Unclassified

IntegrateWise — Complete Platform Features

A Knowledge Workspace over the Spine, powered by AI.
Everything the platform does, organized by layer.

Canonical Internal Doctrine Guardrail

User Workbench = the single projection-native customer/product surface.
Twin = silent contextual AI Co-worker behavior inside the Workbench, not a
separate Workbench, panel or shell.
Governance = shared platform module and hard execution/promotion gate.
Neon = canonical tenant truth; D1/KV = bounded operational state; R2 =
evidence bytes.
When this doc uses public-facing language like "Workbench", interpret that internally as the User Workbench unless stated otherwise.
Twin runtime surfaces use the canonical Forest + Paper system language.
Midnight Executive is investor-deck / fundraising language only.

1. THE SPINE (Single Source of Truth)

The Spine is the canonical tenant data model. Connector data first enters the
Operational Spine, is normalized, resolved and staged, and reaches canonical
truth only through promotion and Governed Spine Writers.

What it does

Entity-based storage: 150+ entity types in tenant-scoped entity tables;
department never determines a physical table namespace.
Tenant isolation: Row-Level Security on Spine DB. Workspace A cannot see Workspace B. Ever.
Entity type routing: Pipeline maps connector-native objects to canonical
entity contracts and records source bindings and lineage.
Adaptive projections: observed source fields are mapped through governed
field definitions; RBAC projections evolve without mutating the physical
schema per department.
Audit trail: Every write logged with source, timestamp, and trace ID.
12 role/department projection configurations: Customer Success, Sales,
Marketing, RevOps, Product Engineering, Finance, Support, Procurement, IT
Admin, BizOps, Education and Personal select curated views, KPIs, fields and
capabilities.
Industry overlays: industry context changes vocabulary, fields, KPIs,
recommended connectors and projection composition—not storage isolation.

Entity projections by department

The table below is a projection catalog. The listed entities remain tenant-level
canonical entities and may appear across departments when RBAC permits.

Department	Entity Types	Examples
Customer Success	16	account_master, success_plan, risk_register, engagement_log, renewal, stakeholder_outcome, initiative, capability
Sales	12	deal, opportunity, lead, pipeline_stage, quote, sales_sequence, call_log, competitor_intel
Marketing	12	campaign, audience, landing_page, email_campaign, social_post, ab_test, attribution_touchpoint
RevOps	8	quota, forecast, territory, comp_plan, revenue_metric, attribution, segment_rule
Product Engineering	10	feature, bug, sprint, release, repository, pull_request, deployment, roadmap_item, feedback
Finance	12	invoice, payment, expense, budget, revenue_entry, tax_filing, financial_report, vendor, cost_center
Support	8	ticket, csat, sla_policy, escalation, queue, agent_performance, knowledge_article
IT Admin	12	system, device, user, change_request, vulnerability, backup, certificate, network_device, license
BizOps	6	activity, workflow, okr, cross_dept_initiative, ops_metric
Procurement	6	purchase_order, spend_category, savings_initiative, approval_request, compliance_check
Education	8	student, course, assignment, grade, attendance, discussion, learning_objective, intervention
Personal	4	calendar_event, bookmark, daily_reflection, note

2. THE PIPELINE (8-Stage Normalizer)

Every record from every source passes through the same 8 stages. No shortcuts. No bypasses.

The 8 stages

Stage	Name	What it does
S1	Analyze	Detect entity type, resolve tenant schema, fingerprint data
S2	Classify	Assign category, priority, sensitivity level
S3	Filter	Check entity type against tenant schema — reject if not allowed
S4	Refine	Normalize field names, split composite fields, extract cross-reference IDs
S5	Extract	Map to canonical schema, keep only priority fields (schema-driven)
S6	Validate	Required field checks, type validation, integrity rules
S7	Sanity	Anomaly detection, business rule validation (score must be ≥ 70 to pass)
S7.5	Identity Resolution	Detect potential cross-source duplicates (never auto-merge)
S8	Sectorize	Route to correct Spine table, write canonical record, enqueue downstream

Pipeline features

Retry with backoff: Each stage retries 3 times with exponential backoff before sending to DLQ
Dead Letter Queue: Failed messages preserved with full context (stage, error, payload, trace ID)
Idempotency: Content-hash dedup keys prevent duplicate processing
Version tracking: Each entity gets a monotonic version number for conflict resolution
Stage-level metrics: Latency, throughput, and error rate tracked per stage
Schema-driven extraction: S5 only extracts fields defined in the tenant's schema — no data leakage

3. ENTITY 360 (Unified Data Surface)

Entity 360 is the single API that assembles the complete picture of any entity. Twin reads ONLY from Entity 360 — never from scattered sources.

