LMbox Agent SDK — Documentation

LMbox Agent SDK

This site hosts the stable design documents for the LMbox Agent SDK — the kernel-agnostic toolkit partners use to write sovereign AI agents that run on LMbox appliances.

Source code: https://github.com/rlacotte/lmbox-cli Partner program: https://lmbox.eu/partenaires

Architecture Decision Records

ADRs are the source of truth for design choices. They are stable — changes here require a new ADR, never an in-place rewrite.

#	Title	Status
001	Agent Manifest format (`lmbox.eu/v1`)	Accepted

More ADRs land as the SDK matures (kernel abstraction, deployment pipeline, eval harness…). They will appear in the table above.

Quick start

pip install lmbox-cli

lmbox agent new my-first-agent
lmbox agent new contract-review --template legal-document --vendor sopra
cd contract-review
$EDITOR manifest.yaml
$EDITOR prompts/system.md
lmbox agent validate

Schema

The manifest format is published as a JSON Schema:

agent_v1.schema.json

Validate any LMbox manifest against it with lmbox agent validate or any standard JSON Schema validator.

What is LMbox?

LMbox is an on-prem AI appliance for European mid-market companies (ETI) that need generative AI without their data leaving their infrastructure. The Agent SDK is the toolkit that integrators (ESN, consultancies) use to build vertical agents on top of the LMbox runtime.

Read the partner program detail at lmbox.eu/partenaires.