lionpride 1.0.0a0

The kernel layer for production AI agents - protocol-based, type-safe, zero framework lock-in

lionpride

Production-ready multi-agent workflow orchestration framework.

Alpha Release - API may change. Originated from lionagi v0, elevated and hardened for production use.

Features

• Model Agnostic - Built-in providers for OpenAI-compatible APIs, Anthropic, Gemini, Claude Code
• LNDL - Domain-specific language for LLM structured output and enhanced reasoning (JSON fallback supported)
• 75% Token Reduction - Efficient instruction representation for complex agentic workflows (e.g., 20+ tools)
• Async Native - Operation graph building, dependency-aware execution, auto-extensions
• Modular Architecture - Protocol-based composition, zero framework lock-in

Installation

pip install lionpride

Quick Start

from lionpride import Session
from lionpride.services import iModel

# Create session with model
session = Session()
model = iModel(provider="openai", model="gpt-4o-mini")
session.register_service(model)

# Create branch and operate
branch = session.create_branch()
result = await branch.operate(
    instruction="Analyze this data",
    imodel=model,
)

Roadmap

• Formal mathematical framework
• Rust core for performance-critical paths

License

Apache-2.0