TinkerNothing sent yet.
Chapter 4 of 7
Many tools, many steps
The assistant can look up an order and the host catches bad calls. Real requests need more: two orders at once, or a refund that depends on what a lookup returns.
This chapter: How does the assistant handle a request that needs several tool calls?
You will learn
- Why independent lookups can go out together, in one turn
- Why a call that needs an earlier result has to wait for it
- How the host caps the loop, so one request cannot call tools forever
Words you’ll meet (4)
- Turn
- One reply from the model: either an answer, or one or more tool calls.
- Parallel calls
- Several tool calls the model asks for in one turn, because none needs another’s result.
- Sequence
- Calls made one after another, because each needs the result of the one before.
- Call cap
- A rule in the host: the most tool calls one request may make before the loop stops.
The picture on the left is live. Each step below asks you to do one thing in it; the story continues here as you go.
Step 1 of 7
Hollis Hardware has two orders open and wants news on both. The host offers get_order. Before you run it: