Trust infrastructure
for AI agents

Hash-chained, Ed25519-signed audit trails. Prove what your agent did — to auditors, customers, and yourself.

How it works

Three steps to verifiable agent behavior.

1

Attest

Record every action with cryptographic proof. Each receipt is Ed25519-signed and timestamped.

2

Chain

Build tamper-proof audit trails per agent. Each receipt hashes its predecessor — no gaps, no rewrites.

3

Verify

Anyone can verify any receipt, anytime. Open format, no vendor lock-in, independently auditable.

For developers

Drop in via MCP or use the SDK directly. Two lines to your first attestation.

Claude Code — MCP config
// .claude/settings.json
{
  "mcpServers": {
    "sigil": {
      "command": "uvx",
      "args": ["sigil-notary"],
      "env": {
        "SIGIL_API_KEY": "sg_..."
      }
    }
  }
}
Python SDK
from sigil import SigilClient

client = SigilClient(api_key="sg_...")

receipt = client.attest(
    "file_write",
    {"path": "/src/main.py"}
)

# receipt.receipt_hash  → "a3f8..."
# receipt.seq           → 17
# receipt.signature     → Ed25519

Pricing

Open source today. Hosted services coming soon.

Hosted Verification

Coming Soon
For production deployments
  • Hosted verification endpoint
  • Dashboard & analytics
  • Webhook integrations
  • Priority support
Star on GitHub

Enterprise

Custom
For regulated industries
  • HSM key management
  • Data residency controls
  • SLA & SSO
  • Compliance reporting
Contact Us

Built in the open

The Sigil MCP server and SDK are MIT-licensed. Audit the code that audits your agents.