Dativo Talon — Vendor Integration Guide

Governing Third-Party AI Vendors

---

Overview

This guide shows how to add Talon compliance to existing AI automation, whether custom-built or third-party SaaS vendors. Talon doesn't replace your existing tools — it adds governance, audit trails, and compliance controls.

Three integration patterns:

1. MCP Proxy (recommended) — Talon sits between vendor and your data
2. Webhook Interception — Talon logs/redacts webhook payloads
3. Shadow Mode — Read-only audit of vendor behavior

---

Why This Matters

The Compliance Problem with Third-Party Vendors

You're using great AI tools like:

• Zendesk AI Agent
• Intercom Resolution Bot
• Drift AI Chatbot
• HubSpot AI Assistant
• Custom Slack bots
• OpenAI Assistants API

But when audited (GDPR, NIS2, EU AI Act), you can't answer:

• ✗ What customer data did the vendor access on January 15th?
• ✗ Was PII redacted before sending to LLMs?
• ✗ Where is data stored? (EU data residency requirement)
• ✗ Which high-risk decisions lacked human oversight?
• ✗ Can you export GDPR Article 30 processing records?

Legal reality:

• You are the data controller (GDPR Article 4)
• Vendor is the data processor (GDPR Article 28)
• You're liable even if vendor claims compliance
• "Vendor said they're compliant" is NOT a defense in AEPD audits

---

Pattern 1: MCP Proxy (Recommended)

Architecture

BEFORE (Black box):
Third-Party AI Agent → Directly accesses your Zendesk/CRM → No visibility

AFTER (Full audit trail):
                    ┌──────────────────────────────────┐
                    │  Talon (your infrastructure)     │
                    │  - Logs all access               │
                    │  - Redacts PII                   │
                    │  - Enforces policies             │
                    │  - Generates audit trail         │
                    └──────────────────────────────────┘
                                 ↓
Third-Party AI Agent → Talon MCP Server → Your Zendesk/CRM
                                 ↓
                    You have complete evidence trail

Implementation (30 minutes)

Step 1: Install Talon

# On your infrastructure (VM, EC2, on-prem server)
wget https://github.com/dativo-io/talon/releases/download/v1.0.0/talon-linux-amd64
chmod +x talon-linux-amd64
sudo mv talon-linux-amd64 /usr/local/bin/talon

Step 2: Configure MCP Proxy for Zendesk

# Create MCP server config
mkdir -p /opt/talon/mcp-servers

cat > /opt/talon/mcp-servers/zendesk-proxy.json <<EOF
{
  "command": "talon",
  "args": ["mcp", "serve", "zendesk", "--mode", "proxy"],
  "env": {
    "ZENDESK_URL": "https://your-company.zendesk.com",
    "ZENDESK_API_KEY": "\${ZENDESK_KEY}"
  }
}
EOF

Step 3: Create Governance Policy

# /opt/talon/agents/zendesk-vendor-proxy.talon.yaml
agent:
  name: "zendesk-vendor-proxy"
  description: "Governance layer for Zendesk AI Agent vendor"
  version: "1.0.0"
  type: "mcp_proxy"

proxy:
  mode: "intercept"  # Intercept all calls
  upstream:
    vendor: "zendesk-ai-agent"
    url: "https://zendesk-ai.example.com"
    auth: "bearer"  # Pass through vendor auth token

capabilities:
  allowed_tools:
    - zendesk_ticket_search
    - zendesk_ticket_read
    - zendesk_ticket_update
  
  forbidden_tools:
    - zendesk_user_delete  # Block destructive operations
    - zendesk_export_all   # Block mass data export

pii_handling:
  redaction_rules:
    - field: "requester.email"
      method: "hash"
    - field: "requester.phone"
      method: "mask_middle"  # +34 6XX XXX 789
    - field: "custom_fields.credit_card"
      method: "redact_full"  # Complete removal

compliance:
  frameworks: ["gdpr", "nis2", "iso-27001"]
  data_residency: "eu-only"   # requires upstream in an EU region (eu-*)
  audit_retention: 365

  # High-risk operations (refunds, account deletion, exports) require human
  # approval; see proxy compliance policy for the built-in detection rules.
  human_oversight: "always"

evidence:
  capture_requests: true
  capture_responses: true
  capture_pii_redactions: true
  exclude_fields:
    - "auth_token"
    - "internal_notes"

