Core capabilities

Libra OS supplies the runtime services below. Your application decides which agents to call, what users may do, and how to present results and review work.

Application need	Runtime feature	Start here
Execute a prompted behavior	Persona and skill agents	Create an agent
Reuse configuration across agents	Employee records and inheritance	Employee YAML
Retrieve company knowledge	Collections, bindings, and Knowledge Packs	Knowledge
Call business systems	Custom tools, callbacks, and configured policies	Tools
Route models by task	Answer, planner, and skill model slots	Model settings
Retain context	Conversations, observations, persisted fields, house profiles	Memory
Run work beyond one connection	Native jobs, persisted progress, outcome inspection	Background tasks
Inspect installed agents	Managed-agent and operator registry APIs	Listing agents
Build a review workflow	Approval groups and action records	Desk example

Knowledge Packs and skill packs

A Knowledge Pack supplies knowledge and associated configuration. A collection is the indexed document resource an agent retrieves from. A skill/tool pack supplies callable capabilities. Installing one does not imply that every agent has access to it; configure bindings and tool availability.

Models and data flow

Model IDs are served by your configured endpoint; they are not models bundled inside the Libra OS binary. Configure local or hosted processing for every tier, including embeddings and optional memory workers. Task complexity, model choice, and available tools determine cost and latency.

See Security for data flow and Portability for capabilities that require external services or specific storage backends.

SDK and applications

Install the Python package with pip install libraos-sdk and import from libraos import Client. Source, API contracts, and language clients live in libraos/sdk.

Libra Desk is an application on this platform. A developer can use the same runtime services from a different UI or a backend workflow.