Lár is an open-source, provider-agnostic agentic framework for software engineers, enabling the development of deterministic, auditable, multi-agent AI workflows. Built on top of LiteLLM, Lár works equally with OpenAI, Gemini, Mistral, and local Ollama models, without locking developers into a single LLM provider.

Key Capabilities#

• Orchestrates autonomous agents using structured tool-calling and message abstractions, ensuring deterministic and verifiable task execution across supported LLM providers.
• Captures detailed execution logs, decision traces, and state transitions to support full auditability and compliance-ready AI workflows.
• Supports agentic RAG architectures, allowing agents to dynamically retrieve and synthesize information from external vector databases.
• Manages agent state, context windows, and execution flow to optimize latency, cost, and reliability across different model backends.

Deployment & Use Cases#

Designed for local deployment or containerized environments, Lár integrates seamlessly with existing CI/CD pipelines, prioritizing data privacy and operational transparency. It is well-suited for use cases such as automated financial reporting, legal document analysis, and enterprise-grade AI assistants, including scenarios requiring EU AI Act–ready auditability.