Deploy, manage, and route robot and vision models across local GPUs, facility clusters, and cloud. Offline-resilient, data-safe, unified runtime control from prototype to fleet.
$ curl -fsSL https://get.odinedge.com | sh
Downloading Odin Edge Node Runtime v1.0.2...
Checking compatibility with NVIDIA Jetpack SDK: [OK] CUDA 12.2 detected.
Generating node private key and issuing mTLS certificates...
$ odinctl auth login --token env_tok_seattle_0a911xf
✓ Node enrollment succeeded. Name: Thor-Jetson-01 [Seattle Facility]
$ odinctl deploy apply ./examples/hello-inference/deployment.yaml
✓ Staged 1 active deployment. Routing mode: LOCAL_PRIMARY. Local policy loaded.
The unified compute and inference plane built for industrial robots, camera grids, and physical workspaces.
A lightweight, memory-safe supervisor layer installed on customer computers. Manages CUDA context, validates schemas, signs outputs, and tracks sub-millisecond latencies.
Odin Model Packages (OMP) sign and version neural weights, preprocessing specs, output schemas, camera calibration contracts, and safe motion bounds.
A dynamic client-side router that decides where inference runs — local GPU, shared facility server, or cloud fallback — based on latency, data policy, and hardware capacity.
Standardized adapter layer matching ROS 2, Fanuc, Kuka, or UR joint protocols to input observation formats with microsecond alignment.
Automated canary deployments, live shadow execution comparisons, instant rollback triggers, and strict local-only policy enforcement for industrial safety.
Odin provides the local GPU and model orchestrator; Eon monitors physical tasks, operator workflows, maintenance dispatches, and enterprise production outcomes.
Odin bridges high-performance hardware and flexible neural runtime configurations.
Owns the customer workflows, facility models, target tasks, maintenance safety, and business dashboards.
Coordinates local GPUs, manages cache, enforces security policies, and monitors execution latencies.
TensorRT within millisecond deadlines.
Lightweight K3s edge clusters.
Secured remote routing with full policy data sanitization.
Lightweight agent deployed on your edge hardware. CUDA, drivers, and mTLS configured automatically.
Your GPU node connects to your facility workspace. Telemetry streams live within seconds.
Push OMP packages via canary or shadow mode. Policy router optimizes where inference runs.
Latency, accuracy, and drift tracked per deployment. Rollback or promote with one click.
No abstract usage formulas. Pay clearly for nodes and support SLA guarantees.
Model testing, SDK prototyping, and single-cell trials.
Production inference with unified fleet canary management.
Air-gapped facility requirements with on-premises storage.