Skip to content
Forward Deployed
Forward Deployed
Agent Foundations: sections and modules
T7.1

The Agent Loop & Tool Calling

Implement tool use and ReAct-style planning with raw API calls and no framework.

Lessons
8 lessons
By
Sourav Banerjee
Prerequisites
None

Lessons

Start module
8 lessons
  1. 1
    The Agent Loop & Tool Calling: overview

  2. 2
    Agentic patterns from scratch: reflection, tool use, ReAct planning

  3. 3
    A real-world agent on raw API calls (Hacker News chatbot)

  4. 4
    Defining tools and binding them to a model

  5. 5
    The manual tool loop and the augmented LLM in LangGraph

  6. 6
    Tool errors and evaluating tool calls

  7. 7
    Hosted vs client-side tools: code execution and image generation

  8. 8
    The Agent Loop & Tool Calling: interview questions

Practise and Check

Deliverable
Retriever as a tool, by hand: wrap your T6 retriever as a tool, run it through your own tool loop with no framework, and log every tool call and error.
Assessment
Answer the five Interview questions without notes, then run the lab.
Interview probe
Larkspur's support agent returns a 500 whenever a customer mistypes a server name ("Is web-9 up?"). The logs show `ValueError: Unknown server 'web-9'`. The tool works for valid names. What is wrong, and what do you change?
Common mistakes
  • No max_rounds cap on a hand-written ReAct loop
  • Vague tool descriptions that cause wrong or missing calls
  • Letting tool exceptions escape instead of returning an error ToolMessage

Your Progress

How confident are you with this module?

1 means "I could not explain it yet"; 5 means "I could teach it". Rate again any time.

 

Saved in this browser when you leave the box.

Related Modules