LLM 0.32.1 pins OpenAI dependency to restore broken fresh installs after httpx removal

The breakage occurred because LLM did not explicitly declare httpx as a direct dependency. Instead, it relied on OpenAI’s Python library to pull in httpx transitively. When OpenAI removed httpx, fresh installs of LLM failed because the missing dependency was no longer installed. This is a classic example of how transitive dependencies can create hidden failure modes in software that appears stable.

The immediate fix in LLM 0.32.1 is to pin the OpenAI dependency to a version that still includes httpx. This restores functionality but does not address the root cause. Pinning dependencies is a common short-term solution, but it can lead to technical debt if not followed by a more permanent fix. The upcoming 0.33 release will replace httpx with httpx2, which suggests the project is moving away from the older library entirely.

For engineers using or maintaining LLM, this incident underscores the importance of explicitly declaring all dependencies, even those pulled in transitively. It also highlights the risks of relying on third-party libraries that may change their own dependencies without warning. Projects that depend on OpenAI’s Python library should audit their dependency trees to ensure they are not silently relying on transitive dependencies that could disappear in future updates.