6 layers assembled in parallel

Layer	Source	What it contains
Truth	Spine (Spine DB)	Canonical entity data with freshness indicator (live/recent/stale)
Context	Knowledge service	Emails, documents, meeting notes, conversation summaries (Flow B)
Signals	Signals table	Active alerts with severity, evidence chain, confidence score
Memory	Knowledge consolidated_memories	AI-extracted knowledge: decisions, preferences, insights, rules (Flow C)
Goals	Goals table	Linked goals with progress, status, target dates, health scoring
Relationships	Entities table	Parent/child/related entity graph (account → contacts, etc.)

API

GET  /api/entity360/:id                    — single entity, all layers
POST /api/entity360                        — single with layer selection
POST /api/entity360/batch                  — multiple entities
GET  /api/v1/cognitive/twin/360/:id        — entity + twin trigger insights

Features

Parallel fetch: All 6 layers fetched simultaneously — assembly time < 200ms
Layer selection: Request only the layers you need (e.g. just truth + signals)
KV cache: 60-second TTL on Cloudflare KV for hot entities
Freshness scoring: Truth layer tagged as "live" (< 1h), "recent" (< 24h), or "stale" (> 24h)
Source attribution: Every data point traces back to its origin system
Progressive hydration: Profile loads first, context and signals load async

4. TWIN TRIGGER ENGINE (AI Reasoning)

The Twin evaluates Entity 360 data against a set of triggers to produce constrained, evidence-backed insights.

10 triggers

Trigger	Fires when	Severity	Example output
health_drop	Health score drops 15+ points	Critical	"Acme health fell from 82 to 48"
renewal_approaching	Renewal within 90/60/30 days	Critical/High/Medium	"Renewal in 32 days, health 48"
engagement_drop	No activity for 30+ days	High/Medium	"No activity since March 2"
arr_change	ARR changes 10%+	High/Info	"ARR dropped 22% ($125K → $97K)"
goal_at_risk	Linked goal status = at_risk	High	"Goal 'API adoption' at 23% progress"
support_escalation	5+ open support tickets	Critical/High	"8 open tickets, 3 are P1"
stale_data	Entity data older than 24h	Medium	"Last sync: 3 days ago"
memory_conflict	2+ low-confidence AI decisions	Medium	"Conflicting memories about budget"
context_gap	High-value account, 0 context items	Medium	"$125K ARR, no emails or docs linked"
signal_cluster	3+ critical/high signals active	Critical	"5 signals firing simultaneously"

Guardrails

Max 3 insights per entity per day — prevents alert fatigue
Severity-first sorting — critical always surfaces before medium
Evidence required — no insight without at least 1 evidence item from Entity 360
Four-question standard — every insight answers: what happened, why it matters, what to do, what happens if ignored
Source layer attribution — every evidence item tagged with origin layer (truth/context/signals/memory/goals)
Confidence scoring — 0.60 to 0.95 based on evidence strength and source count

5. THREE DATA PATHS

Structured — Truth (CRM, billing, support → Spine)

Connector OAuth → Loader (creamy/full/delta) → Pipeline Queue → 8-Stage Normalizer → Spine

Handles: accounts, contacts, deals, tickets, invoices, subscriptions
Sync modes: creamy (30 days), full (all history), delta (changes only), webhook (real-time)
All data becomes canonical truth in the Spine

Unstructured — Context (emails, docs, meetings → Spine + Knowledge)

Connector OAuth → Loader → Pipeline → Spine (metadata) + Knowledge (embeddings)

Handles: Gmail, Slack, Teams, Notion, Google Drive, Dropbox, OneDrive, Calendly
Structured metadata goes to Spine, unstructured content goes to Knowledge for embedding
Feeds Entity 360 Context layer

AI Sessions — Approval-Based Personal Memory (reasoning → Knowledge only, NEVER Spine)

AI Source → D1 Buffer → Triage Bot → Confidence Scoring → Human Approval → Knowledge → Entity 360 Memory

Handles: MCP, OpenAI, Claude, Moonshot, Grok, Gemini, any AI source
AI-generated content NEVER writes to Spine directly
Triage bot (Workers AI) extracts entities, sentiment, key facts
Auto-approve or HITL review based on flow_c_config rules
Only approved, verified knowledge enters the system
Rejected content is logged but never stored as fact
Approved content feeds Entity 360 Memory layer

5.1 APPROVAL-BASED PERSONAL MEMORY (No-Hallucination Memory)

The single most unique capability in IntegrateWise. No other platform has this.

