Product

Not a tool.
A runtime platform.

Stop rebuilding robot runtime scaffolding for every deployment.

AEROS is not a library you import. It is the infrastructure layer that makes embodied agents runnable, governable, and evolvable — out of the box.

v0.15.0 Runtime Core · ECM Package System · Governance Layer · Frozen Identity · Adaptive Persona · Layer 2 Memory · Active Dreaming · Evolution · Fleet · Marketplace · Unitree G1 locomotion — all shipped, all open source.
View on GitHub → Architecture deep dive → Changelog →
Audience

Who this is for

Robot OEM

Ship products faster with a standard governed runtime instead of rebuilding scaffolding per customer.

Fleet Operator

Run heterogeneous robot fleets under unified lifecycle control, policy enforcement, and audit.

Production Robotics Team

Add governance, rollback, and auditability to production robot-agent workflows.

Open Source Developer

Prototype on an open-source runtime with real execution control, not demo scripts.

AGPL-3.0

Agent Runtime Core

The problem: How do you keep an embodied agent running continuously, stably, and recoverably in the real world?

  • Lifecycle Management Init, start, pause, resume, stop, restart — full agent state machine
  • State Persistence Agent state survives crashes, restarts, and migrations
  • Task Scheduling Priority-aware scheduling with preemption and deadline guarantees
  • Execution Loop Sense-decide-act closed loop with timeout protection
  • Fault Recovery Four-level structured recovery: retry, fallback, degrade, halt
  • Distributed Tracing OpenTelemetry-native tracing across the full execution chain
governed runtime 
$ aeros run "restock shelf B3" --robot mock
# the agent proposes; every action passes admission,
# policy, contract, and safety gates before executing

$ aeros migrate export -o unit-07.aer
exported: unit-07.aer
  identity_hash=9f3ac2…71e
  episodic_rows=1247

$ aeros migrate verify --bundle unit-07.aer
verified: unit-07.aer
AGPL-3.0 kernel · Apache-2.0 interfaces

ECM Package System

The problem: How do you standardize, version, and safely upgrade the capabilities of embodied agents?

  • ECM Standard Format Declarative capability modules with metadata, dependencies, and contracts
  • Install / Upgrade / Rollback Package operations with atomic transactions and instant rollback
  • Semantic Versioning Breaking vs. non-breaking changes tracked at the contract level
  • Six-Dimensional Contracts Perception, cognition, action, safety, resource, interface constraints
  • Compatibility Checking Pre-install validation against hardware, runtime, and policy constraints
  • Shadow Mode New capabilities run in parallel without affecting production behavior
ecm operations 
$ aeros ecm package ./dual_grasp
packaged: dual_grasp-2.1.0.aecm
  capability: manipulation.dual_grasp
  version:    2.1.0

$ aeros ecm install dual_grasp-2.1.0.aecm
# Ed25519 signature gate checks dual_grasp-2.1.0.aecm.sig
installed: manipulation.dual_grasp
  contract: bridge/configs/contracts/manipulation.dual_grasp.json

$ aeros ecm list
AGPL-3.0

Governance Layer

The problem: How do you enforce constraints, audit behavior, and maintain control over agents operating in the real world?

  • Policy DSL Declarative policy language for defining what agents can and cannot do
  • Watcher Engine Real-time monitoring with configurable triggers and escalation rules
  • Structured Recovery Four-level orchestrated recovery: log, alert, intervene, halt
  • Full Audit Trail Ed25519-signed log of every decision, action, and state change — queryable
  • Adaptive Persona Asymmetric trust matrix: tighten = auto, relax = operator-signed. Deep dive
  • Layer 2 Semantic Memory Deterministic SQL aggregation, outbox-pattern emit durability. Deep dive
operator review 
$ aeros agent persona-events --kind relax --limit 3
# read-only audit dump; needs-review rows surface separately

$ aeros agent review --recent 20
# interactive operator review: relaxations only land
# with your Ed25519 countersignature

$ aeros agent freeze-persona
# kill-switch: adaptive surface off, identity untouched
Impact

What changes after AEROS

Before

Rebuild runtime scaffolding for every project

→
After

One governed runtime, reused across deployments

Before

Capability upgrade = risk

→
After

Shadow mode testing + instant rollback

Before

Incident review by hand

→
After

Full-chain audit trail, every decision recorded

Roadmap

Where we're headed

Phase 1

Runtime Core

Agent lifecycle, state, scheduling, execution, recovery, tracing

Phase 2

ECM Package System

Standard format, contracts, versioning, install/upgrade, shadow mode

Phase 3

Governance Engine

Policy DSL, watchers, structured recovery, audit, conflict detection

Phase 4

Cross-Embodiment

ECM portability across robot morphologies; MuJoCo sim backends (Franka, Go2, G1) shipped, ROS 2 / real-robot bring-up hardening

Phase 5

Ecosystem

Marketplace infrastructure shipped in-repo; next: community ECM library, reproducibility benchmark suite + leaderboard, governed CaP-Agent0 integration (in progress), partner integrations

See AEROS in action

Clone the repository, or schedule a technical walkthrough of the runtime, ECM system, and governance layer.

View on GitHub Book a Demo