Core capabilities

Orchestration kernel

Plans, routes, and validates hundreds of agent calls per task. Seven layers of agents, each routed to the model that fits the job — the reason a complex task costs a fraction of what a naive approach spends.

Multi-provider model routing

100+ models, cloud or local, routed per query. Frontier models where they help, local models where sensitivity requires it. A provider changes terms or gets blocked? Switch. You are never locked to one vendor.

Sensitivity routing

Only the brain agent handling your most sensitive material is suggested to stay local — the other layers are free to use whatever model suits the task. You choose where on the spectrum to sit, from fully cloud to fully air-gapped.

Governance dashboard

Per-agent activity, token consumption, task volume, and firewall catch rate — the same dashboard we use to run deployments is the one you get.

Knowledge Packs

Industry knowledge as installable modules, built from licensed and public content. Add a vertical's expertise without a data project.

Capability modules

Language, voice, and vision are additive. Start with documents; switch on voice and video when you're ready. Nothing gets rebuilt.

SDK and app kit

libra-os-sdk (Apache-2.0) for teams building their own front ends on top. Channel adapters for the tools you already run — they get managed, not replaced.