Building the infrastructure
embodied agents deserve
Open-core runtime infrastructure for governed embodied deployment.
AEROS builds the runtime layer that turns embodied AI from lab demos into governed, production-grade systems. We make deployment safe, auditable, and continuous.
We build governed, evolvable runtime infrastructure for embodied agents.
Not another robot OS. Not another model wrapper. A runtime platform that closes the gap between "AI demo" and "AI in production" — making it manageable, safe, and auditable for every deployment.
A runtime, not a demo
AEROS started from a single observation: the gap between "AI demo" and "AI in production" for embodied systems isn't a model problem or a hardware problem. It's a runtime problem.
So we built one. AEROS is the governed runtime layer that sits between the LLM planner and the actuator — handling identity, persona, capability evolution, audit, and rollback for agents that need to run continuously, safely, under governance, in the real world.
Today, AEROS is a fully open-source runtime — dual-licensed AGPL-3.0 + Apache-2.0 — with a growing capability packaging system and a governance layer that is built in, not bolted on. The whole thing is public at github.com/s20sc/aeros-core.
Why embodied agents need runtime infrastructure
Models are necessary but not sufficient
Foundation models give agents the ability to reason. But reasoning without governance, state management, and lifecycle control is a demo, not a product.
Hardware diversity is accelerating
Humanoids, quadrupeds, AMRs, drones, collaborative arms — each with different capabilities. Capabilities must be modular, portable, and versionable.
Production demands governance
In warehouses, hospitals, and public spaces, "what happened?" and "who authorized that?" are not optional questions. Runtime governance makes them answerable.
The timing is right
Models are ready
Foundation models can now reason about physical actions, objects, and environments. The intelligence layer is no longer the bottleneck.
Hardware is diversifying
Humanoids, quadrupeds, AMRs, collaborative arms — the fleet is getting heterogeneous. A shared runtime layer is no longer optional.
Governance is becoming unavoidable
As robots move from labs to warehouses, hospitals, and public spaces, policy enforcement, audit, and accountability become regulatory requirements, not nice-to-haves.
Thesis-driven, engineering-first
Most robotics infrastructure is built bottom-up from middleware. AEROS is designed top-down from a governance-first thesis and implemented as production-grade open-source software.
Single-agent thesis
A clear architectural principle that every design decision traces back to — one identity, one runtime, one audit chain.
Open-source runtime
Not slides. Running code, fully open source, with a passing test suite and a measured benchmark.
Governance is the wedge
The runtime layer between the LLM and the actuator is where audit, persona, and rollback have to live — and where the commercial value sits.
Milestones
AEROS MVP + research series
Runtime core running with a passing test suite and measured benchmarks; the first AEROS papers land on arXiv.
Hardware + bench v2
Real-hardware adapter seams for Unitree G1 and Franka Panda; multi-LLM benchmark matrix.
Full runtime open-sourced
v0.15.0 public at github.com/s20sc/aeros-core — the entire runtime, dual-licensed AGPL-3.0 + Apache-2.0.
Runtime Core v1.0
Production-ready runtime with full lifecycle management, ECM support, and tracing.
First pilot deployment
Proof-of-concept deployment with a warehouse automation partner.
Ecosystem
Community ECM library, reproducibility benchmark suite + leaderboard, governed CaP-Agent0 integration.
For investors and partners
AEROS is defining the governed runtime layer for embodied AI — a systems category that doesn't exist yet. We're looking for partners who understand infrastructure plays and are willing to invest in the platform layer.