ohdearquant/lionagi

An intelligence orchestra

What it solves

lionagi provides a governed multi-agent orchestration framework that allows developers to build complex agent workflows without surrendering control to an opaque runtime. It solves the problem of "black box" agent chains by providing typed, inspectable state and durable runs that can be resumed or monitored in real-time.

How it works

The framework uses a set of core concepts to manage AI interactions:

  • Branches: Individual conversation threads containing history, tools, and model configurations.
  • Sessions: Coordinators that manage multiple branches to execute DAG (Directed Acyclic Graph) workflows.
  • Flows: Orchestration patterns where a lead agent plans a DAG of specialist workers who execute as dependencies are resolved.
  • CLI Integration: The li CLI allows users to run single agents, parallel fan-outs, or complex flows directly from the terminal, integrating both API-based models and subscription-based CLI agents (like Claude Code).
  • Persistence: Every run is automatically saved locally, allowing users to resume, reattach, or schedule recurring tasks.

Who it’s for

It is designed for developers and AI engineers who want a transparent, governed approach to multi-agent orchestration, specifically those who need structured output, durable execution, and the ability to mix API models with local CLI-based agents.

Highlights

  • Governed Execution: Includes permission policies for tool calls and guard hooks to block destructive commands.
  • Durable Runs: Local persistence of all runs allows for easy resumption and monitoring via the li monitor command.
  • ** uma-first CLI**: First-class support for running agents and complex DAG flows directly from the terminal.
  • Lion Studio: A client-side web UI for inspecting runs, managing playbooks, and visualizing execution DAGs.
  • Structured Output: Native support for Pydantic models to ensure AI responses follow a specific schema.

Related

  • Project
    wanxingai/LightAgent

    A modular, lightweight Python framework for building secure, multi-agent AI applications with support for complex workflows, persistent memory, and human-in-the-loop approvals.

  • Project
    AgentEra/Agently

    Agently is a Python framework that turns LLM calls into structured, observable, and restart‑safe runtime units. It provides contract‑first output schemas, streaming of partial results, tool/action integration, versioned skills, managed execution resources, and an event‑driven workflow engine (TriggerFlow) for building reliable AI services.

  • Project
    huangjunsen0406/py-xiaozhi

    py-xiaozhi is a cross-platform, asynchronous Python framework that enables real-time voice and vision AI interactions on desktop and embedded hardware.

  • Project
    mudler/LocalAGI

    LocalAGI is a self‑hosted Go platform for building and running AI agents locally. It provides a web UI for no‑code agent creation, OpenAI‑compatible REST APIs, built‑in RAG/knowledge base, multimodal model support, and extensible Go‑based custom actions. Agents can be linked into teams, scheduled, and connected to Discord, Slack, Telegram, GitHub, and IRC. Docker‑Compose files let you run on CPU or various GPUs, keeping all data on‑premise for privacy.

  • Project
    vortico/flama

    Flama is a production framework that packages predictive and generative AI models into portable files and serves them as high-performance APIs with built-in support for agentic tools and standard LLM protocols.