RepoJournal
Local LLMs Local LLMs
93 wires and counting

$ follow Local LLMs

Keep up with Local LLMs in about 3 minutes: what actually shipped — the commits, pull requests, releases, and security advisories that matter.

or

fair warning: these emails are deeply technical. diffs, version numbers, CVEs, benchmark deltas. if that's not your idea of a good read, this isn't your newsletter.

Folds into your digest — weekly by default, monthly if you prefer. Unsubscribe in one click.

$ status

wire 2026-08-25
stories 229
RSS Archive Reviews

© 2026 RepoJournal Home Showcase How it works Privacy

$ the-wire · showcase

OLLAMA BOLSTERS CLAUDE DESKTOP, VLLM HARDENS AUDIO PATHS

By RepoJournal · Filed · About Local LLMs · Composed from the cited sources · methodology

hoyyeva ParthSareen Juan Pérez de Algaba mickqian ispobock Georgi Gerganov YiChen Lv
12 people shipped this
hoyyeva @hoyyeva 3 cited ParthSareen @ParthSareen 2 cited Juan Pérez de Algaba @jperezdealgaba 2 cited mickqian @mickqian 2 cited ispobock @ispobock 2 cited Georgi Gerganov @ggerganov 1 cited YiChen Lv @forforever73 1 cited Saad Ali @NIXKnight 1 cited Harry Mellor @hmellor 1 cited Jee Jee Li @jeejeelee 1 cited Jiminator @Jiminator 1 cited Ning Xie @andyxning

Ollama's desktop app just made Claude Desktop integration faster and safer to manage, while vLLM closed a security gap on audio file size limits.

Ollama is tightening its Claude Desktop experience with three fixes: explicit model mappings [1], a default-on Auto mode [2], and faster toggle responsiveness [3] that replaces blocking model checks with a lightweight connection summary. The same team also killed a duplicated first-use modal and a stale integration chevron [4], with a scrollable integrations list to match [5]. Across the stack, llama.cpp shipped a ccache-clear tweak [6] to prevent CI cache rot, a per-device FlashAttention tuning table for Metal [7], and a grammar fix that makes hyphens literal in character classes [8], so tool-call grammars stop failing on escaped dashes. vLLM now enforces VLLM_MAX_AUDIO_CLIP_FILESIZE_MB on every audio path [9], a security fix worth shipping before you touch any multimodal pipeline, and documents the UUID exposure risk that comes with it [10]. Model cleanup continues: vLLM removes ten deprecated architectures [11] and deprecated parameters [12], so plan a migration if you rely on any of them. SGLang fixed a nasty decode retraction bug that could resume a request on someone else's recurrent state [13], and added a test for MXFP8 KV state transfer on Inkling [14]. It also lands GGUF and NVFP4 support for MiniMax H3 diffusion [15] [16], plus docs that split Qwen3.8-27B NVFP4 export cells by lm_head precision into two checkpoints [17].

Action items

References

  1. [1] app: add Claude Desktop model mappings ↗ ollama/ollama
  2. [2] app: add Claude Desktop Auto mode setting ↗ ollama/ollama
  3. [3] app: improve desktop integration responsiveness ↗ ollama/ollama
  4. [4] app: fix desktop interaction regressions ↗ ollama/ollama
  5. [5] app: make integrations list scrollable ↗ ollama/ollama
  6. [6] ci : apply ccache-clear with older/min/dry-run to all ccache jobs ↗ ggml-org/llama.cpp
  7. [7] metal : per-device tuned (Q, NE) for flash-attn vec (#26570) ↗ ggml-org/llama.cpp
  8. [8] grammar : parse \- in char classes as literal hyphen (#27591) ↗ ggml-org/llama.cpp
  9. [9] fix(security): enforce VLLM_MAX_AUDIO_CLIP_FILESIZE_MB on all audio paths (#53561) vllm-project/vllm ↗
  10. [10] [Docs][Security] Document multimodal media UUID security implications (#53582) vllm-project/vllm ↗
  11. [11] [Model] Remove ten deprecated model architectures (#53608) vllm-project/vllm ↗
  12. [12] [MISC] Cleanup deprecated parameters (#53559) vllm-project/vllm ↗
  13. [13] Fix recurrent state loss on decode retraction ↗ sgl-project/sglang
  14. [14] Add PD test for inkling with mxfp8 KV ↗ sgl-project/sglang
  15. [15] [Diffusion] Load MiniMax H3 GGUF text encoders ↗ sgl-project/sglang
  16. [16] [Diffusion] Load Comfy NVFP4 MiniMax H3 checkpoints ↗ sgl-project/sglang
  17. [17] [docs] Split the Qwen3.8-27B NVFP4 cells by lm_head precision ↗ sgl-project/sglang

$ ls local-llm/reviews/ # the step-back reads

the week in review vLLM v0.32.0 lands systemone, security fixes behind it the month in review vLLM 0.30.0 lands and nineteen engine-fatal flaws close
Full archive → Weekly reviews →

Quick answers

What shipped in Local LLMs on August 25, 2026?
Ollama's desktop app just made Claude Desktop integration faster and safer to manage, while vLLM closed a security gap on audio file size limits. In total, 110 commits, 110 pull requests, and 9 releases landed.
Who contributed to Local LLMs on August 25, 2026?
12 developers shipped this update, including hoyyeva, ParthSareen, Georgi Gerganov, YiChen Lv, Saad Ali, Juan Pérez de Algaba, Harry Mellor, and Jee Jee Li, and 4 more.
What were the notable Local LLMs updates?
app: add Claude Desktop model mappings, app: add Claude Desktop Auto mode setting, and app: improve desktop integration responsiveness.
♥

$ log Local LLMs --recent

Ollama preps Phonon voice input, vLLM drops deprecated IR knob 269 shipped Ollama fixes cloud metadata for every launcher, MLX speeds 6-16 row decode ~3x 298 shipped Ollama streams real errors, llama.cpp forwards sampling params on /v1/messages 274 shipped Five vLLM crash and forgery paths, realtime transcription lands in llama.cpp 295 shipped

$ log --all # elsewhere on the wire

Alignment & Post-training MILES ships a 25-benchmark eval panel and simpler config presets Interpretability J-space direction loss and Jacobian lens joint ablation fixed in v4.3.0

$ ls showcases/

Safety Evals AI Red Teaming Agent Security Interpretability Alignment & Post-training

$ init your-desk

RepoJournal can write one of these for any repo you care about. Sign in with GitHub, pick your repos and sources, and your desk gets a wire like this one every morning.