Why Events Matter: AI Control & Data Sovereignty

The business case for Synap's event architecture

---

The Problem

Traditional personal knowledge systems have a binary choice when it comes to AI and external integrations:

Model	User Control	Flexibility	Risk
All-or-Nothing	❌ Low	High	High
Manual Only	✅ Full	❌ Low	Low

The All-or-Nothing Dilemma

AI Agent: "Can I access your data?"

Option A: Yes → AI can do ANYTHING
Option B: No  → AI blocked from EVERYTHING

Result: Users either:

• Give up control for convenience
• Limit AI capabilities for safety

We need a middle ground.

---

The Solution: Requested → Validated Flow

Synap introduces a two-stage approval model:

┌─────────────────┐
│ AI/External     │
│ Proposes Action │
└────────┬────────┘
         │ .requested event
         ▼
    ┌──────────┐
    │   User   │
    │ Reviews  │
    └────┬─────┘
         │ Approve/Reject
         ▼
┌─────────────────┐
│ Change Applied  │
│ .validated      │
└─────────────────┘

Key Innovation

Every action has intent and execution separated:

• Intent: .requested - Anyone can propose
• Execution: .validated - Only after approval

---

Business Benefits

1. Granular AI Control

Problem: Traditional AI tools are all-or-nothing

Solution: Configure per-AI, per-action permissions

Example:

// Data pod configuration
{
  "ai_permissions": {
    "gpt-4": {
      "auto_approve": [
        "entities.create.requested",  // Can create notes
        "tags.create.requested"       // Can create tags
      ],
      "require_review": [
        "entities.delete.requested",  // Must ask before deleting
        "documents.update.requested"  // Must ask before editing docs
      ]
    },
    "claude-3": {
      "auto_approve": [],  // No auto-approvals
      "require_review": ["*"]  // User reviews everything
    }
  }
}

Result: Fine-grained control over what each AI can do

---

2. Audit Trail for Compliance

Problem: No visibility into what AI agents did

Solution: Every action is logged as an immutable event

Use Case: Regulatory Compliance

Auditor: "Show me all AI actions in Q4"

Query: 
SELECT * FROM events 
WHERE source = 'ai-agent'
  AND timestamp >= '2024-10-01'

Result: Complete, tamper-proof audit trail

---

3. Revocable AI Access

Problem: Once AI has access, can't revoke retroactively

Solution: Change permissions anytime, AI respects instantly

Scenario:

Day 1: GPT-4 can auto-approve entity creation
  → User sees too many AI-created notes
  
Day 2: Change to "require_review"
  → GPT-4 immediately starts asking for approval
  
Day 3: Review queue
  → User approves/rejects each proposal

Result: Dynamic, user-controlled AI behavior

---

4. Data Sovereignty

Problem: Data synced to external systems without control

Solution: Approve every external sync

Use Case: Cross-Pod Sharing

Friend's Pod: "Can I read your notes tagged #research?"

Flow:
1. Friend sends request → entities.read.requested
2. You review in your data pod
3. You configure: "Yes, read-only, expires in 7 days"
4. Share approved → entities.read.validated

Result: You control who sees your data, when, and for how long

---

Real-World Scenarios

Scenario 1: AI Task Extraction

Context: AI reads your emails and suggests tasks

Traditional Approach:

• AI creates tasks automatically
• User finds inbox full of unwanted tasks
• Disables AI entirely

Synap Approach:

1. AI reads email
   → entities.create.requested (type: 'task')
   
2. User sees notification: "AI suggests task: Call John"
   
3. User decides:
   Option A: Approve → entities.create.validated → Task created
   Option B: Reject → Discarded, no task created
   Option C: Edit → Modify title/details → Approve

Result: AI proposes, user approves, everyone happy

---

Scenario 2: External Calendar Sync

Context: Sync Google Calendar with Synap data pod

Configuration:

{
  "external_integrations": {
    "google_calendar": {
      "sync_policy": {
        "read": "auto_approve",    // Can read events freely
        "create": "require_review", // Must ask before creating
        "update": "require_review",
        "delete": "block"           // Never allow
      }
    }
  }
}

