Building the world's first nervous system and platform for robotic general intelligence.
Earth
Joined March 2022
- Tweets146
- Following43
- Followers68
- Likes14
IronSocketLayer 0.5.1 is flying:
◾ Certificate policies in config AND in the session report
◾ CNSA 1.0 + 2.0 profiles reconciled clause-by-clause with RFC 9151/8603
◾ Received data wiped after read
◾ Trust anchors as name + key
And, IronEvidence: every project's control mappings as OSCAL that assessor tooling reads.
Validated against Wycheproof, NIST ACVP/CAVP, tlsfuzzer AND TLS-Anvil. Four independent suites. All AGPL-3.0.
We secure the world of agents 🦾
Read or listen here:
nervosys.ai/blog/ironsecurit…
🌐 IronSocketLayer brings the model to TLS 1.3 + QUIC for HTTP/3.
Post-quantum by default, sans-I/O, `no_std`, and built on IronCrypto. Agents get named profiles, SessionReports, structured failures, mTLS, QUIC v1/v2, ECH, revocation checks, PSKs, resumption & 0-RTT.
The goal: security infrastructure agents can inspect, reason about, use, and deploy at scale without guessing.
Code here:
github.com/nervosys/IronCryp…
github.com/nervosys/IronPriv…
github.com/nervosys/IronSock…
Read or listen here:
blog.nervosys.ai/blog/iron-s…
Introducing IronCrypto — agentic-first cryptography in pure Rust. 🛡️
It provides zero third-party dependencies, hardware acceleration, post-quantum crypto, `rustls` integration, MCP support, and a machine-readable security ontology.
github.com/nervosys/IronCryp…
🤖 Made with AI
IronCrypto includes:
- AES-GCM, ChaCha20-Poly1305, SHA-2/3
- P-256/P-384/P-521, X25519, Ed25519, RSA
- ML-KEM-768 + ML-DSA-65
- Deterministic CycloneDX SBOMs
- Timing-leak analysis
- `rustls` CryptoProvider
- MCP tools
IronCrypto is trustworthy cryptography for autonomous AI 🛡️
Read or listen to our IronCrypto blog announcement here:
nervosys.ai/blog/announcing-…
per aspera ad astra! 🚀
AI inference engines, training runtimes, and agent dispatch loops all want the same base: a Rust API over hardware that's fast, correct, and unsurprising.
Today we're announcing IronAccelerator — an agentic hardware-agnostic substrate faster than cudarc.
nervosys.ai/blog/announcing-…
On an RTX 3090 Ti against cudarc:
- host→device wins at every size tested (up to ~1.28×, CI-confirmed)
- device→host is at parity
- opt-in MemPool allocator hits ~70× (!) on warm alloc/free by caching the driver function table and skipping per-call thread-context rebinding.
CUDA is the only backend that's production-ready today — ROCm, Metal, Vulkan, QNN, and others compile and enumerate devices but haven't had the same optimization pass. We want to get the best tools in your hands today.
AGPL-3.0 + commercial license.
github.com/nervosys/IronAcce…
