Your First Credential

Register an Agent

Before your AI editor can access secrets, it needs an identity:

sanctum agent register cursor-agent \
  --description "Cursor AI editor"

✅ Agent 'cursor-agent' registered
   Identity: Ed25519 keypair generated
   Public key: dwP8...k4Fm
   Config written to: ~/.sanctum/agents/cursor-agent/

Each agent gets a unique Ed25519 keypair. This keypair is used for challenge-response authentication — the agent proves its identity before accessing any secret.

Create an Access Policy

Grant the agent access to specific secrets:

sanctum policy add cursor-access \
  --principal "agent:cursor-agent" \
  --resources "openai/*,anthropic/*" \
  --actions retrieve \
  --max-ttl 300

✅ Policy 'cursor-access' created
   Principal: agent:cursor-agent
   Resources: openai/*, anthropic/*
   Actions: retrieve
   Max TTL: 300s (secrets expire after 5 minutes)

This policy says: cursor-agent can retrieve any secret under openai/ or anthropic/, and each retrieved value expires after 5 minutes.

Note: Deny by default. Without a policy, agents can’t access anything. The github/token secret is not accessible to cursor-agent because no policy grants it.

Check the Audit Log

Every operation is recorded in a tamper-evident HMAC-chained log:

sanctum audit log --last 5

┌─────────────────────┬───────────────┬──────────────────┬──────────┬────────┐
│ Timestamp           │ Agent         │ Resource         │ Action   │ Result │
├─────────────────────┼───────────────┼──────────────────┼──────────┼────────┤
│ 2026-02-10 23:20:01 │ cursor-agent  │ openai/api_key   │ retrieve │ ✅ OK  │
└─────────────────────┴───────────────┴──────────────────┴──────────┴────────┘

Next Step

→ Connect to AI Tool: Wire up MCP to your editor

Quickstart Connect to AI Tool