LLM Command-Line Tool Gets Patch After OpenAI Library Change

👁 117 reads

Simon Willison released LLM 0.32.1 on 21 August 2026 to repair broken fresh installs of his popular command-line tool, after a change in the OpenAI Python library exposed a fragile transitive dependency on httpx that had gone unnoticed until it stopped working.

What broke and why

The failure traced back to the OpenAI Python library removing its own use of httpx. Because LLM never declared httpx as a direct dependency - it arrived only as a side-effect of installing the openai package - any environment that pulled in the newer OpenAI library suddenly found httpx missing. The result was that fresh installs of LLM failed entirely, with no graceful fallback.

This class of breakage, sometimes called a transitive or phantom dependency, is a well-known hazard in Python packaging. A project works reliably for months or years because a dependency it relies on happens to bring along a second library the project also needs. When the first dependency is updated and drops that second library, the downstream project breaks without having changed a single line of its own code. For end users who simply ran a standard install command, the tool stopped working for no apparent reason on their end.

Willison described the chain of events plainly in his release notes: the OpenAI Python library dropped its usage of httpx, and it turned out LLM depended on that library but only installed it via a transitive openai dependency.

The fix in 0.32.1 - and what comes next

The immediate solution in 0.32.1 is a pin: the release constrains the openai dependency to versions below 3, expressed as openai<3. That keeps the older OpenAI library - which still carries httpx along with it - in place for users who install LLM today. It is a conservative, defensive move that restores functionality without requiring any architectural changes to LLM itself.

Pins like this are a standard emergency measure in Python projects. They restore stability quickly but carry their own cost: a version ceiling on a core dependency means LLM users cannot take advantage of features or fixes in newer OpenAI library releases until the pin is lifted. Left in place too long, it can create its own compatibility headaches as the rest of the ecosystem moves forward.

Willison has acknowledged that limitation directly. A forthcoming 0.33 release, described as soon-to-drop in his notes, will switch LLM from httpx to httpx2. That migration will sever the implicit reliance on the OpenAI library as a vehicle for delivering a networking dependency, and should allow the openai pin to be removed. The 0.33 release would also bring LLM in line with the direction the broader Python HTTP ecosystem appears to be taking, as httpx2 supersedes the original httpx library.

Why this matters for developers building with LLM

LLM is a widely used open-source tool that lets developers and power users query large language models directly from the command line, without writing application code or navigating a web interface. Because it sits at the intersection of several fast-moving libraries - openai, httpx and others - it is particularly exposed to the kind of dependency churn that has accelerated alongside rapid iteration in AI tooling.

For individual developers or small teams who rely on LLM in scripts, automation pipelines or daily workflows, a silent install failure is a significant disruption. Unlike a crash inside a running application, a broken install produces no useful error message traceable to the actual cause without some investigation. Users unfamiliar with Python dependency resolution might spend time debugging their own environment before identifying that the breakage originated upstream.

The episode is also a reminder of how tightly the AI developer tooling ecosystem is coupled to a small number of core libraries. The OpenAI Python library is a dependency - direct or indirect - for an enormous number of projects. When it makes a breaking change, the downstream effects ripple outward in ways that are difficult to anticipate entirely, even for experienced maintainers.

Willison maintains a large and active body of open-source work in the AI tooling space, and the speed of the response - a patch release and a clear roadmap for the longer-term fix published the same day the problem became apparent - reflects the kind of maintenance posture that keeps widely used developer tools trustworthy over time.

Users on affected versions should upgrade to 0.32.1 immediately. Those watching for the httpx2 migration should expect it in the 0.33 release, which Willison indicated is close to shipping. Until then, the openai pin provides a stable, if temporary, floor.

// read moreLLM 0.33 Adds Repeatable Templates, OpenAI Library 3.x SupportAI ToolsLLM-Anthropic Plugin 0.27 Updates for Python Library 1.0AI ToolsOpenAI Develops AI Tools for Seniors: Everyday ApplicationsNews

Kinetyra