feat(openai): add optional aiohttp backend support for improved concurrency

[NirAyalon1]

Introduction

The OpenAI Python SDK now supports an optional aiohttp backend that removes concurrency limits seen with HTTPX. This PR exposes an opt-in path in LangChain's OpenAI integration to allow http_client=DefaultAioHttpClient() or an environment variable toggle.

Based on this issue

What changed

• New optional kwarg: Enhanced existing http_client and http_async_client parameters to support DefaultAioHttpClient
• Environment variable: Added LC_OPENAI_USE_AIOHTTP environment variable for convenient opt-in
• Graceful fallback: Warns when env var is set but openai[aiohttp] is not installed
• No default behavior changed: All changes are optional and backward compatible
• Unit tests: Comprehensive test coverage for all scenarios
• Documentation: Updated README with installation and usage examples

Why

Improve throughput for parallel chains, graphs, and agents without breaking existing users. The default HTTPX async client in OpenAI has concurrency limitations that can be removed by using the aiohttp backend.

How to enable

Option 1: Install and set environment variable

pip install "openai[aiohttp]"
# Set LC_OPENAI_USE_AIOHTTP=1 in your environment

from langchain_openai import ChatOpenAI

llm = ChatOpenAI()  # Will automatically use aiohttp if available

Option 2: Pass explicit client

pip install "openai[aiohttp]"

from openai import DefaultAioHttpClient
from langchain_openai import ChatOpenAI

llm = ChatOpenAI(
    http_client=DefaultAioHttpClient(),
    http_async_client=DefaultAioHttpClient()
)

Backwards compatibility

✅ No breaking changes - this is an optional parameter only
✅ Existing behavior preserved - default httpx clients still used by default
✅ Graceful degradation - warns and falls back if aiohttp not available
✅ Proper precedence - explicit client parameters override environment variables

Testing

• Unit tests for explicit client passing
• Unit tests for environment variable enabling/disabling
• Unit tests for fallback behavior with warnings
• Unit tests for truthy environment variable values
• Unit tests for precedence (explicit overrides env var)
• No network calls in tests (mocked OpenAI constructors)
• All existing tests still pass
• Linting and formatting checks pass

References

• OpenAI Python README - aiohttp backend documentation
• OpenAI Python issue #1596 - concurrency bottleneck with HTTPX
• LangChain contribution guidelines

Code Quality Checklist

• Breaking Changes: ✅ No public API changes
• Type Hints: ✅ All functions have complete type annotations
• Tests: ✅ New functionality is fully tested with comprehensive coverage
• Security: ✅ No dangerous patterns (proper import handling, graceful fallbacks)
• Documentation: ✅ Google-style docstrings and README updates
• Code Quality: ✅ make lint and make format pass
• Architecture: ✅ Clean separation of concerns, backward compatible design
• Commit Message: ✅ Follows Conventional Commits format

---

Ready for review 🚀

[vercel]

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

1 Skipped Deployment

Project	Deployment	Preview	Comments	Updated (UTC)
langchain	Ignored	Preview		Sep 24, 2025 8:37pm

[codspeed-hq]

CodSpeed Instrumentation Performance Report

Merging #33084 will not alter performance

_{Comparing NirAyalon1:feat/openai-aiohttp-backend (0866607) with master (b92b394)¹}

Summary

✅ 6 untouched
⏩ 11 skipped²

Footnotes

1. No successful run was found on master (50012d9) during the generation of this report, so b92b394 was used instead as the comparison base. There might be some changes unrelated to this pull request in this report. ↩

2. 11 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

[NirAyalon1]

Note: The PR title labeler workflow is failing due to permissions restrictions for fork PRs. The workflow should use pull_request_target trigger instead of pull_request to work properly with forks.