Deterministic security infrastructure for AI agents.
We build formal enforcement systems grounded in 40 years of Information Flow Control research.
| Layer | Tool | Install | What it secures |
|---|---|---|---|
| Runtime | MVAR | pip install mvar-security |
How information flows |
| Execution | ClawZero | pip install clawzero |
What the agent does |
| Memory | ClawSeal | pip install clawseal |
What the agent remembers |
| Layer | Tool | Current Version | Role |
|---|---|---|---|
| Runtime | MVAR | 1.4.3 |
Runtime enforcement |
| Execution | ClawZero | 0.4.0 |
Execution boundary |
| Memory | ClawSeal | 1.1.5 |
Cryptographic memory |
Install the complete stack:
pip install mvar-security clawzero clawsealExecution boundary for AI agents. Blocks prompt injection from reaching tool execution.
- ~1ms per decision (measured benchmark on release hardware)
- Deterministic enforcement (same input = same decision)
- 5 framework adapters (OpenClaw, LangChain, CrewAI, AutoGen, generic)
- 50-vector attack pack across 9 categories
- Cryptographic witness chain for every decision
- Apache 2.0
Information Flow Control engine for LLM agent runtimes. Dual-lattice IFC with cryptographic provenance.
Cryptographic memory for AI agents. Tamper-evident QSEAL signatures, zero database dependencies, human-readable YAML storage.
- Provisional patent filed February 24, 2026 (24 claims)
- Submitted to NIST RFI Docket NIST-2025-0035
- Preprint published on SSRN (February 2026)
- Built on: Denning lattice, Jif, FlowCaml, FIDES
Built by Shawn Cohen. Apache 2.0. Patent pending.