The Problem

Teams use multiple AIs (Claude, GPT, Grok, Gemini). Each AI:

Gives different answers to the same question
Doesn't remember what you told it last week
Hallucinates facts with no way to verify
Has no shared context with other AIs on your team

The Solution

One verified knowledge base. Every AI reads from it. No AI writes without approval.

How It Works

ANY AI (Claude, GPT, Grok, Gemini, MCP, your own models)
    │
    ├── READ: Yes — all AIs read from Entity 360 Memory
    │         (approved, verified knowledge only)
    │
    └── WRITE: Only through the approval gate
              │
              ├── AI generates insight/fact/decision
              ├── Triage Bot extracts entities + scores confidence
              ├── High confidence + trusted source → prioritized human review
              ├── Low confidence or new source → review with evidence gaps
              ├── Approved → enters Knowledge → feeds Entity 360 Memory
              └── Rejected → logged but never stored as fact

What Makes It Different

Approach	What it does	Problem
RAG (retrieval)	Fetches docs before generating	Still hallucinates — just with context
Grounding	Checks AI output against sources	Catches hallucinations after they happen
Approval-Based Personal Memory	AI output goes through triage → scoring → human approval BEFORE entering knowledge base	Hallucinations never enter the system

Configuration (per tenant)

{
  trusted_sources: ["claude", "system"], // affects review priority only
  min_confidence: 0.8,                   // review-priority threshold
  max_priority_proposals_per_day: 50,    // rate limit
  require_entity_link: true,             // must link to an entity
  notifications: {
    notify_on_auto_approve: true,
    notify_on_triage: true
  }
}

Results

Every AI on the team has the same verified context
Knowledge compounds over time — verified, not hallucinated
Switch AI providers anytime — the knowledge stays
Full audit trail on every piece of knowledge (who approved, when, from which AI)
Memory types: decision, preference, insight, action, rule, fact

6. IDENTITY RESOLUTION (S7.5)

Cross-source duplicate detection with human-in-the-loop resolution.

Detection (automatic, in pipeline)

Runs after S7 Sanity, before S8 Sectorize
Generates signature from identifying fields (email, domain, name, phone)
Queries for matches: same tenant + entity type + different source system
Supported types: account, contact, lead, ticket, opportunity, people_team
Entity ALWAYS written to Spine — detection is non-blocking

Matching confidence

Match type	Confidence	Example
Email exact	0.95	john@acme.com from Salesforce = john@acme.com from Zendesk
Domain + name	0.80	acme.com + "Acme Corp" from HubSpot = acme.com + "Acme Corporation" from Stripe
Name + phone	0.75	"John Smith" + +1-555-0123
Domain only	0.50	acme.com (low confidence, goes to HITL)

Resolution (human, via API or UI)

Merge: Combine two records into one canonical entity
Keep Separate: Mark as not duplicates, suppress future matching for this pair
Defer: Revisit later
Full audit trail on every resolution decision

7. GOVERN (Approval Gate)

Every autonomous action requires human approval. No exceptions.

What Govern gates

Entity merges (identity resolution)
Bulk field overrides
Action proposals from Twin
Connector disconnects
Schema changes

How it works

Twin or system proposes an action
Action created with status "pending"
Routed to appropriate approver (workspace admin, CSM, compliance officer)
Approver reviews evidence and approves/rejects with reason
Approved actions execute via the standard write path
Every decision logged to governance audit trail

8. TRUST LAYER

Every data point in the system carries provenance.

What's tracked

Source system: Which connector provided this data (Salesforce, Zendesk, etc.)
Sync timestamp: When this data was last synced
Confidence score: 0-1 score based on source reliability and data recency
Freshness: live (< 1h), recent (< 24h), stale (> 24h)
Evidence chain: For every Twin insight — which data points triggered it, from which layers

What's visible in UI

Source badges on every data field
Confidence percentage with color-coded bar
Evidence panel on every Twin insight (expandable)
"Why this insight" explanation with source flow visualization
"Risk if ignored" warning
Data freshness indicator on Entity 360