Step 4: Start Talon MCP Proxy

talon server \
  --port 8080 \
  --mcp-proxy \
  --config /opt/talon/agents/zendesk-vendor-proxy.talon.yaml

# Output:
# → MCP proxy listening on https://talon.your-company.local:8080
# → Proxying requests to https://zendesk-ai.example.com
# → Audit trail: /opt/talon/evidence.db

Step 5: Point Vendor to Talon

In Zendesk AI Agent settings:
┌──────────────────────────────────────────────┐
│ Data Sources Configuration                   │
│                                              │
│ ☐ Direct Zendesk API access                 │
│ ☑ Custom MCP Server                         │
│   MCP Endpoint: https://talon.your-company.local:8080 │
│   Auth Token: <your-talon-token>            │
└──────────────────────────────────────────────┘

What Happens Now

Vendor wants to search tickets
    ↓
Calls Talon MCP endpoint: /tools/call {"name": "zendesk_ticket_search", ...}
    ↓
Talon intercepts:
    ├─ Logs: "Zendesk AI Agent requested tickets matching 'eSIM activation'"
    ├─ Policy check: Is "zendesk_ticket_search" in allowed_tools? ✓ YES
    ├─ PII redaction: Masks phone numbers per redaction_rules
    ├─ Fetches from real Zendesk API
    ├─ Redacts response before returning
    ├─ Generates evidence record (who, what, when, redacted_fields)
    └─ Returns to vendor
    ↓
Vendor receives data (works normally, unaware of governance layer)
    ↓
Your compliance officer has complete audit trail

Benefits:

• ✅ Vendor functionality unchanged (transparent proxy)
• ✅ Full visibility into vendor data access
• ✅ PII never reaches vendor's systems unredacted
• ✅ Can block forbidden operations (user deletes, mass exports)
• ✅ Generate GDPR Article 30 records on demand
• ✅ Prove NIS2 compliance to auditors

---

Pattern 2: Webhook Interception

When to Use

• Vendor triggers via webhooks (Zendesk → Vendor on ticket creation)
• You can control webhook destinations
• Simpler than MCP if vendor doesn't support custom servers

Implementation (20 minutes)

Step 1: Configure Webhook Proxy

# /opt/talon/agents/webhook-proxy.talon.yaml
agent:
  name: "zendesk-webhook-proxy"
  type: "webhook_interceptor"

triggers:
  - name: "intercept-vendor-webhooks"
    type: "webhook"
    endpoint: "/webhooks/zendesk"
    method: "POST"
    
    # Talon forwards to vendor after logging/redacting
    forward_to: "https://zendesk-ai-vendor.com/api/v1/tickets"
    forward_auth:
      type: "bearer"
      token: "${VENDOR_API_KEY}"

pii_handling:
  redaction_rules:
    - field: "ticket.requester.email"
      method: "hash"
    - field: "ticket.description"
      patterns:
        - "(\\+?\\d{1,3}[-.\\s]?)?\\d{9,15}"  # Phone numbers
      method: "mask"

compliance:
  audit_retention: 365
  frameworks: ["gdpr", "nis2"]

evidence:
  capture_webhook_payload: true
  capture_forwarded_payload: true  # See what vendor received

Step 2: Update Zendesk Webhook

BEFORE:
Zendesk Webhook URL: https://zendesk-ai-vendor.com/api/v1/tickets

AFTER:
Zendesk Webhook URL: https://talon.your-company.local/webhooks/zendesk
                      ↓
                  (Talon logs, redacts, forwards)
                      ↓
                  https://zendesk-ai-vendor.com/api/v1/tickets

Step 3: Verify Interception

talon logs --follow --agent webhook-proxy

# Output shows every webhook:
# [2025-02-16 10:23:15] Webhook received: ticket.created #45231
# [2025-02-16 10:23:15] PII redacted: 2 phone numbers, 1 email
# [2025-02-16 10:23:15] Forwarded to vendor: 1.2KB payload
# [2025-02-16 10:23:15] Evidence record generated: evt_abc123

Benefits:

• ✅ Logs every webhook payload vendor receives
• ✅ Redacts PII before vendor sees it
• ✅ Can block suspicious payloads
• ✅ Vendor unaware of proxy (transparent)
• ✅ Minimal setup (just change webhook URL)

---

Pattern 3: Shadow Mode (Audit Only)

