Research updates, technical deep-dives, and announcements from the frontier of embodied AI.
IronProof proves properties of the assembly rustc emits and the PTX our inference engine sends to the GPU: constant time, memory bounds, exact outputs, race freedom. It has proved 335 of 342 obligations across IronCrypto, and it found a real out-of-bounds GPU write that no test would ever have hit.
Five days after introducing three projects, the IronSecurity stack is six: SLH-DSA and LMS signatures, a shared certificate validator, one interface for HSM and TPM key custody, OSCAL control evidence — and a high-severity Poly1305 fix we found with Wycheproof.
Three Rust projects that give agents cryptography, file privacy, and TLS 1.3 they can discover and verify instead of guess at — including the first public release of IronSocketLayer.
Meet IronPrivacyGuard: file encryption, signatures, explicit trust policies, and machine-readable contracts for AI agents, built in Rust on IronCrypto.
Listen to this articleYour browser does not support the audio element. Most cryptographic failures aren't caused by somebody breaking AES. They're caused by perfectly good cryptography being used incorrectly: a GCM nonce gets reused, a password is fed into a fast hash, an unauthenticated cipher mode gets treated like an AEAD, a MAC gets compared with ==, or an application silently substitutes the algorithm it has for the one its security policy actually requires. Humans are supposed to le
Listen to this articleYour browser does not support the audio element. Every serving loop, training runtime, and agent dispatch system sitting on top of a GPU wants the same thing at the bottom of the stack: a Rust interface over the vendor driver that's fast, correct, and unsurprising. On CUDA, that's usually cudarc — the de facto choice, and a good one. But its per-call thread-context verification and per-buffer event-fence Drop add real overhead in a hot dispatch loop: roughly 350 nanose
Listen to this articleYour browser does not support the audio element. Every agentic workflow eventually has to move a file — pull a dataset, push a report, sync a directory to a remote host. The tools available for that job were all built for humans typing commands one at a time: FTP, rsync, curl. None of them were designed to be driven by another program, and it shows — free-text output, inconsistent exit codes, no standard way to ask "what can you actually do." Today we're announcing AF
Listen to this articleYour browser does not support the audio element. Every programming language you've used was designed for a human to read and type. That made sense for sixty years. It makes less sense now that the entity writing most of the code in front of you is a language model. Agents have different constraints than we do. They pay by the token, not by the keystroke. They don't benefit from familiar syntax — they benefit from a grammar that parses on the first try. They don't ship
Listen to this articleYour browser does not support the audio element. The AI supply chain runs on trust it hasn't earned. You download a model from a hub, read a number off a leaderboard, pull weights into a pipeline — and at every step you are trusting a claim that nobody signed and nobody can check. Where did these weights come from? Are they the bytes the publisher released? Did anyone actually run the benchmark, or reproduce it? For software we solved this years ago with signatures, SB
Listen to this articleYour browser does not support the audio element. The web was built for people reading documents. Agents aren't people, and they aren't reading. They stream tokens, call tools, advertise capabilities, ship embeddings, and coordinate in swarms — and they do all of it over a transport stack designed for browsers fetching HTML. Today we're announcing SPINE (Synaptic Pathways INterconnecting Entities), an agentic-first web stack that treats the things modern LLM agents act
Listen to this articleYour browser does not support the audio element. AI agents are starting to run real workloads — building software, executing tools, driving long-horizon tasks — and every one of those workloads needs somewhere safe to run. The honest answer today is a tangle of containers, cloud VMs, and shell scripts glued together with brittle CLI scraping. None of it was designed for an agent that wants to discover what it can do, reserve a GPU, spawn a sandbox, and clean up after i
Listen to this articleYour browser does not support the audio element. PDFs are everywhere an AI agent needs to look — contracts, papers, invoices, manuals — and almost none of them were built to be read by a machine. A single document can run to hundreds of pages, the text arrives as a soup of positioned glyphs rather than sentences, and most libraries insist on loading the whole thing into memory before they will tell you anything. For an agent paying by the token and the millisecond, tha
Every shell that wants to work with AI agents claims it was built for them. We decided to stop claiming it and start measuring it. So this release does two things: it introduces a brand-new open-source tool, agentic-eval, that scores any program for how well it works with AI agents — and it ships AetherShell 1.5, which uses that tool to harden and benchmark itself. On a zero-to-ten score across four areas, AetherShell scores 9.6. Nushell scores 2.3, PowerShell 2.2, and the traditional Bash, Zsh
Three new open-source, agentic-first Rust tools from NERVOSYS: Lit (version control), Louie (TUI framework), and Dewey (GUI framework). Structured data by default, a machine-readable ontology for agent discovery, and hardened agent-facing entry points.
Today we are releasing AgenticBlockTransfer (abt), an open-source cross-platform disk imaging tool built in Rust that reads, writes, verifies, hashes, and formats block devices via CLI, TUI, and GUI interfaces — with native support for AI agent orchestration through JSON-LD ontology, OpenAPI, and the Model Context Protocol (MCP). abt is both an agentic-first CLI successor to dd and a human-first GUI/TUI successor to balenaEtcher, Ventoy, Rufus, Fedora Media Writer, and rpi-imager. Simple. Relia
Today, we are excited to announce three new open-source tools from NERVOSYS, each built in Rust and designed from the ground up for the age of AI agents. Together they form a powerful triad: AetherShell, Simon, and Chasm. AetherShell — The AI-Native Command Line Traditional shells were designed for humans typing text commands. AetherShell reimagines the command line for a world where AI agents are first-class participants. AI agents today are forced to interact with shells designed decades a
Below is a transcript of my interview with Ahalya Ravendran of the Australian Centre for Field Robotics at the University of Sydney, conducted as part of the Coffee with a Researcher series for the IEEE International Conference on Robotics and Automation (ICRA 2022): How can robots learn to interact with and reason about themselves and the world without an intuitive feel for either? Communication is at the heart of biological and robotic systems. Inspired by control theory, information theory,
This is Nervosys Blog, the official blog of Nervosys that is just getting started. Things will be up and running shortly, but you can subscribe in the meantime if you'd like to stay up-to-date and receive emails when new content is published!