MechanismArticle
A content-addressable message store modeled after Git solves the duplicate-JSON logging problem inherent in chat-completions-style APIs where message sequences are appended to on every request.
To avoid logging duplicate JSON for every turn in a growing message sequence, LLM 0.32 introduces a Git-like content-addressable message store that the log commands can reconstruct into human-readable format. ✦ AI generated
Simon Willison · Simon Willison's Weblog · 2026-08-04 · original ↗
The bigger challenge with that kind of API concerns logging. If we're going to support the pattern where the message sequence is appended to on every request, ideally we can avoid logging all of that duplicate JSON for every turn. The solution is the new content-addressable message store, modeled after Git. You can see the new schema for that in the documentation, but the llm logs and llm logs --json commands have both been upgraded to convert that format back into something that's easy to consume.
Read full article ↗excerpt · fair-use quotation
Around this claim
This moment responds to
explains mechanism → Content-addressable hash IDs for stored messages enable deduplication in the database and allow LLM to represent trees of messages for forked conversations.Release author · Simon Willison's Weblogprovides context → Content-addressable logs in LLM 0.32rc1 are designed to support OpenAI Chat Completion style requests where each incoming message extends the previous conversation.Author · Simon Willison's Weblog