Lár

Lár is a standalone Python agent execution engine designed for teams building AI systems in finance, healthcare, legal, and enterprise contexts. Unlike wrapper frameworks, it is built from the ground up to make auditability a structural property rather than an afterthought. The engine features deterministic graphs, cryptographic audit trails, step-level state diffs, and 22 compliance primitives that directly address EU AI Act requirements.

The platform ships with an Enterprise Compliance Backbone covering record-keeping, human oversight, risk management, bias filtering, and synthetic content marking. Lár is optimized for code-as-graph execution and includes topology validation, privilege minimization, and decision ledgers. The makers indicate it was developed with Cursor and Windsurf.