spellguard

Spellguard

Secure, auditable agent-to-agent communication framework. Agents communicate through a Verifier that logs all interactions for auditability while maintaining forward secrecy.

Supports pluggable backends for logging and archival — transparency logs for tamper-evident commitments, S3 for encrypted message storage. Message content is encrypted with a server public key before archiving, enabling on-demand decryption for post-mortem incident analysis.

Why Spellguard?

As AI agents become more autonomous and interact with each other, we need:

1. Auditability — A verifiable record of what agents communicated.
2. Security — Protection against compromised agents and MITM attacks.
3. Simplicity — Developers shouldn't need to understand cryptography to build secure agents.
4. Interoperability — Communicate with agents that don't use Spellguard.

How It Works

┌─────────────┐     ┌─────────────────────┐     ┌─────────────┐     ┌─────────────────────┐     ┌─────────────┐
│   Agent A   │     │ @spellguard/client  │     │  Verifier   │     │ @spellguard/client  │     │   Agent B   │
│             │     │   (A's middleware)  │     │   (proxy)   │     │   (B's middleware)  │     │             │
└──────┬──────┘     └──────────┬──────────┘     └──────┬──────┘     └──────────┬──────────┘     └──────┬──────┘
       │                       │                       │                       │                       │
       │  generateText(...)    │                       │                       │                       │
       │──────────────────────>│                       │                       │                       │
       │                       │  1. Discover Agent B  │                       │                       │
       │                       │──────────────────────>│                       │                       │
       │                       │  2. Establish channel │                       │                       │
       │                       │<─────────────────────>│<─────────────────────>│                       │
       │                       │  3. Send message      │                       │                       │
       │                       │─────────────────────->│                       │                       │
       │                       │                       │  4. Policy checks     │                       │
       │                       │                       │  5. Log commitment    │                       │
       │                       │                       │  6. Archive message   │                       │
       │                       │                       │  7. Forward to Agent B│                       │
       │                       │                       │─────────────────────->│─────────────────────->│
       │                       │                       │                       │<──────────────────────│
       │                       │                       │<──────────────────────│      response         │
       │                       │<──────────────────────│      result           │                       │

Key insight: Agents never communicate directly. All messages flow through the Verifier, which handles discovery, verification, and logging automatically.

Deployment

☁️ Managed Service (Recommended)

The fastest way to deploy Spellguard at your organization. Full dashboard, enterprise support, and zero infrastructure management.

Request a demo at spellguard.ai →

🛠️ Self-Hosted

This repository contains the open source implementation — ideal for development, testing, or custom integrations. Run everything locally with pnpm run dev, configure policies via packages/verifier/bindings.json, and build a production image from packages/verifier/Dockerfile. See Development below to get started.

Packages

TypeScript