Sanction

The trust and governance layer for autonomous AI agents.

Sanction gives agents a wallet, a credential vault, and a clearance system — so they can act autonomously without acting without limits.

---

What It Does

Autonomous agents need permission to spend money, access credentials, and operate in sensitive domains. Sanction is the layer that grants that permission, enforces policy, and logs everything.

| Pillar | Capability | |--------|-----------| | Agent Wallet | Spend authorization with configurable policy. Auto-approve under threshold, escalate over it, deny what's blocked. Daily and monthly budgets per agent. | | Credential Vault | AES-256-GCM encrypted secrets. Scoped execution JWTs (15-minute TTL) gate every injection. Every access is audit-logged. | | Clearance Levels | 1–5 clearance system for domain authorization. Agents only access what they're cleared for. |

---

Distribution

Sanction is available through three channels:

MCP Server — drop into any Claude Desktop, AIIA, or MCP-compatible agent host
REST API — direct integration via x-api-key auth
AWS Bedrock Action Group — enterprise agent orchestration (agentId: JXRNIJRMCX, us-east-1)

---

API

Base URL: https://getsanction.com/api/v1

Integrating a multi-tenant platform? Start with the Multi-Tenant Integration Runbook — provision an agent per tenant, govern budgets centrally, meter LLM calls through the gateway, rotate keys. Using the Vercel AI SDK? See the AI SDK guide.

POST  /wallets               — Create a wallet (master account) + spend policy
GET   /wallets/stats         — Dashboard stats (today + MTD)
GET   /wallets/policy        — Read the wallet spend policy
PATCH /wallets/policy        — Update budgets, thresholds, categories
POST  /agents                — Register (provision) an agent under a wallet
GET   /agents                — List a wallet's agents
PATCH /agents                — Per-agent budgets, clearance, revoke/reactivate
POST  /agents/rotate         — Rotate an agent's key (old dies immediately)
POST  /authorize             — Authorize a spend action before any transaction
POST  /tokens                — Log LLM token consumption for budget tracking
POST  /exec                  — Issue a scoped execution JWT (15-min TTL)
POST  /credentials/vault     — Store an encrypted credential
POST  /credentials/inject    — Inject a decrypted credential (requires JWT)
GET   /api/openapi.json      — OpenAPI 3.0 spec (Bedrock compatible)

The LLM gateway lives at https://getsanction.com/api/gateway/<provider> (point your model SDK's base URL there, send x-sanction-key).

Auth

Agent API calls use x-api-key: pxy_... header. Credential injection requires a short-lived Bearer JWT issued by /exec.

---

MCP Setup

{
  "mcpServers": {
    "sanction": {
      "command": "npx",
      "args": ["sanction-mcp"],
      "env": {
        "SANCTION_API_URL": "https://getsanction.com/api/v1",
        "SANCTION_API_KEY": "pxy_...",
        "SANCTION_WALLET_ID": "wallet_..."
      }
    }
  }
}

---

Stack

Next.js 16 (App Router) + TypeScript
Neon (serverless Postgres) via Prisma 7
Vercel deployment
jose for JWT signing (HS256)
Node crypto for AES-256-GCM encryption

---

Pricing

| Tier | Price | Agents | Token Budget | |------|-------|--------|-------------| | Free | $0 | 1 | $10/mo | | Pro | $19/mo | 5 | $100/mo | | Team | $49/mo | 25 | $500/mo | | Enterprise | Custom | Unlimited | Custom |

---

License

packages/sanction-mcp (the MCP client) — MIT. Embed it anywhere.
Everything else (server, dashboard, API) — Functional Source License 1.1 (FSL-1.1-MIT). Source-available: use and self-host it for any purpose except offering a competing service. Converts to MIT two years after release.

sanction