Cryptographic trust
for AI agents.

Integrate AgentGuard into your stack. Read agentguard.run/llms.txt and follow the quickstart.

npm install agentguard-cb GitHub
In production. Patent-backed.
Live at AgWaterAI 4 US provisional patents filed Feb 2026 Public on npm
agentguard · verify
$ agentguard verify --trace latest
actionpayment:initiate
agentagw-prod-watchdog-v20
amount$142.50/ cap $500.00
timestamp2026-05-21T19:42:31.004Z
────────────────────────────────────────────
signatureed25519:a3f2b1c9d4e5f6a7b8c9d0e1f2a3b4c5d6e7f8a9b0c1d2e3f4a5b6c7d8e9f0a1
chainsha256:9e4d8fa2b3c1d5e6f7a8b9c0d1e2f3a4b5c6d7e8f9a0b1c2d3e4f5a6b7c8d9e0
  Signature valid
  Spend cap not exceeded
  DAG chain integrity verified
  Receipt tamper-evident

Overview

Tamper-evident by design.

Every agent action is Ed25519-signed at execution time. No database admin, infrastructure operator, or post-hoc modification can forge or alter a signed receipt. Centaur stores Postgres rows. AgentGuard signs them.

Harness agnostic.

Works with Centaur, LangChain, Claude Code, Amp, Stripe Agents, or any custom stack. AgentGuard operates at the SDK layer, below orchestration. One npm install. No platform lock-in.

Spend caps with signed receipts.

Per-agent financial limits enforced at the SDK level, not the application layer. Each cap check produces a signed, verifiable receipt. Auditable by regulators, investors, and compliance teams.

DAG trust chain.

Multi-agent workflows produce a cryptographically linked attestation graph. Every hop is independently verifiable. No central authority required. Holds up to external audit without trusting the database.


The ecosystem.

Core SDK
AgentGuard CB

The circuit-breaker and control-bus layer for agent pipelines. Drop-in npm package. Intercepts actions, applies policy, emits signed receipts. Works with any harness or model provider.

npm install agentguard-cb →
Audit
AgentGuard Trace

Signed, tamper-evident audit trail for any agent stack. Ed25519 receipts chain across turns and agents. Cannot be altered post-execution, even by the infrastructure operator. In production.

Integration guide →
Finance
AgentGuard Spend

Per-agent spend caps enforced at the SDK layer with signed receipts at each enforcement point. Verifiable limits, not application-layer trust. Shipped v0.1.0 with Ed25519 chain and tamper-detection verifier.

View on GitHub →

Where AgentGuard sits.

The cryptographic primitive below orchestration, above the model.
Orchestration handles control flow. AgentGuard proves what happened.

Orchestration
Centaur · LangChain · Claude Code · Amp · Custom
AgentGuard SDK
Ed25519 signing  ·  Spend caps
DAG attestation  ·  Trust chain
Model
Claude · GPT-4 · Gemini · Custom
Signed receipts export to   compliance  ·  audit  ·  regulators  ·  M&A diligence  ·  DAG verification