LAGRANGE · Physical Agent Infra

Enter the field.
Complete the task.

Building Agentic OS for AI agents working with real machines.

Agentic OS connects intelligent decisions to dependable execution. Physical RSI is our longer-term direction for turning field feedback into evaluated capability improvements.

Field practice Physical RSI

Earth, Moon, and the five Lagrange points: L1, L2, L3, L4, and L5 — L1Field exploration
L2Task completion
L3Capability capture
L4Reliable operation
L5Scaled reuse
Connecting models to machines and physical capabilities to real operations.
Among the five Lagrange points, L1–L3 require ongoing correction; L4–L5 are relatively stable.

Scroll to field practice

Field practice

From isolated actions to complete tasks.

Robots work in existing operations alongside on-site equipment.

We start with bounded production tasks in existing workflows and work toward reliable completion in the field.

Entering real operations

Read the field article (Chinese)

Physical RSI

Improving physical intelligence through real work.

Physical RSI is our direction for improving models and systems from real tasks, then returning only evaluated capabilities to the field.

Read our technical view (Chinese)

PHYSICAL RSI

Continuous improvement driven by real tasks

Execute → Learn → Evaluate → Redeploy

01
Field execution
Task · Execution
Tasks, devices, and operating workflows
02
Execution feedback
Trace · Failure
State, failures, and human intervention
03
Capability improvement
Research · Engineering
Changes to models and software systems
04
Evaluation
Simulation · HIL · Real
Simulation, hardware in the loop, and field evaluation
05
Release and reuse
Release · Deploy
Versioned release, rollback, and reuse