Fetching from the wire…
01
02
03
04
05
06
07
08
Infra2026-07-26 · source-backed
A Rust/io_uring storage engine posted to Show HN today moves KV cache off the GPU into a petabyte-scale store co-located on GPU hosts, so a prefix computed on one GPU serves to others (GitHub). Claims million-plus IOPS within 1ms and 66x TTFT speedup on cache hits at 128K context. The demo serves Gemma4-31B on an H100 at 256K context, plugging into vLLM through an openlake-vllm connector via --kv-transfer-config. Apache-2.0, 2.3K stars, 423 forks, 164 commits, 91 open issues. Early, but the most concretely useful long-context cost lever posted this week.
Each link below shares sources, entities, or timing with this story.
Triple-stream retrieval (BM25 keyword, vector embeddings, knowledge-graph traversal) fused via Reciprocal Rank Fusion on the iii engine, with SQLite for state and an in-memory vector index, no external database. The economic claim: ~170K tokens/year (~$10) versus ~650K tokens...
| Project | Stars | What It Does | Why It Matters | |---------|-------|-------------|----------------| | Worktrunk | 2.7K | Rust CLI for Git worktrees with AI agents | Auto-squash/rebase/merge, LLM commit messages, build cache sharing. From PRQL creator. | | Timber | 545 | Com...
macro-inc/macro hit #4 on GitHub trending at 2,821 stars total, shipping email, chat, docs, tasks, calls and CRM that all @-link into a shared context graph agents read as team-level memory. Explicitly "fully open source, not open core," with commercial licensing sold separate...
Published August 4, Exponent Aligned ANS splits BF16 into its 8-bit exponent field and its sign/mantissa, aligns exponents to a block-local center, then entropy-codes only the low-entropy exponent stream while storing sign and mantissa raw. 1.51x lossless, 622 GB/s median deco...
Created August 10, Apache-2.0, at 824 stars. Structure first, pixels only when the agent actually needs them, so page fetch, extraction, and search stay cheap. GitHub The default in browser-agent stacks is headless Chrome per agent, which is absurd on cost and memory when 90%...
+106 stars today to 1,467, Apache-2.0, installs with npm install -g @magnitudedev/cli and runs fully offline, claiming to work out of the box on any hardware rather than requiring a GPU class check. GitHub Different bet from the dominant pattern this week, where nearly every n...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.