1min.

1MIN · PYTHON

LLM 0.37.1 preserves system prompts when messages are also supplied

LLM 0.37.1 fixes a Python API path where supplying both system and messages could drop the system prompt for models constructing requests from message lists. The release also repairs endpoint configuration after removing the last saved endpoint and removes an extra newline from binary embedding output. These are integration fixes: a successful response alone would not reveal that important input had been omitted.

Simon Willison · LLM · Nov 3, 2026

Why it matters

Client libraries need verification of the request they send and the bytes they return, not just a check that a model answered.

More context

Try the released maintenance version on a small reproduction of the affected API call. It also handles non-streaming responses without usage metadata and repairs several log-search cases. LLM is a Python CLI and library; model credentials, provider charges and installed plugins remain separate setup concerns.

Keep in mind

The system-prompt fix concerns the documented combination of arguments and request-building behavior; it is not a guarantee that a model obeys every instruction it receives.

Original read

More engineering ideas, one minute at a time.

Explore more stories