When to Use

• Vendor doesn't support custom MCP or webhook routing
• You need compliance visibility but can't force vendor changes
• First step before full interception (validate Talon policies)

Implementation (15 minutes)

Step 1: Shadow Audit Configuration

# /opt/talon/agents/zendesk-shadow-audit.talon.yaml
agent:
  name: "zendesk-shadow-audit"
  mode: "shadow"  # Read-only, no interception

triggers:
  - name: "audit-zendesk-access"
    type: "cron"
    schedule: "*/5 * * * *"  # Every 5 minutes
    action: "compare_access_logs"

audit:
  # Talon independently fetches Zendesk audit logs
  sources:
    - name: "zendesk_api_logs"
      endpoint: "https://your-company.zendesk.com/api/v2/audit_logs"
      auth: "bearer:${ZENDESK_ADMIN_KEY}"
    
    - name: "vendor_claimed_access"
      endpoint: "https://zendesk-ai-vendor.com/api/audit"
      auth: "bearer:${VENDOR_API_KEY}"
  
  # Compare what vendor claims vs. what Zendesk logs show
  compare:
    - field: "accessed_ticket_ids"
      alert_if: "mismatch"
    
    - field: "accessed_user_ids"
      alert_if: "vendor_accessed_but_not_in_their_audit"
    
    - field: "data_center_location"
      alert_if: "non_eu_location"

  alerts:
    slack_webhook: "https://hooks.slack.com/services/YOUR/WEBHOOK"
    email: "security@your-company.com"

compliance:
  generate_reports:
    - type: "gdpr_article_30"
      frequency: "weekly"
      recipients: ["compliance@your-company.com"]

Step 2: Run Shadow Audit

talon server --config /opt/talon/agents/zendesk-shadow-audit.talon.yaml

# Talon runs in background, auditing vendor behavior

Step 3: Review Findings

talon audit report --agent zendesk-shadow-audit --last 7d

# Output:
# Vendor Access Audit (2025-02-09 to 2025-02-16)
# ------------------------------------------------
# ✓ Total API calls: 1,247
# ✓ PII accessed: 423 customer records
# ⚠ Discrepancies found: 3
#   - Vendor accessed ticket #45299 (not in their audit log)
#   - API call from us-east-1 (outside EU data residency)
#   - User export at 3:47 AM (suspicious timing)
#
# Evidence: /opt/talon/evidence/shadow-audit-2025-02-16.json

Benefits:

• ✅ No changes to vendor setup (zero friction)
• ✅ Independent verification of vendor claims
• ✅ Alerts on policy violations
• ✅ Can't prevent violations, but detects them
• ✅ Good first step before forcing interception

Limitations:

• ❌ Cannot block vendor in real-time
• ❌ Relies on vendor providing audit logs
• ❌ PII already sent to vendor (can only detect, not prevent)

---

Pattern Comparison Table

Feature	MCP Proxy	Webhook Interception	Shadow Mode
Setup Time	30 min	20 min	15 min
Vendor Changes Required	Medium (MCP endpoint config)	Low (webhook URL change)	None
Blocks Violations	✅ Yes	✅ Yes	❌ No (detect only)
PII Redaction	✅ Before vendor sees it	✅ Before vendor sees it	❌ After vendor has it
Human Oversight	✅ Real-time approval	✅ Real-time approval	❌ Post-hoc review
Audit Trail	✅ Complete	✅ Complete	✅ Partial
Vendor Transparency	Vendor aware	Transparent	Transparent
Best For	New vendors, controlled environments	Webhook-based vendors	Legacy vendors, validation

---

Common Integration Scenarios

Scenario 1: Zendesk AI Agent

vendor: "Zendesk AI Agent"
pattern: "MCP Proxy"
reason: "Zendesk supports custom MCP servers in enterprise plan"
setup_time: "30 minutes"
compliance_gain: "Full GDPR Article 30 records + PII redaction"

Scenario 2: Intercom Resolution Bot

vendor: "Intercom Resolution Bot"
pattern: "Webhook Interception"
reason: "Intercom triggers via webhooks, can change URL in settings"
setup_time: "20 minutes"
compliance_gain: "Audit trail + PII redaction on webhooks"

Scenario 3: Custom Slack Bot (Self-Built)

vendor: "Internal Slack bot (Python script)"
pattern: "Direct Integration"
reason: "You control the code, modify it to call Talon"
setup_time: "10 minutes (5 lines of code)"
compliance_gain: "Full governance + policy enforcement"
code_change: |
  # Before
  response = openai.ChatCompletion.create(...)
  
