README
¶
Multi-Round Tool Calling
Demonstrates an agent that performs multiple sequential tool-calling rounds, where the result of one tool feeds the arguments of the next.
go run ./examples/multi_round_tools/
A small deterministic mock model.Provider drives the loop, so the example runs
with no API keys and no network access.
The scenario
The user asks: "How many orders does Ada Lovelace have?" Answering it requires two tools in sequence:
resolve_customer(name="Ada Lovelace") ──▶ { customer_id: "C-1007", tier: "gold" }
│ (A feeds B)
▼
fetch_orders(customer_id="C-1007") ──▶ { order_count: 3 }
│
▼
"Customer C-1007 has 3 recent orders."
Agent.Chat runs the tool-calling loop: it executes each requested tool, feeds
the result back to the model, and repeats until the model stops asking for
tools.
How the mock sequences the rounds
The mock chooses its next action purely from the most recent message:
| Most recent message | Mock's response |
|---|---|
| the user's question | call resolve_customer(name) |
resolve_customer result |
read customer_id, call fetch_orders(id) |
fetch_orders result |
emit the final natural-language answer |
Because the mock reads resolve_customer's output out of the tool-result
message and places the customer_id into fetch_orders's arguments, the "A
feeds B" data dependency is real, not scripted with hard-coded ids.
Key APIs
tool.Definitionwith a realHandlerandPermAllow.agent.New(...).AddTool(...).WithModel(...)builder.model.Providerimplemented locally (Chat,StreamChat,Name,Model).model.StopReasonToolCall/model.StopReasonEndto drive the loop.
Documentation
¶
Overview ¶
Example: multi_round_tools demonstrates an agent performing MULTIPLE sequential tool-calling rounds, where the result of tool A feeds the arguments of tool B.
A small deterministic mock model.Provider drives the loop, so the example runs with NO API keys and NO network access. The mock decides its next action purely from the most recent message:
round 0: user asks a question -> call resolve_customer(name) round 1: resolve_customer returned an id -> call fetch_orders(customer_id) (A feeds B) round 2: fetch_orders returned orders -> emit the final natural-language answer
Chronos' agent loop (Agent.Chat) runs the tools, feeds each result back to the model, and repeats until the model stops requesting tools.
go run ./examples/multi_round_tools/
Source Files