Skip to content

Getting Started

Installation

pip install lllm-core

Set your LLM provider API key:

export OPENAI_API_KEY=sk-...        # OpenAI
# or
export ANTHROPIC_API_KEY=sk-ant-... # Anthropic
# or any other LiteLLM-supported provider

5-Line Quick Start

No config files. No folder structure. No subclassing.

from lllm import Tactic

response, agent = Tactic.quick("Good morning.", model="gpt-4o", return_agent=True)
agent.open("new_chat")
agent.receive("What is the capital of France?")
print(agent.respond().content)

Tactic.quick() creates an agent from a plain string system prompt. The open / receive / respond pattern maps to: start a conversation, add a message, get a reply.

To use Anthropic instead:

response = Tactic.quick("Good morning.", model="claude-opus-4-6")

LiteLLM handles all provider differences automatically.

Single Script

For experiments and one-off scripts, everything can stay inline. No config needed — just run it.

from lllm import Tactic

response, agent = Tactic.quick(
    "Good morning.", 
    model="gpt-4o", 
    system_prompt="You are a helpful assistant.",
    return_agent=True
)
agent.open("session1")
agent.receive("Summarize quantum computing in two sentences.")
print(agent.respond().content)

Ready to Grow?

When your prompts get long, you need multiple agents, or you want to reuse components across projects, it's time to move to a proper package.

Start a standard project with:

lllm create my-app
cd my-app
uv sync --extra dev
cp .env.example .env
uv run python main.py

The default scaffold creates lllm.toml, prompts/, configs/, tactics/, main.py, and a smoke test. The --template flag accepts a bundled template name or a local folder containing lllm-template.toml.

Built-in templates:

Template Purpose
minimal Smallest runnable app
pipeline Planner/writer/reviewer workflow
service FastAPI-ready service
proxy API/tool proxy integration
research Experiment and batch workspace

The Tutorial: Build a Full Package walks through it step by step — from a single file to a complete multi-agent system with logging and advanced customization.

Or understand the model first: