Meanwhile, after more than a month of waiting, Apple has released my app for managing actions on GitHub. Want to try it out? Available for iPhone: apps.apple.com/us/app/bosun...

Bosun App App - App Store

Download Bosun App by Daniele Scaratti on the App Store. See screenshots, ratings and reviews, user tips, and more apps like Bosun App.

apps.apple.com

I've been struggling with METAL ICB for two days now... This Indirect Command Buffer keeps crashing my terminal interface every time it throws an exception.

A little bit of @gleam.run tonight. Let's celebrate the new version 1.18 with a little game I made using étui. étui: hex.pm/packages/etui #minesweeper #gleam #tui #etui

Hey everyone, I built a lightweight vector database written in #Rust called Skeg. It’s disk-first, so it stays extremely RAM-frugal. repo is here: ⭐ github.com/skegdb/skeg Pruduct Hunt: 🆙 www.producthunt.com/products/git... thanks for the support! #opensource #producthunt

GitHub - skegdb/skeg: A multi-tenant vector database focused on extreme RAM efficiency. Lightweight, scalable, and optimized for high-density deployments.

A multi-tenant vector database focused on extreme RAM efficiency. Lightweight, scalable, and optimized for high-density deployments. - skegdb/skeg

github.com

I train models and run @real-comfyui.bsky.social on @runpod.io The point of that work is that you go do something else while it runs. Instead you sit there checking a dashboard. So I built Poddy Monitor. Native macOS menu bar! apps.apple.com/us/app/poddy... #cloud #llm #macos

Poddy Monitor App - App Store

Download Poddy Monitor by Daniele Scaratti on the App Store. See screenshots, ratings and reviews, user tips, and more apps like Poddy Monitor.

apps.apple.com

Skeg 0.7 almost out! 🔥 Off-thread maintenance just landed: p99 query latency under churn dropped from ~200ms to ~17ms. Default tier is tq2 (half the RAM, same recall). Need even less memory? tq1 is ready. github.com/skegdb/skeg Stars & Issue welcome! #VectorSearch #Rust

GitHub - skegdb/skeg: A multi-tenant vector database focused on extreme RAM efficiency. Lightweight, scalable, and optimized for high-density deployments.

A multi-tenant vector database focused on extreme RAM efficiency. Lightweight, scalable, and optimized for high-density deployments. - skegdb/skeg

github.com

Why are we allowing vector DBs to camp on GB of RAM? 🤯 skeg is a disk-first, vector engine built in Rust. Full vectors on SSD; quantized set stays in RAM. The 100K benchmark: skeg: 47MB | 1.0 Recall | 2.5ms • Qdrant: 885MB RAM | 0.997 Recall | 2.6ms 👉 github.com/skegdb/skeg

GitHub - skegdb/skeg: A multi-tenant vector database focused on extreme RAM efficiency. Lightweight, scalable, and optimized for high-density deployments.

A multi-tenant vector database focused on extreme RAM efficiency. Lightweight, scalable, and optimized for high-density deployments. - skegdb/skeg

github.com

I train models and run @real-comfyui.bsky.social on @runpod.io The point of that work is that you go do something else while it runs. Instead you sit there checking a dashboard. So I built Poddy Monitor. Native macOS menu bar! apps.apple.com/us/app/poddy... #cloud #llm #macos

Poddy Monitor App - App Store

Download Poddy Monitor by Daniele Scaratti on the App Store. See screenshots, ratings and reviews, user tips, and more apps like Poddy Monitor.

apps.apple.com

I've released a small app for reading #manga on #macOS. No distractions. No libraries or accounts. Just drag the archive onto the icon in the dock and start reading. You can even view pages side by side! No history. No awkwardness. maki.altumdream.com

Maki, a minimalist manga reader for macOS

Drop a ZIP or folder on the Dock and read. Immersive, distraction-free, RTL-first, local-only. No library, no accounts, no telemetry.

maki.altumdream.com

I've released a small app for reading #manga on #macOS. No distractions. No libraries or accounts. Just drag the archive onto the icon in the dock and start reading. You can even view pages side by side! No history. No awkwardness. maki.altumdream.com

Maki, a minimalist manga reader for macOS

Drop a ZIP or folder on the Dock and read. Immersive, distraction-free, RTL-first, local-only. No library, no accounts, no telemetry.

maki.altumdream.com

Is it a rat? Is it a mutant? It doesn't matter… because in the end, you'll see it dance. Stay tuned.

Skeg update: if everything goes well, I should be able to halve the current RAM usage while keeping nearly identical recall and slightly better latencies. Fingers crossed 🤞 github.com/skegdb/skeg

GitHub - skegdb/skeg: Vector database for AI agents. Context layer, multi-tenant, RAM-frugal. Runs anywhere your model runs.

Vector database for AI agents. Context layer, multi-tenant, RAM-frugal. Runs anywhere your model runs. - skegdb/skeg

github.com

Three skeg instances. Three separate corpora. One answer. The federation layer queries each instance independently. No central store. Each agent keeps its own memory, partitioned, on its own footprint. RAG without collapsing everything into one bucket. Running on an MacBook Pro M1 with 16GB.

