What we're building

Every piece from Intermediate, in one flow: an Agent (Lesson 14) connected to a real Gemini model, given both a Custom Component tool (Lesson 12-13's WordLengthTool) and a built-in tool (add_calculator_tool), its API key coming from a Global Variable (Beginner Lesson 7), the whole thing uploaded and run over the REST API (Lesson 11), not run_flow_from_json.

Do this yourself

  1. Build the flow fresh, from an empty canvas:

    • Chat Input.
    • Google Generative AI, model gemini-3.5-flash-lite, API key picked from the GOOGLE_API_KEY Global Variable.
    • A New Custom Component, code pasted in from WordLengthTool in lesson.py (Lesson 13's tool, exactly).
    • Agent, Calculator toggle on, Model connected from Google Generative AI's model output, Tools connected from Word Length's tool output.
    • Chat Output.

  2. Wire it exactly like Lesson 13's flow, plus the Calculator toggle from Lesson 14.

    No canvas.png ships with this lesson, the same reason as Lesson 13: re-importing a saved flow that wires a model into an Agent's Model field trips a frontend re-import quirk in this Langflow version (the edge gets silently dropped with an "invalid connections" notice, even though the flow runs correctly). Build it by hand following the steps above and the wiring holds.

  3. Upload the flow and run it over the REST API (or just use the Playground, same underlying call), ask it something that needs the custom tool ("how many letters in 'checkpoint'?") and something that needs the built-in one ("what's 12 times 12?"). Confirm both come back correct, that's proof the Agent is picking the right tool for each question, not just always answering from what the model already knows.

Try this yourself

  • Add a second Custom Component tool of your own (a Fahrenheit-to-Celsius converter is a good small one) and confirm the Agent picks it correctly when asked a question that needs it.
  • Swap add_calculator_tool off and ask the same arithmetic question again, confirm Gemini still attempts an answer on its own, without a tool, and think about when you'd actually want that fallback versus forcing tool use.
  • Remove the Global Variable and hardcode the API key instead (Lesson 7's exercise, repeated here in a bigger flow), confirm it still runs, then put the Global Variable back.

If you can make these changes confidently, you're ready for the Advanced tier, starting at Lesson 17.