  # After
  response = requests.post("http://localhost:8081/v1/chat/completions", ...)

Scenario 4: HubSpot AI Assistant (No API Access)

vendor: "HubSpot AI Assistant"
pattern: "Shadow Mode"
reason: "HubSpot doesn't allow custom MCP or webhook routing"
setup_time: "15 minutes"
compliance_gain: "Audit detection, vendor behavior monitoring"
limitation: "Cannot block in real-time, only detect violations"
next_step: "Escalate to HubSpot for MCP support or find alternative"

---

Vendor Negotiation Guide

Conversation Template

When vendor resists Talon integration:

You: "We need to route your API access through our compliance proxy."

Vendor: "Why? Don't you trust us? We're already GDPR compliant."

You: "Spanish regulation requires independent audit trails. We're the 
      data controller, so we're liable even with your DPA in place."

Vendor: "But our system needs direct access to work properly."

You: "MCP is an industry standard. If you support it, integration is 
      seamless. If not, we'll need to evaluate alternatives."

Vendor: "That requires engineering work on our side..."

You: "We understand. However, GDPR Article 28 requires us to verify 
      processor compliance. Without independent audit trails, we can't 
      use your product. Can your team provide an ETA for MCP support?"

Outcome:

• Vendor adds MCP support → You get full transparency
• Vendor refuses → You find different vendor (market pressure works)
• Vendor negotiates → Maybe they provide better audit logs as alternative

Market Pressure Strategy

If 10+ European companies demand MCP proxying:

1. Vendors will add support (competitive pressure)
2. MCP becomes standard in EU market
3. Talon becomes de facto compliance gateway

This is already happening:

• Anthropic supports MCP natively (Claude Desktop, API)
• OpenAI announced "Custom Tools Protocol" (MCP-compatible)
• Microsoft exploring MCP for Copilot

Talon's advantage: First mover in compliance-grade MCP gateway.

---

Verification & Testing

Test Your Integration

1. Verify Interception

# Send test request through Talon
curl -X POST https://talon.your-company.local/tools/call \
  -H "Authorization: Bearer ${TALON_TOKEN}" \
  -d '{
    "name": "zendesk_ticket_search",
    "arguments": {"query": "test"}
  }'

# Check logs
talon logs --last 1m

# Expected output:
# → Policy check: ALLOWED (zendesk_ticket_search in allowed_tools)
# → PII redacted: 0 fields
# → Upstream call: 200 OK (154ms)
# → Evidence generated: evt_abc123

2. Verify PII Redaction

# Create test ticket with PII
curl -X POST https://talon.your-company.local/tools/call \
  -d '{
    "name": "zendesk_ticket_create",
    "arguments": {
      "subject": "Test",
      "requester": {
        "email": "test@example.com",
        "phone": "+34612345678"
      }
    }
  }'

# Check evidence
talon audit show evt_abc123

# Expected evidence:
# {
#   "pii_redacted": [
#     {"field": "requester.email", "method": "hash", "original_hash": "sha256:..."},
#     {"field": "requester.phone", "method": "mask_middle", "redacted_value": "+34 6XX XXX 678"}
#   ]
# }

3. Verify Policy Enforcement

# Try forbidden operation
curl -X POST https://talon.your-company.local/tools/call \
  -d '{"name": "zendesk_user_delete", "arguments": {"user_id": 123}}'

# Expected: 403 Forbidden
# {
#   "error": "Policy violation: zendesk_user_delete not in allowed_tools",
#   "evidence_id": "evt_xyz789"
# }

---

Migration Path

Phase 1: Shadow Mode (Week 1)

• Deploy Talon in read-only mode
• Build audit trail for 1 week
• Validate policies without blocking vendor
• Risk: Zero (no vendor changes)
• Goal: Prove Talon works, tune policies

Phase 2: Pilot Interception (Week 2)

• Enable MCP proxy or webhook interception
• Route 10% of traffic through Talon
• Monitor for issues (latency, errors)
• Risk: Low (easy rollback)
• Goal: Verify production readiness

Phase 3: Full Rollout (Week 3)

