Briglia CLI 0.2.17
Briglia 0.2.17 adds a model, removes a request flag that a provider backend has started rejecting, and closes a storage edge case. Ordinary turns are unchanged in everything except that flag.
DeepSeek V4.1 Flash. OpenCode serves it under the id deepseek-flash (text and image input, reasoning, tools, 1M context). It is now in the curated catalogue, so it appears in /model, the Telegram model keyboard, setup and Quick Setup. The reasoning classification also recognises the new id: its reasoning is stored and replayed between rounds and turns like the other DeepSeek models. Region availability follows your OpenCode workspace settings.
reasoning_history is no longer sent. A fresh install on glm-5.3-flash failed every request with HTTP 400 "unknown field reasoning_history": OpenCode now serves that model from two backends, chosen per session, and the newer one rejects the flag. The same happened to Kimi K3 on 1 September. Measured live with Briglia's request shapes on Kimi K2.6, GLM 5.3 Flash (both backends), Qwen 3.8 Max and DeepSeek, the flag never changed what any backend did: reasoning output, replay visibility, tool loops, images, streaming and billed prompt tokens were identical with and without it. It is therefore removed for every model rather than special-cased again. Reasoning replay itself, the reasoning_content on earlier assistant messages, is untouched. If your install was stuck on the rejecting backend, /upgrade is the fix; /rotateaffinity is no longer needed.
Untracked raw archive files are reconciled at startup. If an index write completed its rename but failed on the directory flush, a restart could leave a raw archive file and its sidecar referenced by neither the chunk index nor the pending index. Startup recovery now removes such files, only when both indexes are present and decode, only for files older than one hour, and never as a consequence of a failed save. A missing or dangling index is treated as missing metadata, not as an empty index, so recoverable history is preserved.
The pinned Chat Completions request baselines pass through a reviewed migration that strips exactly the removed 32-byte flag from the recorded requests; everything else remains byte-identical.
Update with /upgrade in Briglia or briglia upgrade in the terminal.
Signed release, sequence 77. Clients authenticate manifest.sig.json with the pinned Ed25519 key before trusting any asset.