Skip to main content
A minimal Python CLI that helps you find and book plane tickets using tool calling and reasoning to solve multi-step workflows. For example: 
Book the cheapest flight from Barcelona to Belgrade on 2026-01-31
This project showcases the power of LFM2.5-1.2B-Thinking, a small Language Model that excels at tasks that require reasoning, logic and strong tool calling skills. The best part? The model can run on edge devices.
Flight search assistant demo

​
Quick start

1

Install uv

uv --version
If the previous command fails, install uv following these instructions.
2

Build the project

uv sync
3

Run the assistant

Ask the flight search assistant to help you find and book plane tickets:
uv run flight_search.py --query "What flights are available from New York to Paris on 2026-01-19?"

​
How does it work?

The model has access to 2 tools:
  • search_flights - to retrieve contextual information
  • book_flights - to act on the outside world
The search_flights and book_flight functions are mocked using synthetic data for demonstration purposes. You can integrate with real flight data APIs (e.g., Amadeus, Skyscanner, or Kiwi) for production use.
Given a user request, for example Book the cheapest flight from Barcelona to Belgrade on 2026-01-31, the model iteratively:
  1. Generates a response possibly with tool calls
  2. Executes any tool calls
  3. Regenerates response
The model is served fully locally using llama.cpp.

​
Next steps

  • Add an evaluation dataset and loop
  • If necessary, boost model performance with GRPO fine-tuning with verifiable rewards

​
Source code

View the complete source code on GitHub.

Build docs developers (and LLMs) love

Get started for freeTalk to us