task: assess the v0.33.2 upstream sync — blocked on ADR 0009 - #232

Merged

glennneuber merged 23 commits into

mainfrom

task/upstream-sync-0.33.2

Aug 29, 2026

Commits on Aug 26, 2026

cmake: make external compat patches idempotent (#17948)

pd95
and
codex
authored
ad94d52
MLX: version bump (#17955)
dhiltgen
authored
d465dc7
llama.cpp: version bump b10630 (#18003)
dhiltgen
authored
3d86a55
llm: remove the unused Grammar completion-request field
jessegross
committed
7027546
mlx: return exact types and evaluate arrays in the value readers
jessegross
committed
7623501
mlxrunner: add structured output support

jessegross
and
dhiltgen
committed
147509c
mlxrunner: avoid Metal GPU timeouts when loading models from slow storage
jessegross
committed
77e3b0a
docs: document Claude Desktop integration (#18004)
ParthSareen
authored
9dbf139
docs: claude docs (#18006)
ParthSareen
authored
e2c6c7e
MLX: Qwen3.8 Flash Next support (#18032)
dhiltgen
authored
18ea9de
linux: fix llama.cpp docker build (#18040)
dhiltgen
authored
d3a8af4
app: prevent Claude model picker scroll jumps (#18041)
hoyyeva
authored
a773b7a
launch: respect OLLAMA_HOST when configuring Pi (#18039)
dhiltgen
authored
91cf995
mlx: remove the text-only gemma3 model
jessegross
committed
3b96a89
app: consume Claude recommendation mappings (#18042)
ParthSareen
authored
13f2fb8

Commits on Aug 27, 2026

Commits on Aug 28, 2026

lint fix (#18081)
dhiltgen
authored
f96e7aa

Commits on Aug 29, 2026

docs/maxusai: assess the v0.33.2 sync; blocked on ADR 0009

glennneuber
and
claude
committed
c59112d
Merge upstream v0.33.2; adopt upstream's MLX grammar engine (ADR 0033)

glennneuber
and
claude
committed
975b884