Multi Tool

Real tasks require multiple capabilities: search the web, then query a database, then format results, then send an email. Multi-tool orchestration is what separates a demo agent (one tool) from a production agent (coordinated tool suite).

Multi-tool orchestration is like a chef using knife, stove, and oven in sequence — not just having tools available, but knowing which to use when and how outputs flow between them.

Scenario

Task: 'Find our top customer this month and draft a thank-you email.' Step 1: query_database('top customer Q4') → {name: 'Acme Corp'}. Step 2: search_web('Acme Corp recent news') → news summary. Step 3: draft_email(name, news) → email draft. Three tools, sequential chain.

Solution

In Agent Foundations, apply Multi Tool to this scenario: Task: 'Find our top customer this month and draft a thank-you email. Identify the inputs, run the technique, validate the output, and note one thing you would monitor in production.

Do this before moving to the next module — reading alone is not enough.

Open the Code Walkthrough below and run it locally. Change one parameter related to Multi Tool (e.g. model, temperature, top_k, or tool name), observe the difference in output, and write 2–3 sentences explaining what changed.

Highlighted lines show where Multi Tool happens in the code.

1import asyncio  # import dependencies2
3async def execute_tools_parallel(tool_calls: list) -> list:  # key line for Multi Tool4    async def run_one(tc):5        fn = TOOL_REGISTRY[tc.function.name]  # key line for Multi Tool6        args = json.loads(tc.function.arguments)7        return await asyncio.to_thread(fn, **args)  # return the result8    return await asyncio.gather(*[run_one(tc) for tc in tool_calls])  # return the result9
10async def execute_tools_sequential(steps: list[dict]) -> list:  # key line for Multi Tool11    results = []12    for step in steps:13        context = {"previous_results": results}14        result = await asyncio.to_thread(15            TOOL_REGISTRY[step["tool"]], **step["args"], **context  # key line for Multi Tool16        )17        results.append(result)18    return results  # return the result

• Too many tools — LLM selects wrong ones
• Sequential execution of independent tools — wasted latency
• No fallback when primary tool fails
• Not passing context between chained tools
• No analytics on which tools fail most

Quick recap — the most important points from this module.