9. CONNECTORS (70+ integrations)

By category

Category	Count	Connectors
CRM	6	Salesforce, HubSpot, Pipedrive, Zoho CRM, Freshsales, Close
Support	9	Zendesk, Freshdesk, Front, Help Scout, Intercom, Kustomer, WhatsApp Business, Crisp, Drift
Billing	6	Stripe, Chargebee, Paddle, QuickBooks, Xero, FreshBooks
Marketing	8	HubSpot Marketing, Mailchimp, Marketo, Braze, Segment, SendGrid, Customer.io, Klaviyo
Analytics	6	Google Analytics, Mixpanel, Amplitude, Heap, Tableau, Metabase
Project	6	Jira, Asana, Linear, Monday, ClickUp, Trello
Engineering	3	GitHub, GitLab, Bitbucket
Commerce	3	Shopify, WooCommerce, BigCommerce
HR	1	BambooHR
Communication	7	Gmail, Outlook, Slack, Teams, Discord, Zoom, Google Calendar
Productivity	6	Notion, Google Drive, Dropbox, OneDrive, Airtable, Calendly
AI	4	MCP, OpenAI, Claude, Moonshot
Design	1	Figma

Connector features

OAuth 2.0 / API key / OAuth 1.0 authentication
Automatic field mapping to canonical schema
Creamy load (30 days), full sync, delta sync, webhook ingestion
Rate limit handling with exponential backoff
Connector health monitoring (status, last sync, error count)
Domain-based filtering in UI (show only relevant connectors for department)

10. PRODUCT SURFACES

Account Success (Universal Product)

Accounts Hub — Daily operating view for anyone managing accounts

Works for CSMs, CAs, hotel managers, freelancers, agency owners — anyone with accounts/clients
My Day cards: engagements today, tasks due, open risks, upcoming renewals/due dates
My KPIs: average health score, total value, account count
Health filter: all / healthy / at-risk / critical
Account list sorted by health score
AI signals section with Twin insights
Industry-adaptive: shows ARR for SaaS, invoice status for CAs, booking frequency for hospitality

Intelligence Center — Portfolio management for managers and leaders

6 KPIs: avg health, total value, at-risk value, upcoming renewals, open risks, AI signals
Health distribution bar (healthy / at-risk / critical with percentages)
Team performance table: owner name, accounts, avg health, value managed, at-risk count
Accounts needing attention (sorted by worst health)
Works for CS Directors, CA firm partners, hotel chain managers, school principals

Strategic View — Executive dashboard for VP, C-suite, board

Big 4: total value, retention %, avg health, value renewing in 90d
Value by segment/industry horizontal bar chart
Initiative portfolio: active / blocked / total
At-risk accounts: health < 70 with upcoming renewals/due dates

Business Ops (Universal Product)

Ops Cockpit — Daily operating view for founders and operators

6 mini KPIs customized to business type (MRR/sales/revenue, pipeline, clients, deals, tasks, AI signals)
AI signals: "What needs your attention" section
Active deals/clients/projects sorted by value
Account/client health sorted by worst first
Works for SaaS founders, retail business owners, agency owners, freelancers

CEO View — Company-wide dashboard

Big 4: revenue, pipeline, customers, OKR progress
Critical signals section
Strategic objectives with progress bars

COO View — Operations command center

6 ops KPIs: active workflows, initiatives, KPIs tracked, ops metrics, tasks done, ops signals
Cross-department initiatives with status badges
KPI tracker with progress bars

CIO/CTO View — Technology dashboard

4 tech KPIs: integrations, active incidents, tech metrics, tech signals
Integration landscape table: workflow name, status, execution count
Technology signals and metrics

Cross-Domain Features (all views)

Content Router: Every view lazy-loaded via module registry. 12 domain maps + BizOps cross-domain access.
Progressive Hydration: B0→B7 bucket system. Views render progressively as data arrives.
Domain Sidebar: Context-aware navigation that changes based on selected department.
L2 Cognitive Overlay: Twin insights, evidence panels, decision memory, drift detection — accessible from any view.
Command Palette: Quick navigation across all modules and domains.

11. REAL-TIME SYSTEM

Signal Stream (Durable Objects)

WebSocket/SSE for live signal delivery
Tenant-scoped broadcasting via SignalStreamDO
Signal filtering by severity, type, entity_type
Recent signals cache (100 max per tenant)

Presence Tracking

User online/away/busy status via PresenceDO
Cursor position sharing for collaboration
Room-based collaboration via RoomDO

HITL Gate (Durable Object)

HITLGate DO manages approval workflows
Real-time notification when approval is needed
Timeout handling for unresolved approvals

12. PROGRESSIVE HYDRATION (B0 → B7)

The system progressively activates capabilities as data flows in.

