Aug 24 to Aug 30, 2026

Mac router you can switch off again that sends Grok Bot's model calls to Codex or OpenRouter. It checks the app version before patching anything.

Restart-resumable local agent stack that runs Qwen3.8 on one RTX 3090/4090/5090.

Convenience script that boots the model on a single DGX Spark or variant.

Reproducible Strix Halo setup with receipts for speed and long context.

Another route to fit the NVFP4 checkpoint on one Spark by offloading its embedding table.

Manuel Casalta Petitjean ported the Windows client to Linux until the official Linux builds shipped.

Tool to add knowledge through the PLE n-gram table.

Custom 4-bit llama.cpp build for AMD Strix Halo mini PCs.

Early notes on fitting the 126GiB checkpoint onto one DGX Spark.

Open-source community port that runs the macOS/Windows-only Grok Bot desktop client on Linux. Shared on Hacker News, where it reached 33 points.

One prompt installs a panel to add API or ChatGPT-login providers and switch with /gs commands across three protocols.

A local OpenAI-compatible proxy translating Grok Build requests into Cursor's native inference protocol, with multi-turn tool calls.

262K context with vision and FP8 KV cache on four used GPUs.

Verified guide for running the model on a high-memory Mac.

Fits the model by quantising the embedding table too.

Open-source system of bots specialised in marketing and distribution.

262K context with exact sampled MTP on a single workstation card.

A read-only CLI that reports Cursor, Grok Bot and SuperGrok usage so bots can check consumption before burning it.

Day-one single Spark build.

TP2 with expert parallelism and MTP speculation.

A homelab owner's practical write-up and configs.

Ampere-compatible reader plus benchmark receipts.

Compresses the embedding table so a single Spark holds the whole thing at long context.

Draft-head speculation and carefully measured Vulkan settings that lift Flash-Next speed on Strix Halo. Notable for publishing the measurement method as well as the numbers.