Module 3 cheat sheet: Tools, Agents & Guardrails

The LLM asks for tools, your code runs them; create_agent automates the loop and middleware keeps it safe.

Key ideas

@tool: name + docstring + type hints = schema. The LLM only asks. Lab 12 →
create_agent(model, tools, system_prompt, checkpointer, response_format). Lab 13 →
Middleware runs around model and tool calls. Lab 14 →

The manual tool loop

while True:
    ai = llm_with_tools.invoke(messages); messages.append(ai)
    if not ai.tool_calls: break
    for c in ai.tool_calls:
        r = tools_by_name[c["name"]].invoke(c["args"])
        messages.append(ToolMessage(str(r), tool_call_id=c["id"]))

Agent memory

agent = create_agent(llm, tools, checkpointer=InMemorySaver())
cfg = {"configurable": {"thread_id": "ajeet-1"}}
agent.invoke({"messages": [...]}, cfg)

Middleware

Middleware	Use
PIIMiddleware	redact / mask PII
HumanInTheLoopMiddleware	approve risky tools
ToolCallLimitMiddleware	stop loops
ModelFallbackMiddleware	backup model
SummarizationMiddleware	long chats
@before_model / @after_model	custom checks

Common mistakes

Vague tool docstrings.
No limit on tool calls.
Money tools without human approval.
Asking customers for OTP / PIN in chat.
Forgetting the checkpointer for HITL.

Created by Ajeetkumar · EasyLearnSpace