Quickstart

This guide walks through the shortest path to a working tool-using agent. It focuses on the core workflow: load configuration, create a client, attach tools, and run a task.

Before you run

1. Install the framework (see Installation page).
2. Create a .env file with your model endpoint settings.
3. Ensure your endpoint is OpenAI-compatible.

Minimum required variables:

LLM_BASE_URL=...
LLM_API_KEY=...
LLM_MODEL=...

Your First Agent

The script below loads config from your environment, opens the async model client, and gives the agent two filesystem tools. The agent can decide when to call these tools as part of its reasoning loop.

import asyncio
from llm_framework.core import LLMClient, Agent
from llm_framework.tools.filesystem import list_directory, read_file

async def main():
    async with LLMClient.from_env() as client:

        agent = Agent(client=client, tools=[list_directory, read_file])

        result = await agent.run("List the files in my home directory.")
        print(result["answer"])

if __name__ == "__main__":
    asyncio.run(main())

What this does

1. LLMClient.from_env() reads LLM_BASE_URL, LLM_API_KEY, LLM_MODEL and opens the HTTP client.
2. LLMClient handles chat completion calls and connection lifecycle.
3. Agent runs a ReAct loop and can invoke the provided tools.

Typical next steps

• Add approval gating for sensitive tools.
• Add guardrails for input/output safety.
• Move shared tools to MCP servers when multiple agents need them.