LionAGI

lionagi is an open source Python framework for orchestrating AI agents. Its PyPI page describes it as an Intelligence Operating System, and its README calls it a governed multi agent orchestration framework. It is installed with pip install lionagi and can be used from Python code or from the terminal through the li command line tool. The framework supports a single agent, parallel fan outs, and directed graph flows in which an orchestrator plans work for specialist agents. Its core concepts are Branches, which hold individual conversations, Sessions, which coordinate Branches, flows for graph execution, teams that exchange inbox messages, and an operate function that combines tool use with structured output. Coding command line tools such as Claude Code and Codex can run in the same orchestration graph as models called through an API. API providers include OpenAI, Anthropic, Gemini, Ollama, NVIDIA NIM, Perplexity, Groq and OpenRouter. Runs are stored on disk, can be resumed and can be scheduled. Governance features include per tool call permission policies, guard hooks that block destructive commands, and git worktree sandboxing for speculative edits. A built in web interface called Lion Studio lets users inspect projects, schedules, playbooks and runs against a local daemon. The framework suits developers building multi step agent workflows who want control and traceability. It requires Python 3.10 or later and whatever provider credentials the chosen models need. lionagi is free and released under the Apache 2.0 license. The README notes that the architecture is moving toward a new runtime in the same repository, described as backward compatible, with version 1.0 planned for around the end of October 2026.