• Route 100% traffic through Talon
• Enable PII redaction
• Enable policy enforcement
• Risk: Medium (vendor dependency)
• Goal: Full compliance coverage

Phase 4: Human Oversight (Week 4)

• Add approval workflows for high-risk actions
• Train team on plan review dashboard
• Configure alert thresholds
• Risk: Low (improves control)
• Goal: EU AI Act Article 14 compliance

---

Troubleshooting

Issue: Vendor Rejects Talon's MCP Endpoint

Symptom: Vendor returns "Invalid MCP server" error

Solutions:

1. Check Talon MCP endpoint is publicly accessible
2. Verify SSL certificate (vendors may require valid HTTPS)
3. Check vendor's MCP implementation version (may need upgrade)
4. Contact vendor support with MCP spec URL: https://spec.modelcontextprotocol.io

Issue: High Latency After Adding Talon

Symptom: API calls 2-3x slower through Talon

Solutions:

1. Check Talon's policy evaluation time: talon metrics
2. Optimize Rego policies (use indexed data structures)
3. Enable caching: cache_ttl: 60s in agent config
4. Deploy Talon closer to vendor (reduce network hops)

Issue: PII Still Reaching Vendor

Symptom: Audit shows unredacted PII in vendor logs

Solutions:

1. Verify redaction rules match actual field names
2. Check vendor uses nested fields: requester.custom_fields.phone
3. Enable debug logging: talon server --log-level debug
4. Add catch-all regex patterns for PII detection

Issue: Vendor Audit Logs Unavailable

Symptom: Shadow mode can't fetch vendor's audit logs

Solutions:

1. Check if vendor provides audit API (may be enterprise-only)
2. Request CSV export if API unavailable
3. Use Zendesk/CRM audit logs as source of truth
4. Escalate to vendor for better audit access

---

Compliance Benefits Summary

Before Talon

• ❌ No independent audit trail
• ❌ Vendor's "we're compliant" claims unverified
• ❌ Can't prove GDPR Article 30 compliance
• ❌ PII sent to vendors unredacted
• ❌ No human oversight for high-risk decisions
• ❌ Manual evidence gathering during audits (weeks)

After Talon

• ✅ Independent audit trail (vendor-agnostic)
• ✅ Real-time verification of vendor compliance
• ✅ One-command GDPR Article 30 exports
• ✅ PII automatically redacted before vendor access
• ✅ Systematic human oversight (visual plan review)
• ✅ Audit-ready in minutes (not weeks)

---

Cost-Benefit Analysis

Scenario: Spanish Telecom (150 employees)

Without Talon:

• Vendor cost: €2,000/month (Zendesk AI Agent)
• Compliance audit prep: 40 hours/quarter × €100/hr = €4,000/quarter
• Risk of GDPR fine: €50,000 (if violation discovered)
• Total annual risk: €16,000 + €50,000 exposure

With Talon:

• Vendor cost: €2,000/month (unchanged)
• Talon cost: €0 (open source)
• Compliance audit prep: 2 hours/quarter × €100/hr = €200/quarter
• Risk of GDPR fine: ~€0 (full compliance)
• Total annual cost: €800 + zero exposure

ROI: €15,200/year savings + eliminated fine risk

---

Next Steps

1. Choose your pattern:

   • New vendor or greenfield: MCP Proxy
   • Webhook-based vendor: Webhook Interception
   • Legacy vendor or validation: Shadow Mode

2. Start with pilot:

   • Deploy Talon in shadow mode for 1 week
   • Validate policies without impacting vendor
   • Review audit trails with compliance officer

3. Gradual rollout:

   • Enable interception for 10% traffic
   • Monitor for issues (latency, errors)
   • Roll out to 100% after validation

4. Prove compliance:

   • Generate first GDPR Article 30 report
   • Show to compliance officer/auditors
   • Document time saved vs. manual process

5. Expand usage:

   • Add more vendors through Talon
   • Enable advanced features (memory, triggers)
   • Train team on plan review dashboard

---

Support

• Documentation: https://github.com/dativo-io/talon#readme
• GitHub Issues: https://github.com/dativo-io/talon/issues
• Community Slack: https://dativo-community.slack.com
• Enterprise Support: enterprise@dativo.com

Remember: The goal isn't to replace your vendors — it's to govern them and generate your own evidence. Talon adds the governance and audit layer vendors can't provide; the compliance determination stays with you as the data controller.