Simon Willison has released version 0.32rc1 of his LLM tool, completing the architectural work begun in the April alpha. This update introduces a new schema design that captures prompt and response details from the latest model families much more effectively. The most significant change involves using content-addressable hash IDs for stored messages. This approach enables database de-duplication and allows the tool to represent trees of messages for forked conversations.
The release also adds support for three new OpenAI models, specifically gpt-5.6-sol, gpt-5.6-terra, and gpt-5.6-luna. Because the update involves a significant schema change with new tables, existing logs.db files remain unaffected. Users should still run a backup before upgrading to the release candidate.
* New tables only are added to the database structure
* Old data is preserved without migration issues
* Backup command is `llm logs backup logs-backup.db`
