Installation¶

Install LionAGI¶

pipuv

pip install lionagi

uv add lionagi

Requirements: Python >= 3.10

Configure API Keys¶

LionAGI loads API keys from environment variables or .env files automatically (via pydantic-settings and python-dotenv). Set at least one provider key.

Environment variables¶

# OpenAI (default provider)
export OPENAI_API_KEY=sk-...

# Or any other provider
export ANTHROPIC_API_KEY=sk-ant-...
export GEMINI_API_KEY=...
export NVIDIA_NIM_API_KEY=nvapi-...
export GROQ_API_KEY=gsk_...
export PERPLEXITY_API_KEY=pplx-...
export OPENROUTER_API_KEY=sk-or-...

`.env` file¶

Create a .env file in your project root:

OPENAI_API_KEY=sk-...
ANTHROPIC_API_KEY=sk-ant-...

LionAGI checks .env, .env.local, and .secrets.env in that order.

Ollama (local models)¶

Ollama requires no API key. Install and run Ollama, then:

from lionagi import Branch, iModel

branch = Branch(
    chat_model=iModel(provider="ollama", model="llama3.2")
)

Verify Installation¶

import lionagi

print(lionagi.__version__)

To verify an API connection works:

import asyncio
from lionagi import Branch

async def main():
    branch = Branch()
    reply = await branch.communicate("Say hello in one word.")
    print(reply)

asyncio.run(main())

If this prints a response, your installation and API key are working.

Optional Extras¶

LionAGI ships with optional dependencies for specific features:

# Install with specific extras
pip install "lionagi[ollama]"    # Ollama client library
pip install "lionagi[mcp]"      # MCP (Model Context Protocol) server support
pip install "lionagi[reader]"   # Document reading (docling)
pip install "lionagi[rich]"     # Rich terminal output
pip install "lionagi[schema]"   # Schema code generation
pip install "lionagi[graph]"    # Graph visualization (matplotlib, networkx)
pip install "lionagi[postgres]" # PostgreSQL storage
pip install "lionagi[sqlite]"   # SQLite storage
pip install "lionagi[xml]"      # XML parsing

# Install everything
pip install "lionagi[all]"

Provider Reference¶

Provider	`provider=`	Environment Variable	Notes
OpenAI	`"openai"`	`OPENAI_API_KEY`	Default provider. GPT-4.1, GPT-4o, o-series.
Anthropic	`"anthropic"`	`ANTHROPIC_API_KEY`	Claude models via Messages API.
Google Gemini	`"gemini"`	`GEMINI_API_KEY`	Uses OpenAI-compatible endpoint.
Ollama	`"ollama"`	(none)	Local models. Requires Ollama running on `localhost:11434`.
NVIDIA NIM	`"nvidia_nim"`	`NVIDIA_NIM_API_KEY`	NVIDIA inference microservices.
Groq	`"groq"`	`GROQ_API_KEY`	Fast inference on LPU hardware.
Perplexity	`"perplexity"`	`PERPLEXITY_API_KEY`	Search-augmented responses.
OpenRouter	`"openrouter"`	`OPENROUTER_API_KEY`	Access 200+ models via single API.

Custom / OpenAI-compatible endpoints¶

Any endpoint that implements the OpenAI chat completions API:

from lionagi import iModel

model = iModel(
    provider="my_custom_provider",
    model="my-model-name",
    api_key="your-api-key",
    base_url="https://your-endpoint.com/v1",
)

Defaults¶

When you create a Branch() without specifying a model, LionAGI uses:

Provider: openai (configurable via LIONAGI_CHAT_PROVIDER env var)
Model: gpt-4.1-mini (configurable via LIONAGI_CHAT_MODEL env var)

# Override defaults via environment
export LIONAGI_CHAT_PROVIDER=anthropic
export LIONAGI_CHAT_MODEL=claude-sonnet-4-20250514

Next Steps¶

Ready to write code? Continue to Quick Start.