Bucket	State	What's available
B0	No data	Empty workspace, onboarding prompts
B1	Schema resolved	Entity types known, empty views with structure
B2	Creamy load done	30 days of data, basic truth in Spine
B3	Full load done	Complete historical data, all entity types populated
B4	Context linked	Flow B active — emails, docs, meetings in Entity 360
B5	Signals active	Twin triggers start firing, insights appear
B6	Memory active	Flow C feeding Entity 360, AI knowledge accumulating
B7	Full cognitive	All 6 Entity 360 layers operational, full Twin reasoning

13. PLATFORM INFRASTRUCTURE

Services (21 Cloudflare Workers)

Service	Port	Purpose
Gateway	8786	API routing, auth, CORS, rate limiting
Spine-v2	8795	Entity storage, Entity 360 API, Identity Resolution API
Pipeline	8794	Pipeline orchestration
Normalizer	8793	8-stage processing + S7.5 identity detection
Knowledge	8789	Context storage, memory consolidation, triage bot
Think	8798	Twin triggers, SignalEngine, AI reasoning
Act	8780	Action execution
Govern	8787	Approval workflows, policy enforcement
Intelligence	8788	Signal processing, Python AI client
Workflow	8800	Real-time signals, Durable Objects, presence
Connector	8785	Connector management, sync scheduling
Tenants	8797	Tenant management, OAuth flows
Billing	8783	Subscription management, webhook replay
Admin	8781	Admin operations
Agents	8782	Agent orchestration
L2	8790	L2 cognitive insights
Loader	8791	Data extraction from connectors
Store	8796	File storage (R2)
MCP Connector	8792	MCP tool server
Webhook Ingress	8799	Inbound webhook processing
HubSpot Sync	8784	HubSpot-specific sync logic

Storage

Spine DB: Primary database (PostgreSQL + RLS + Realtime + Auth)
Cloudflare D1: Edge cache for entity signatures, merge candidates, pipeline state
Cloudflare KV: Hot entity cache, session state, rate limits, signal cache
Cloudflare R2: File storage for documents, exports
Cloudflare Queues: Pipeline queue, knowledge queue, accelerator queue, intelligence queue, signal queue, DLQ

Security

Spine DB RLS enforces tenant isolation at database level
OAuth 2.0 with PKCE for connector authentication
Token encryption for stored credentials
Secrets Store for all sensitive configuration
Govern gate on all destructive operations
Full audit trail on every data change

14. SYNC ENTITLEMENTS (Plan-Based)

Capability	Free	Starter	Professional	Enterprise
Sync interval	24h	4h	1h	15min
Records per sync	1,000	5,000	25,000	100,000
Records per month	10,000	50,000	500,000	Unlimited
Historical days	30	90	365	Unlimited
Max connectors	1	3	10	Unlimited
Delta sync	Yes	Yes	Yes	Yes
Full refresh	No	Yes	Yes	Yes
Two-way sync	No	No	Yes	Yes
Webhooks	No	No	Yes	Yes

15. WHAT MAKES IT DIFFERENT

Capability	IntegrateWise	Gainsight	Totango	CDPs	Internal dashboards
Universal (any industry)	Yes — SaaS, finance, hospitality, SMB, freelancers	CS-only	CS-only	Marketing only	Custom per team
Unified entity across all sources	Yes (Entity 360)	Partial	Partial	Marketing only	No
Identity resolution	S7.5 + HITL	Manual	Manual	Yes but marketing-focused	No
AI with evidence	Twin + trust layer	Rules only	Rules only	No	No
Approval-Based Personal Memory	Yes — triage + approval gate, any AI reads verified knowledge, none write without approval	No	No	No	No
Governed actions	Govern gate	No	No	No	No
Schema-driven (no manual config)	150+ types auto-routed	Manual setup	Manual setup	Manual	Manual
Edge deployment	300+ Cloudflare PoPs	Cloud only	Cloud only	Cloud only	Depends
Context (emails, docs)	Unstructured path	No	No	No	No
Multi-AI support	Any AI reads from same verified knowledge	No	No	Single vendor	No
Multi-market pricing	₹999 to $999/mo	$2,500/mo+	$1,200/mo+	$1,000/mo+	Engineering cost
70+ connectors out of box	Yes	~30	~20	Varies	0