LIBREROBOTICS Request evaluation

Robots that obey their owner. Not their manufacturer.

First article
inspection
PASSOwner control
FAI / LR
Detailed engineering illustration of a quadruped: front and top views on the left, a large side view on the right. Bone linework reveals chassis panels, actuators, bearings and replaceable leg assemblies; red inspection marks identify four points.
Side view of the reference quadruped, showing its chassis, replaceable legs, and four red inspection callouts.
Side view
Front view of the reference quadruped with dimension lines.
Front view
Top view of the reference quadruped with its modular chassis.
Top view
Illustrative inspection plate · Dimensions and performance verified during evaluation

Ownership should
survive a disconnection.

A robot carries your cameras, your maps, your tasks. When its intelligence depends on a vendor, so does your ability to use it.

LibreRobotics puts the intelligence, the authority, and the evidence on your side of the perimeter.

Vendor cloud

A connection drops, a service closes, or a contract ends. The machine you bought can stop being useful.

Cut off at source

Remote authority

An update can change the robot’s priorities. A remote kill switch can take the decision out of your hands.

Owner override

Foreign eyes

Operational data leaves your site. Vendor access and jurisdictional exposure become part of your procurement review.

No outbound path

The boundary is physical. The authority is yours.

The sovereign brain.
From model to machine.

A complete robot-intelligence stack running on your hardware, on your premises, under your direction.

Discuss your platform
System assemblyLocal / owner-held

Sovereign models

Hardened open robot foundation models, including LeRobot, OpenPI, and GR00T-class models, fine-tuned on your tasks.

Published weights and code. Signed, reproducible builds.

Local fleet stack

Task orchestration, teleoperation, fleet management, and signed update bundles. Inference and operations stay inside your network.

On-robot or on-premise. Air-gap capable. Zero outbound connections.

First integrations

LeRobot SO-101 arms
Unitree quadrupeds

OEM fleet integration
on request

Attestation is part of the product.

Your procurement team needs evidence it can inspect. The system is designed for independent audit of the model, software, and outbound connections.

Verification transcriptIllustrative record
# Scope: owner-controlled deployment

model.location       local
model.weights        inspectable
build.signature      verified
build.reproducible   yes

outbound.connections 0
cloud.uplink         absent
telemetry            absent
vendor.kill_switch   absent

# Disposition
authority            owner
Inspect independentlyEvidence over assurance

Sample format, not a completed audit. Deployment-specific evidence and verification scope are established during evaluation.

The owner’s
guarantees.

Built into the system.
Open to inspection.

Aligned to the owner
You set the tasks, priorities, and operating boundaries. The manufacturer cannot impose a remote policy change.
Fully local
Inference and fleet operations run on-robot or on-premise, with air-gap operation and no vendor kill switch.
Open and verifiable
Inspect the weights and code. Reproduce the build, fine-tune on your data, and verify the software you deploy.
No foreign eyes
No telemetry or vendor backdoor. Independently auditable attestation lets your security team examine the claim.

Prior art.
On the record.

The foundations are already built and published by the team behind LibertAI. Read the methods, inspect the code, and examine what the work establishes.

Visit LibertAI Labs

Unalignment methodology

Methods / open weights

Agent-run refusal suppression and stacked abliteration, demonstrated on a 320B-parameter open multimodal model. The method is available for scrutiny.

NVFP4 edge inference

Quantization / CUDA

4-bit quantization and hand-written CUDA kernels for running large open models on workstation hardware. Practical foundations for local inference.

Conscio agent runtime

Experimental runtime

Persistent memory, attention, drives, self-monitoring, and autonomous action in an inspectable agent runtime.

Sovereign infrastructure

Confidential compute

Private, decentralized inference on confidential compute through LibertAI. Published infrastructure from the same team building LibreRobotics.

LibertAI Labs / Published workOpen for inspection

Put it through
procurement.

The system’s commitments, entered against the questions your reviewers ask. Validate them against your deployment during evaluation.

Procurement compliance record System commitments / verification in evaluation
RequirementBasis for inspectionDisposition
Owner controlOwner-set alignment and local override.Met
Air-gap operationLocal inference, orchestration, and teleoperation.Met
Cloud uplinkNo vendor cloud in the operating path.Not present
TelemetryNo operational data sent to the vendor.Not present
Foreign eyesAbsence of vendor access independently auditable.Not found

The attestation is the artifact.
Bring your security team.

Request an evaluation

Before you
bring it in.

A few questions from the buyer’s side of the bench.

Ask us directly
Can we use robots we already own?

First integrations are LeRobot SO-101 arms and Unitree quadrupeds. OEM fleet integration is available on request. Evaluation starts with your platform, interfaces, and intended tasks so the integration scope is explicit.

What does owner-set alignment mean?

You define the robot’s tasks, priorities, and operating boundaries. Our published unalignment methodology provides the foundation for removing vendor-set priorities and fine-tuning models to the owner’s requirements.

Does the system need an internet connection?

No. Inference, orchestration, and teleoperation run locally. The system supports air-gap operation and signed update bundles, without an outbound cloud connection or telemetry.

What can our security team inspect?

The published model weights and code, signed and reproducible builds, and the deployment’s network behavior. Attestation is independently auditable; the verification scope and evidence are agreed around your procurement requirements.

Are the hardware specifications certified?

The first-article panel records evaluation requirements for the reference quadruped. It is not a hardware certification or a published benchmark. Payload, endurance, environment, and safety requirements must be verified for the selected platform and use case during the pilot.

How does an evaluation work?

We begin with a discreet conversation, under NDA when needed, then scope a dev-kit pilot around your use case. Tell us about your robot platform, operating environment, tasks, and security constraints at hello@librerobotics.ai.

Own more
tomorrow.

Bring the use case.
We’ll put the system on the bench.

For defense, critical infrastructure, and industrial teams whose robots cannot accept foreign eyes.

Discretion first. A dev-kit pilot built around your platform and requirements, with details under NDA.

hello@librerobotics.ai
Request evaluation / NDA-friendly