All releases
major28 changes

v2.0.0 — Intelligence Engine

Jetwork evolves from a contact database into a relationship intelligence platform. Full scoring engine, knowledge graph, proactive AI insights, 15 config-driven integrations, universal webhook receiver, and real data pipeline — no more mock data.

New

(11)
  • Intelligence pipeline — 8-step post-sync engine: score contacts, backfill timestamps, auto-embed, infer relationships, extract knowledge, compute PageRank, predict decay, generate insights
  • Knowledge graph (brain_entries) — extracts company intelligence and topic expertise from contact data, powering AI queries with structured knowledge
  • PageRank network position scoring — identifies superconnectors, bridge contacts, and isolated clusters in your relationship graph
  • Decay prediction model — personalized half-life per contact based on interaction patterns, predicts when relationships will go cold
  • Proactive insight engine — 6 insight types: dormant VIPs, network gaps, concentration risk, multi-source superconnectors, geographic distribution, unstarred high-value contacts
  • Dashboard insights panel — AI-generated insights shown at the top of the dashboard with dismiss controls
  • GenericContactNormalizer — config-driven sync engine supporting 15 providers (Salesforce, Pipedrive, Freshsales, Intercom, Zendesk, Stripe, Shopify, Monday, Airtable, QuickBooks, GitLab, Discord, Linear, Eventbrite, plus existing 10)
  • Universal webhook receiver — POST /api/webhooks/ingest accepts normalized payloads from Zapier, Make, n8n, and custom integrations
  • Contact enrichment — domain-based company name extraction from email addresses, handles country TLDs and known domains
  • Guided import flows — step-by-step instructions for LinkedIn, X (Twitter), Facebook, Instagram data exports with direct links to each platform's export page
  • Pre-meeting AI briefings — generate dossiers for any contact with talking points, shared context, and relationship warnings

Improved

(5)
  • AI context builder enhanced — now includes brain_entries (knowledge graph), contact_metrics (PageRank, decay), and active insights alongside contacts and interactions
  • VCF parser upgraded — extracts address (city/country), handles multiple emails/phones/URLs, detects LinkedIn URLs, parses multi-line values
  • Connections page — 14 generic providers wired with sync buttons, guided import for social media platforms
  • Knowledge page — superconnectors now ranked by relationship score, dynamic blind spot detection from common industries
  • Dashboard — proper empty state with CTA to connect accounts instead of mock data banner

Infrastructure

(8)
  • New database tables: contact_metrics (RSI, NPS, decay, sentiment), insights (proactive AI), sync_queue (decoupled pipeline)
  • brain_entries extended with temporal columns (valid_from, valid_until, metadata) for tracking knowledge evolution
  • Performance indexes on contacts (company, sources, updated_at), interactions (contact+date, direction), relationships
  • All 20+ API routes use getUserId() from shared api-auth.ts — x-user-id header, cookie, and Bearer token fallbacks
  • Middleware injects x-user-id header after session validation for reliable auth across Vercel edge + serverless
  • Nango proxy resolves connection_id from end_user mapping — fixes mismatch between Supabase IDs and Nango connection IDs
  • All 11 integration sync files use service role client for DB writes — eliminates RLS policy violations
  • Data hook uses triple-method token retrieval: getSession(), onAuthStateChange(), and 3-second timeout fallback

Fixed

(4)
  • Removed all mock data fallbacks — data.ts, use-data.ts, claude.ts, and briefing route no longer fall back to fake 65-contact dataset
  • Fixed cookie forwarding in middleware rewrites — was losing maxAge, secure, sameSite attributes when copying Set-Cookie headers
  • Fixed cron job auth — replaced broken magic-link approach with x-user-id + service role Bearer token
  • Fixed domain enrichment — was extracting "Com" from short .com domains, now requires >3 chars and handles country TLDs

Stay in the loop

Get notified when we ship new features.