I'll never be a composer. But étui is quite rewarding. I need to stop following the news about @ratatui.rs 😂 hex.pm/packages/etui github.com/lupodevelop/... etui.altumdream.com

I'll never be a composer. But étui is quite rewarding. I need to stop following the news about @ratatui.rs 😂 hex.pm/packages/etui github.com/lupodevelop/... etui.altumdream.com

I was thinking... it wouldn't be a bad idea to have a TUI app jam github.com/lupodevelop/...

I'm starting to publish the Skeg benchmarks. This includes the case with Llama 3.2, which uses various databases to perform RAG. ⭐️ github.com/skegdb/skeg 📊 skegdb.github.io/bench/

GitHub - skegdb/skeg: Vector database for AI agents. Context layer, multi-tenant, RAM-frugal. Runs anywhere your model runs.

Vector database for AI agents. Context layer, multi-tenant, RAM-frugal. Runs anywhere your model runs. - skegdb/skeg

github.com

@ratatui.rs inspired all of this. Now Étui is live! #gleam #beam #erlang #js

Daniele@lupodevelop.bsky.social · 2mo ago

Great. I showed you some things earlier, and now the TUI library for Gleam is publicly available. I kept you waiting, so here’s a page with a live demo created by ètui. 📟 etui.altumdream.com ⭐ github.com/lupodevelop/... 📦 hex.pm/packages/etui

I'd love for Skeg to get some love. Or even just some feedback... after 8 months, it would be great to see someone give it a try. github.com/skegdb/skeg

GitHub - skegdb/skeg: Vector database for AI agents. Context layer, multi-tenant, RAM-frugal. Runs anywhere your model runs.

Vector database for AI agents. Context layer, multi-tenant, RAM-frugal. Runs anywhere your model runs. - skegdb/skeg

github.com

étui /eɪˈtwiː/ · n. · a case for holding small precious things It's a French word. I thought it fit well.

Great. I showed you some things earlier, and now the TUI library for Gleam is publicly available. I kept you waiting, so here’s a page with a live demo created by ètui. 📟 etui.altumdream.com ⭐ github.com/lupodevelop/... 📦 hex.pm/packages/etui

I'm almost done preparing the TUI library for release.

Today's schedule is a bit lighter than yesterday's. 🐕 Let's start with the 1.8.0 release of Woof (a logging library for Gleam). 🎁 Then I have a promise to keep. And yes, it's about the TUI library... ⚓ Has anyone had a chance to check it out, Skeg?

Let's try out the Hacker News experience

I built a vector database that fits in the RAM your #Slack tab is using... in any case, always less than a Visual Studio Code process. #vscode #llm #rag #ai #rust

Daniele@lupodevelop.bsky.social · 2mo ago

Tired of vector DBs killing your RAM? Skeg is made for big LLMs on the same machine. - 1M vectors → ~400 MB RAM - 1-4 bit quantization + reranking - Fast Rust Long-term memory for your agent without downgrading the model. github.com/skegdb/skeg #LocalLLM #RAG #AI #Agents

Tired of vector DBs killing your RAM? Skeg is made for big LLMs on the same machine. - 1M vectors → ~400 MB RAM - 1-4 bit quantization + reranking - Fast Rust Long-term memory for your agent without downgrading the model. github.com/skegdb/skeg #LocalLLM #RAG #AI #Agents

GitHub - skegdb/skeg: Vector database for AI agents. Context layer, multi-tenant, RAM-frugal. Runs anywhere your model runs.

Vector database for AI agents. Context layer, multi-tenant, RAM-frugal. Runs anywhere your model runs. - skegdb/skeg

github.com

A little background. Skeg is written entirely in Rust. It started as a way to pass the time by reading through a mountain of research papers.

I'd like to get better at promoting my work. Anyway, I'd love to hear what you think of Skeg. ⭐ Maybe a few encouraging stars?

Daniele@lupodevelop.bsky.social · 2mo ago

1M vectors. 419 MiB RSS. recall@10 ≥ 0.998. p99 < 4ms. qdrant-hnsw needs 4162 MiB for the same corpus. skeg does it in one tenth. disk-resident Vamana index, SIMD NEON kernels, PQ + int8 + TurboQuant tiers. built for Apple Silicon M1 alongside a running LLM.

1M vectors. 419 MiB RSS. recall@10 ≥ 0.998. p99 < 4ms. qdrant-hnsw needs 4162 MiB for the same corpus. skeg does it in one tenth. disk-resident Vamana index, SIMD NEON kernels, PQ + int8 + TurboQuant tiers. built for Apple Silicon M1 alongside a running LLM.

Daniele@lupodevelop.bsky.social · 2mo ago

I know. You've been waiting for my TUI for Gleam. But hey, I've released Skeg, a RAM-efficient context layer. github.com/skegdb/skeg I'll make it up to you soon.

I know. You've been waiting for my TUI for Gleam. But hey, I've released Skeg, a RAM-efficient context layer. github.com/skegdb/skeg I'll make it up to you soon.

GitHub - skegdb/skeg: Vector database for AI agents. Context layer, multi-tenant, RAM-frugal. Runs anywhere your model runs.

Vector database for AI agents. Context layer, multi-tenant, RAM-frugal. Runs anywhere your model runs. - skegdb/skeg

github.com