Grounded context
Combine graph, vector, and keyword retrieval. Grounded output carries citations; ungrounded output is labeled.
Build on LibraOS
Give your AI app a governed runtime.
Use one platform for durable agents, knowledge, memory, permissions, human approvals, and audit. Build your own product—or start from Libra Desk, the reference application built on LibraOS.
The primitives your application should not rebuild.
Durable state
Scope conversational and long-lived memory to the application, end user, and agent that own it.
Permissions & approvals
Keep external actions behind explicit policy and human review instead of hiding control in prompts.
Durable runs
Start long-running jobs, follow their progress, recover from failures, and preserve an execution receipt.
Attributable actions
Trace work to the application, user, employee, and job without inventing a second identity system.
Compatible APIs
Use the LibraOS SDK and native APIs, with OpenAI-compatible model access where it fits your stack.
From install to first application.
STEP 1Run LibraOSInstall the binary or request Cloud beta access.
STEP 2Create identityRegister an application credential with the least privilege it needs.
STEP 3Create an agentDefine the employee identity and runnable behavior.
STEP 4Send workCall the Messages API and receive governed output.
pip install libraos-sdk
# Then follow the maintained three-call example:
# employee.create → agent.create → messages.create
open https://libraos.com/docs/creating-an-agentPrefer a guided start?
The setup wizard generates your starting team.
Tell the wizard about your team, recurring work, knowledge sources, and approval boundaries. It proposes editable employees and tools for you to review before activation.