datasette-llm 0.1a8

The bug fixed here, broken collection of chained LLM responses, is the same class of problem addressed the same day in datasette-llm-accountant 0.1a4, suggesting Willison is doing a coordinated sweep of response-chain reliability across the entire plugin family rather than patching in isolation.

The datasette-llm-accountant 0.1a4 release from the same day explicitly called out a fix for response chain tracking, and this release addresses the upstream hook that accountant depends on. That pairing matters: cost and observability tooling is only as reliable as the response collection layer beneath it. Meanwhile, the llm-gemini 0.32a0 release introduced streaming reasoning tokens with a versioned dependency on llm>=0.32a0, which points to a broader coordinated version bump across Willison's plugin stack. These three same-day releases look less like coincidence and more like a deliberate stabilization pass ahead of wider adoption.

If datasette-llm reaches 0.1 stable (dropping the alpha tag) within the next two months, that would confirm this sweep was a pre-release hardening effort. If the alpha tag persists past summer, the response-chain architecture likely still has unresolved design questions.