Tool chaining passes one tool’s output into the next tool’s input, allowing an agent to complete a larger goal through connected tasks. Like a relay race, each call hands useful information to the next step rather than starting independently. section 1.1.2 section 2.4.2
Use this mental model to distinguish dependent chains from independent tool calls:
- A sequential chain waits for an earlier result: retrieve available research sources, then use each source to obtain its URL. section 2.4.2
- An agent or prompt may define the chain explicitly, or the agent may decide which tools should follow one another. section 2.4.2
- Independent calls can run in parallel, reducing latency; parallel execution does not work when one call needs another call’s output. section 2.4.2
- Trace events expose call order, inputs, outputs, and timing, helping locate incorrect decisions and performance bottlenecks. section 2.4.2
Research planning chain
A research-planning agent first calls get_research_sources(), then passes each returned source to get_resource_url(). It uses those results to produce a research plan restricted to the available sources. Typed structures describe sources and tasks, making the intermediate data inspectable. section 2.4.2
Table 1.2 Example goals, and the tasks and tools needed to carry them out
| Goals |
Tasks |
Tools |
| Create an image |
Create an image |
|
| Travel to Calgary (YYC) |
Search for flights Book flights Book hotels Book transportation |
|
| Buy a computer |
Search for computers Compare features and prices Order a computer |
|
Compare multi-step goals with their task-level tools, including travel chains that search for flights before booking flights, hotels, and transportation. section 1.1.2
