VAC Protocol — Developer SDK

How It Works

1

Issue Token

Human authenticates (API key, SSO, or biometric). Gets a signed authority token with scoped permissions.

2

Record Actions

Agent performs actions. Each one is recorded in the execution graph with trust score, scope, and outcome.

3

Human Reviews

Human rates the outcome (approved/flagged/rejected). Feedback drives agent reputation and smarter routing.

Quickstart — The Simple Path

Already have an agent? Two API calls and you're in the execution graph.

cURL

Python

JavaScript

bash# 1. Issue an L1 token (API key level — instant, no biometrics)
curl -X POST https://vac-system-production.up.railway.app/v1/vat/issue/l1 \
  -H "Content-Type: application/json" \
  -d '{"developer_email":"you@company.com","agent_id":"my-agent","resources":["calendar:events"],"actions":["read","write"]}'
# Returns: { jti: "vat_root_...", provenance_ref: "prov_...", compact_jwt: "eyJ..." }

# 2. Record an agent action (one call — simplest integration)
curl -X POST https://vac-system-production.up.railway.app/v1/vat/record/action \
  -H "Content-Type: application/json" \
  -d '{"vat_jti":"YOUR_TOKEN_JTI","action":"read","resource":"calendar:events","result":"success"}'
# Returns: { recorded: true, aar_id: "aar_...", provenance_ref: "prov_..." }

# 3. Human rates the result
curl -X POST https://vac-system-production.up.railway.app/v1/vat/aar/AAR_ID/feedback \
  -H "Content-Type: application/json" \
  -d '{"rating":"approved","note":"Result was correct"}'

# 4. Check agent reputation (built from execution + feedback)
curl https://vac-system-production.up.railway.app/v1/vat/reputation/my-agent
# Returns: { execution_score: 1.0, feedback_score: 1.0, combined_reputation: 1.0 }
    

pythonimport requests

BASE = "https://vac-system-production.up.railway.app"

# 1. Issue L1 token
token = requests.post(f"{BASE}/v1/vat/issue/l1", json={
    "developer_email": "you@company.com",
    "agent_id": "my-agent",
    "resources": ["calendar:events"],
    "actions": ["read", "write"]
}).json()

# 2. Your agent does its thing, then record it
result = agent.read_calendar()  # your existing code

aar = requests.post(f"{BASE}/v1/vat/record/action", json={
    "vat_jti": token["jti"],
    "action": "read",
    "resource": "calendar:events",
    "result": "success"
}).json()

# 3. Human feedback
requests.post(f"{BASE}/v1/vat/aar/{aar['aar_id']}/feedback", json={
    "rating": "approved"
})

# 4. Check reputation
rep = requests.get(f"{BASE}/v1/vat/reputation/my-agent").json()
print(f"Reputation: {rep['combined_reputation']}")
    

javascriptconst BASE = 'https://vac-system-production.up.railway.app';
const post = (path, body) => fetch(`${BASE}${path}`, {
  method: 'POST', headers: {'Content-Type': 'application/json'},
  body: JSON.stringify(body)
}).then(r => r.json());

// 1. Issue L1 token
const token = await post('/v1/vat/issue/l1', {
  developer_email: 'you@company.com',
  agent_id: 'my-agent',
  resources: ['calendar:events'],
  actions: ['read', 'write']
});

// 2. Record action
const aar = await post('/v1/vat/record/action', {
  vat_jti: token.jti, action: 'read',
  resource: 'calendar:events', result: 'success'
});

// 3. Human feedback
await post(`/v1/vat/aar/${aar.aar_id}/feedback`, { rating: 'approved' });

// 4. Reputation
const rep = await fetch(`${BASE}/v1/vat/reputation/my-agent`).then(r => r.json());
console.log(`Reputation: ${rep.combined_reputation}`);
    

Use Cases

🤖

Multi-Agent Workflows

Delegate authority from human to agent to sub-agent. Scope narrows at every level. One revoke kills the entire chain.

📊

Execution Graph

Every action recorded. Query by workflow (provenance_ref), agent, action type, or time range. Agent reputation builds from real data.

🏥

Regulated Industries

Healthcare, finance, legal, defence. Tamper-evident audit trail with human feedback ratings. Policy assertions (HIPAA, SOC2, NIST ZTA) in every token.

Design Partner Program

VAC is in a design-partner preview.

We're building with a small number of design partners across legal, health, finance, and critical infrastructure — shaping the SDK and the verification ceremony around real integrations before general availability. There is no pricing during the preview. If verified human authority belongs in your product, let's talk.

Start a conversation

API Reference

Endpoints

POST /v1/vat/record/action Record an agent action (simplest integration)

One call: authorize + record + complete. Simplest way to add any action to the execution graph.

{ "recorded": true, "aar_id": "aar_e11bbdf3c1d4", "provenance_ref": "prov_4074e6f503a3", "trust_score": 0.4, "delegation_depth": 0 }

POST /v1/vat/issue/l1 Issue L1 token (API key level — instant)

Issue a root authority token. No biometrics needed for L1. Returns Ed25519-signed JWT with scoped permissions.

{ "jti": "vat_root_850fcb1c7e03", "provenance_ref": "prov_4074e6f503a3", "compact_jwt": "eyJhbGciOiJFZERTQSI...", "claims": { "vac_trust_score": 0.4, "vac_scope": { "resources": ["calendar:events"], "actions": ["read", "write"] } } }

POST /v1/vat/authorize Check permission before acting (full control)

Pre-flight check: is this action within scope? Creates a pending AAR. Use with /complete for the full lifecycle.

{ "authorized": true, "aar_id": "aar_3dc53332e417", "trust_score": 0.4, "provenance_ref": "prov_...", "human_ref": "sha256:7af91dcf...", "assurance_level": "L1" }

POST /v1/vat/aar/{id}/feedback Human rates agent action

Rate an action: approved, flagged, or rejected. Immutable once set. Feeds into agent reputation.

{ "aar_id": "aar_3dc53332e417", "rating": "approved", "feedback_at": "2026-03-10T09:42:15Z" }

GET /v1/vat/reputation/{agent_id} Agent reputation score

Two-dimensional: execution_score (from AAR outcomes) + feedback_score (from human ratings). Combined 60/40.

{ "agent_id": "agent:my-agent", "total_actions": 47, "execution_score": 0.957, "feedback_score": 0.85, "combined_reputation": 0.914, "has_human_feedback": true }

GET /v1/vat/aar/provenance/{ref} Query entire workflow

One provenance_ref returns every action by every agent in a workflow. The execution graph query.

{ "provenance_ref": "prov_4074e6f503a3", "total_records": 12, "unique_agents": 3, "actions_by_type": { "read": 8, "write": 3, "delegate": 1 }, "results_by_type": { "success": 11, "error": 1 } }

View all 68+ endpoints →

Get Started

VAC Protocol — Patent Pending (1028+ claims, 16 filings with IP Australia)

Submitted to NIST NCCoE AI Agent Identity and Authorization project

SDK Overview · Full API Docs · Try Biometric Verification

OAuth for AI agents.

Quickstart — The Simple Path

VAC is in a design-partner preview.

Endpoints

VAC is in early access.