Flow:

Google Calendar creates event
  → entities.create.requested
  → User reviews notification
  → User approves
  → Event appears in Synap

Result: One-way sync (read freely), two-way sync (with approval)

---

Scenario 3: Multi-Pod Collaboration

Context: Two users want to share a project

Flow:

User A: Shares project with User B

1. User A: webhooks.deliver.requested
   → Sends to User B's pod: "User A wants to share project X"

2. User B reviews in their pod:
   - What data? (project + related notes)
   - What access? (read-only vs read-write)
   - How long? (permanent vs 30 days)

3. User B configures:
   {
     "access": "read-only",
     "expires": "2025-01-15",
     "allowed_events": ["entities.read.validated"]
   }

4. User B approves
   → webhooks.deliver.validated
   → User A can read project data from User B's pod

5. If User B changes mind:
   → Revoke access immediately
   → User A loses access

Result: Privacy-preserving collaboration with dynamic consent

---

Future: Federated Data Pods

Vision

Users control data across pods:

Personal Pod (You)
  ↓ Share with approval
Work Pod (Company)
  ↓ Share with approval
Research Pod (University)
  ↓ Share with approval
Public Knowledge Graph

Requested → Validated Enables

1. Cross-Pod Queries

   • Pod A requests data from Pod B
   • Pod B owner reviews + approves
   • Data shared with time limit

2. AI Training with Consent

   • AI company: "Can we use your data for training?"
   • You: "Yes, but anonymize and delete after 90 days"
   • Enforced via smart contracts

3. Decentralized Knowledge Graphs

   • Link entities across pods
   • Users approve each link
   • Revoke links anytime

---

Competitive Advantage

Feature	Traditional Tools	Synap
AI Control	All or nothing	Per-AI, per-action
Audit Trail	Limited or none	Complete, immutable
Data Ownership	Vendor controls	User controls
External Sync	Automatic	Approval required
Revoke Access	Hard/impossible	Instant
Cross-System	Vendor lock-in	Open federation

---

Technical Foundation

This business model is only possible because of the event architecture:

1. Intent vs Execution Separation

// Traditional (unsafe)
function createTask(data) {
  await db.insert(tasks).values(data);  // No approval step
}

// Synap (safe)
function createTask(data) {
  await publishEvent({
    type: 'entities.create.requested',  // Intent logged
    data,
  });
  // Approval step happens in worker
  // Only validated events modify DB
}

2. Immutable Audit Log

// Every action is an event
events = [
  { type: 'entities.create.requested', source: 'ai', timestamp: ... },
  { type: 'entities.create.validated', source: 'system', timestamp: ... },
]

/

/ Cannot modify past events
// Full audit trail
// Replay capability

3. Granular Permissions

// Check permissions before validation
if (event.source === 'ai-agent') {
  const permissions = await getAIPermissions(userId, aiAgent);
  
  if (!permissions.autoApprove.includes(event.type)) {
    await queueForUserReview(event);
  } else {
    await autoValidate(event);
  }
}

---

Getting Started

For Users

1. Set Up Data Pod: Self-host or use Synap Cloud
2. Configure AI Permissions: Choose which AI can auto-approve
3. Review Queue: Check AI proposals in real-time
4. Approve/Reject: Full control over every change

For Developers

1. Integrate via Events: Use .requested → .validated pattern
2. Respect Permissions: Check user config before acting
3. Subscribe to Webhooks: Get notified on validated events
4. Build Approval UIs: Help users review AI proposals

For Enterprises

1. Deploy On-Premise: Full data control
2. Configure Policies: Enterprise-wide AI permissions
3. Audit Compliance: Query event store for reports
4. Federate Securely: Connect multiple pods with approval

---

Next Steps

• Event Architecture - Technical foundation
• Event Types Catalog - All events reference
• Webhooks Guide - External integrations
• Development Guide - Build integrations