fix(core): handle tool call chunks without IDs in streaming mode

[christian-bromann]

Fix tool call streaming for empty schemas

Fixes a bug where tool calls with empty schemas (z.object({})) were being dropped in streaming mode but worked correctly in non-streaming mode.

Problem

The issue occurred because AIMessageChunk constructor was filtering out tool call chunks that didn't have an ID. OpenAI sometimes sends chunks without IDs for tools with empty parameter schemas, causing these tool calls to be completely lost in streaming mode while working fine in non-streaming mode.

Solution

Changes:

• Modified tool call chunk grouping logic in langchain-core/src/messages/ai.ts to assign fallback IDs for chunks without IDs (format: "fallback-{index}")
• Preserve original IDs when present to maintain backward compatibility
• Added comprehensive unit tests for chunks without IDs in langchain-core/src/messages/tests/base_message.test.ts
• Added integration test in libs/langchain-openai/src/tests/chat_models-extended.int.test.ts to verify streaming/non-streaming parity

Result

Tool calls now work consistently in both streaming and non-streaming modes for tools with empty schemas. Users can reliably use parameter-less tools like get_current_time in streaming contexts.

Before:

// Streaming mode - tool calls lost ❌
const stream = await llmWithTools.stream(messages);
// finalChunk.tool_calls === []

After:

// Streaming mode - tool calls preserved ✅  
const stream = await llmWithTools.stream(messages);
// finalChunk.tool_calls === [{ name: "get_current_time", args: {}, id: "fallback-0" }]

Fixes #8518

[vercel]

The latest updates on your projects. Learn more about Vercel for Git ↗︎

Name	Status	Preview	Comments	Updated (UTC)
langchainjs-docs	✅ Ready (Inspect)	Visit Preview		Jul 29, 2025 1:22am

1 Skipped Deployment

Name	Status	Preview	Comments	Updated (UTC)
langchainjs-api-refs	⬜️ Ignored (Inspect)			Jul 29, 2025 1:22am