Libra OS
The AI Operating System — one sovereign AI layer over everything your company knows.
One 59MB binary. Install it inside your own infrastructure and eight supervised AI employees start working — grounded in your documents, screened by an AI firewall, and never sending your data anywhere.
Why Libra OS exists
Every company wants AI on its own work. Almost none can hand its documents to a managed cloud service.
Legal work is privileged. Insurance files carry PII. Student records are FERPA-protected. And AI that answers differently every time, with no citation and no audit trail, isn't usable in a regulated workflow — no matter how capable the model is.
The alternatives all ask something unreasonable. Cloud assistants want your data. Agent frameworks want your engineering team. Governance platforms want you to have already built the agents. Libra OS asks for one thing: a machine to install on.
What you get on day one
Eight supervised digital employees
Pre-installed and ready to work — document analysis, contract review, intake, email triage, compliance review, and more. No agent-building project, no prompt library to assemble, no platform team required.
An AI firewall on every answer
Source grounding, PII redaction, and prompt-injection screening run on every request, in every tier. Answers cite where they came from. Nothing is a black box.
Supervision built into the daily workflow
Approval queues put a human in the loop where it matters. An instant hard-stop freezes any agent mid-task. Nothing ships without review — and the review trail becomes your compliance record.
A knowledge base that stays yours
Graph and vector hybrid search over your own documents. Indexes, agent memory, and audit history accumulate inside your deployment and never leave it.
What makes Libra OS different
Most tools in this space either govern agents you still have to build, or hand you source code you still have to run. Libra OS is the runtime itself — it arrives with the employees already in it, and it works with the network cable unplugged.
In production
- Exabits — an AI data center running its own GPU fleet: internal legal and business decisions on their own hardware.
- EqualDocs — legal AI across 120+ jurisdictions; the first production
libra-os-sdkdeployment. - Bona Sonority School, Nanjing — supervised AI employees in education, from document work to voice and video.
See the case section on libraos.com for the full stories.
How it's priced
Cloud plans start at $249/month — self-serve, the full platform, no feature gating. Enterprise licenses are per node on an annual term. Every tier includes the complete platform: the firewall and the governance layer are never an upsell. Details at libraos.com/pricing.
Where to start
- LibraOS Cloud — hosted, sign-up-and-go: start free in minutes, no install.
- Cloud vs Self-Hosted — the two editions, and how to move between them.
- Getting started (self-hosted) — download, install, and run your first query.
- Deployment — managed cloud, your cloud, or air-gapped on-prem.
- Security — the AI firewall, source-cited answers, and the offline license.
- Model settings & token plans — point the routing tiers at any OpenAI-compatible endpoint, or at your plan's covered models.
- Create your first agent — a working agent in three SDK calls: employee, skill agent, first message.
- Defining employees in YAML — advanced: the declarative file format for employees and agents.
- Customer support agent — the most common deployment, from template to production.
- Core capabilities — the kernel, model routing, packs, and the SDK.
- Benchmarks — measured, not marketed: same-model A/B results.
This site is the user guide and reference for running Libra OS. It is developer-editable: the content lives in Markdown, and updates publish through a normal pull-request workflow.