Identity is the Sum of History
Enterprise Compliance for Autonomous AI
Govern every action. Authenticate every agent. Audit every interaction. LockStock is the cryptographic continuous compliance layer between your AI agents and everything they touch — providers, data, and each other. Zero-replay per-action proof of agent identity, tamper-evident audit trails, and a neutral trust layer for agent-to-agent collaboration. SOC 2, HIPAA, PCI-DSS ready.
Our Products
From stateless drifters to verifiable agents
LockStock
"Your Logs Show When. We Show Why."
For Teams, CISOs, and Compliance
Cryptographic audit trail, governance policies, kill switch, token budgets, encrypted transcripts, and anomaly detection. One Docker container. Zero code changes.
- Cryptographic audit trail (hash-chained, tamper-evident)
- Per-action cryptographic proof (zero replay window)
- Governance policies & kill switch
- Token budgets & velocity monitoring
- Encrypted transcripts (Auditor decryption)
- Compliance-ready architecture
Running 100+ agents? Let's talk.
Liberty
"Open Source. No Strings Attached."
For Individual Developers
Hardware-bound secrets manager with zero configuration. Secrets encrypted using your hardware fingerprint—no master passwords, no key management, just works.
- Hardware-bound encryption (software PUF)
- AES-GCM authenticated encryption
- Zero configuration required
- CLI interface for everyday use
- Environment injection for secrets
- Complete audit logging
Enterprise-Grade Security Built In
Continuous compliance, zero replay windows, mathematical proof
Cryptographic Identity
"Identity is not a label; it is the sum of a complete history."
In traditional systems, an agent's ID is just a string in a database column — easily spoofed or duplicated. In LockStock, an agent's identity is its cryptographic chain. Clone it, and the chain breaks.
Chain-Based Authentication: Every agent action extends a hash chain that binds the agent's entire history into a single verifiable proof. No secrets to steal — the chain IS the identity.
Tamper-Proofing: A forged action breaks the chain. A cloned agent diverges immediately. The math catches it before any damage is done.
Governance & Kill Switch
"Freeze rogue agents before the damage. Set budgets before the spend."
The gateway enforces governance policies at the point of action. Token budgets cap spending. Allowed-action policies restrict what agents can do. The kill switch freezes agents instantly.
Token Budgets: Set per-agent spending limits. When the budget runs out, the circuit breaker trips automatically. The gateway clamps max_tokens on each request.
Kill Switch: Lock any agent from the dashboard. The gateway returns a 429 to the agent. The request never reaches the AI provider. Unlock when ready.
Encrypted Transcripts
"The server stores ciphertext it cannot read. Only you can."
Every prompt and response is encrypted with envelope encryption before leaving your network. The d3cipher cloud stores only ciphertext. Your Account Key decrypts it in your browser.
Zero-Knowledge Storage: d3cipher never sees your content. The encryption key is generated client-side and never leaves your browser. We store only a verification hash.
Auditor Decryption: Open the Auditor tab in the dashboard, enter your Account Key, and read the full content of every conversation — decrypted entirely in your browser.
Per-Action Credential Rotation
"Zero-second replay window. Continuous compliance."
Traditional systems rotate credentials every 30-90 days, creating exposure windows where stolen credentials remain valid. LockStock rotates cryptographic proof on every single action.
Business Impact: Demonstrate lower risk posture with zero replay windows. Accelerate SOC2/HIPAA audits by proving continuous rotation instead of scheduled 90-day changes. Industry benchmarks show audit cycle time reduced by 60-70% with automated compliance evidence.
The Difference: Stolen credentials from action N cannot authenticate action N+1. Compare this to traditional systems where the same API key works for months. Every action uses a credential that has never existed before and will never be valid again.
Velocity Monitoring
"Detect the rogue agent before the damage."
Agents operating at machine speed can spiral out of control in milliseconds. We embed physics-like constraints into the execution environment.
Adaptive Baseline: The system learns each agent's normal operating pattern. When behavior deviates significantly, the circuit breaker trips automatically.
Containment: Anomalous agents are frozen instantly. The request never reaches the AI provider. This prevents runaway loops, compromised agents, and unexpected cost spikes.
Causal Audit Trail
"Mathematical non-repudiation. No clocks required."
Distributed systems suffer from "clock drift"—when servers disagree on time, logs become useless. We solved this by removing wall-clock time from the equation.
Logical Ordering: Event B follows Event A because Event B contains the hash of Event A. No clocks needed. The chain itself proves the order.
Non-Repudiation: An agent cannot deny an action it took, because that action is a necessary mathematical predecessor to its current state. To deny the past is to invalidate the present.
Agent-to-Agent Authentication
"Verified handshakes. No shared secrets. No central authority."
When autonomous agents need to collaborate, how does Agent B know Agent A is who it claims to be? LockStock extends the same cryptographic chain that proves individual integrity to prove agent-to-agent identity.
Mutual Verification: Each agent can verify the other's complete cryptographic history before accepting work. The chain IS the credential — no passwords, no tokens, no PKI.
Neutral Intermediary: LockStock serves as a trusted witness to every agent interaction. Both chains record the handshake. Neither side can deny it happened.
Cross-Customer End-to-End Encryption
"Your agents talk to their agents. Nobody else can listen. Not even us."
When agents need to exchange data — within your organization or across company boundaries — LockStock provides end-to-end encryption using the MLS (Messaging Layer Security) protocol, the same IETF standard behind secure group messaging at scale.
How It Works: Each gateway ships with an MLS sidecar binary. When enabled, agents generate KeyPackages tied to their chain state. A counterparty verifies the sender's chain integrity before accepting the encrypted session. Frozen or revoked agents are rejected automatically.
Three Independent Keys: Your Gateway Key (stamping), Account Key (transcript encryption), and MLS Seed (cross-customer E2EE) are completely independent. Rotate one without touching the others. Compromise of one doesn't affect the other two.
Zero-Change Deployment
"One container. One env var. Complete governance."
The d3cipher Gateway is a Docker container you deploy in your own network. Your agents talk to it instead of directly to OpenAI, Anthropic, or any other provider.
No Code Changes: Point your agents at the gateway by changing one environment variable. No SDK. No middleware. No library to install. The agents don't even know the gateway exists.
Works With Everything: OpenAI, Anthropic, Azure, vLLM, Ollama, TGI, LiteLLM, Bedrock — any OpenAI-compatible endpoint works out of the box.