Parallel Tool Calls via bind_tools

What it does

bind_tools on a LangChain OpenRouter chat model surfaces parallel_tool_calls, letting the model request multiple tool calls in a single response turn. The provider records the preference and the model can return two or more ToolCall chunks simultaneously instead of forcing a strict sequential back-and-forth.

Why it matters

If your agent needs a calculator and a date lookup in one step, parallel_tool_calls removes the round-trip tax. Instead of Assistant → Tool A → Assistant → Tool B → Assistant, you get Assistant → Tool A + Tool B → Assistant. This is especially valuable for independent tools and mirrors how frontier models handle concurrent tool use.

Example

$ model = ChatOpenRouter(model="openrouter/anthropic/claude-3-haiku", api_key="your-openrouter-api-key")
bound = model.bind_tools([get_weather, get_time], parallel_tool_calls=True)
resp = bound.invoke("What's the weather and current time in Vienna?")
print("Tool calls:", resp.tool_calls)

Tool calls: [
    {'name': 'get_weather', 'args': {'city': 'Vienna'}, 'id': '...'},
    {'name': 'get_time',    'args': {'city': 'Vienna'}, 'id': '...'}
]

Common flags

ChatOpenRouter.bind_tools

Binds tools to the chat model, with parallel_tool_calls option

BaseTool.invoke

Synchronous tool execution for @tool decorated functions

strict=True

Enforces JSON-schema strictness for tool args