment flow.
Store per-step branching conditions, validate branch targets, and select campaign contact steps by following the flow graph. Preserve linear progression for steps without branch conditions.
Store per-day schedule windows on campaigns and validate updates before persistence. Use the schedule windows in campaign scheduling, including legacy day/time derivation for existing campaigns.
Treat warmup content as a shared library in settings and admin tables, and realign the conversation pick index with segment-based selection.
Add generation job details and fix warmup placement displays so backend percentages and provider counts render correctly.
Add OpenAI Batch API support for warmup content generation, including job metadata, polling, cancellation, and completed-batch ingestion.
Share the generation prompt between sync and batch modes and humanize/lint generated threads before storing them in the warmup content library.
Store delayed warmup engagement actions in Postgres and drain them from the consumer so read, important, and star actions survive worker restarts.
Keep foldering and spam rescue immediate while routing delayed actions to the mailbox's current worker at fire time.
Track complaint events in campaign progress and evaluate rolling bounce and complaint rates before auto-pausing campaigns.
Emit an early warning webhook when campaign deliverability enters the warning band.
Matthew Meszaros