Inkling: Our Open-Weights Model

380 vimarsh6739 2026-07-15 18:32:16 UTC thinkingmachines.ai Qwen Share 98

The company released Inkling, a 975B-parameter Mixture-of-Experts model with 41B active parameters, supporting text, images, audio, and video, alongside Inkling-Small (12B active) for efficient customization via Tinker. Trained on 45T tokens, Inkling offers controllable reasoning, safety features, and is available on platforms like Hugging Face and Tinker, with partnerships for deployment and ...

Stripe and Advent have made a joint offer to acquire PayPal – sources

251 rvz 2026-07-15 16:32:16 UTC reuters.com Full Share 129

Please enable JS and disable any ad blocker

Duskers, the scary command line game, is getting a sequel

57 spacemarine1 2026-07-15 19:32:16 UTC elbowgreasegames.substack.com Qwen Share 8

Misfits Attic announced *Duskers 2.0*, funded by Max McGuire’s Stray Signal, during PC Gamer’s event, expanding the survival game’s narrative with moral choices and deeper gameplay. EGG’s community-driven funding model, supporting indie developers like Tim through prototypes and partnerships, highlights a new era of collaborative game development.

Running Gemma 4 26B at 5 tokens/sec on a 13-year-old Xeon with no GPU

194 neomindryan 2026-07-15 16:32:16 UTC neomindlabs.com Qwen Share 110

A 13-year-old HP server with Ivy Bridge Xeons runs Google's 26B-parameter Gemma 4 model at ~5 tokens/second via CPU-optimized code, overcoming AVX2 instruction set limitations with custom patches. The project highlights adapting modern AI to legacy hardware through engineering, not just subscriptions, enabling local inference on outdated systems.

Who Is America's Homer?

15 samclemens 2026-07-15 20:32:16 UTC plough.com Qwen Share 10

The article explores candidates for America’s "Homer," including Frost, Dickinson, and Wilder, debating their poetic range and cultural resonance. It argues that America’s fragmented identity and youth hinder a unifying national epic, unlike ancient Greece or Rome.

Brainless: Shadcn components that look like Claude Code, Codex and Grok

30 benswerd 2026-07-15 20:32:16 UTC brainless.swerdlow.dev Full Share 5

Claude Code, Codex, and Grok interfaces as shadcn components.

Collection of Digital Clock Designs

119 levmiseri 2026-07-15 17:32:16 UTC clocks.dev Full Share 31

A collection of thoughtfully designed digital clocks.

Voxatron

26 lsferreira42 2026-07-15 20:32:16 UTC lexaloffle.com Full Share 9

Enter your email address below, choose a payment method, and you will be emailed a download link. Purchasing Voxatron also gives you access to all future updates. If you have any trouble, please feel free to email us.

Show HN: misa77 - a codec that decodes 2x faster than LZ4 (at better ratios)

111 nonadhocproblem 2026-07-15 16:32:16 UTC github.com/welcome-to-the-sunny-side Qwen Share 36

misa77 is an LZ-based codec optimized for write-once, read-many scenarios, offering constant memory use (<5 MB) and faster decompression for highly compressed files, with experimental modes enhancing CPU efficiency. It balances compression ratios and throughput, excelling in decode speed for compressible data while supporting CLI-driven compression and parameter tuning.

Mysteries of Telegram Data Centers (2022)

217 theanonymousone 2026-07-15 13:32:16 UTC dev.moe Full Share 108

Grok Build is open source

10 runesoerensen 2026-07-15 20:48:16 UTC x.ai Qwen Share 2

Your action triggered a security block, possibly due to specific phrases, SQL commands, or malformed data. Contact the site owner with your activity details and the Cloudflare Ray ID for resolution.

Prioritize mental health, and why communication is so important

253 ramon156 2026-07-15 11:32:16 UTC ramones.dev Qwen Share 200

Struggling with severe depression and career setbacks, the user acknowledges personal challenges in motivation and discipline, now focusing on healing with support. They seek stability, improved work discipline, and meaningful contributions while navigating mental health.

MITS: Rockets, Calculators, and Personal Computers

9 BirAdam 2026-07-15 20:32:16 UTC abortretry.fail Prefix Share 0

Henry Edward Roberts was a mustang. He had enlisted in the USAF in May of 1962, and at the age of 27, had gained his commission in 1968 after attending Oklahoma State University and earning a degree in electrical engineering. As a second lieutenant in the Air Force service at the Effects Branch of the Air Force Weapons Laboratory at Kirtland AFB, he’d been tasked with procuring a ...

Towards a harness that can do anything

144 evakhoury 2026-07-15 14:32:16 UTC eardatasci.github.io Qwen Share 72

The author advocates for transparent, Unix-inspired LLM harnesses that minimize cognitive load by leveraging the model's existing systems knowledge, modular tools, and event-driven architectures like Ambiance to enable efficient, self-directed agent workflows. Key principles include deterministic goal pursuit, flat text interfaces, structured data storage, and robust logging to ensure ...

Artie (YC S23) Is Hiring Software Engineers

0 2026-07-15 17:32:16 UTC jobs.ashbyhq.com Full Share

Artie Jobs

Designing APIs for Agents

14 benswerd 2026-07-15 19:32:16 UTC freestyle.sh Qwen Share 4

Designing APIs for agents prioritizes clarity, explicitness, and precise error handling over human-centric utilities, as agents require unambiguous documentation and direct code interpretation to function effectively without relying on defaults or abstractions. The author advocates for minimalistic, self-explanatory APIs with explicit field names and comprehensive guides, emphasizing that ...

Show HN: Low-latency local LLM runner via OpenJDK Panama FFM (Java 22)

10 KingJoker 2026-07-15 19:32:16 UTC github.com/projectargus-cc Qwen Share 2

libargus is a zero-allocation native AI runtime consolidating Vision, Speech, and LLM pipelines via Project Panama FFM, built on GGML/llama.cpp for high-performance, thread-safe execution. It enables JVM integration with Java bindings, supports multimodal processing, and offers zero-copy memory, KV cache quantization, and CUDA acceleration for low-latency inference.

Open-source memory for coding agents, synced over SSH

88 vshulcz 2026-07-15 16:32:16 UTC github.com/vshulcz Qwen Share 19

Deja is a zero-dependency binary that indexes and searches code assistant histories, redacting secrets and syncing across machines via local files or SSH. It offers fast, privacy-focused memory recall without models or network access, leveraging existing logs for retroactive insights.

Book Prizes Don't Work How You Think

7 samclemens 2026-07-15 20:32:16 UTC rebeccamakkai.substack.com Prefix Share 0

Rebecca Makkai's Substack: mostly writing and books! but also Zillow, true crime, etymology, food, general inanity By subscribing, you agree Substack's Terms of Use, and acknowledge its Information Collection Notice and Privacy Policy. I want to start by preventing some heart attacks and assuring the administrators of every book prize I’ve ever judged that I am NOT about to disclose ...

Today I Rescued 7,234 Old GIFs

70 birdculture 2026-07-15 15:32:16 UTC danq.me Qwen Share 5

GlitchyZorua and I automated archiving Ibiblio's 1990s GIF icons by calculating server-side imagemap coordinates for bulk downloads, bypassing manual clicks. We built a searchable static site (ibiblio-icon-archive.danq.dev) to preserve and share the collection.

Show HN: Firefox in WebAssembly

5 coolelectronics 2026-07-15 21:01:16 UTC developer.puter.com Full Share 0

The Gecko engine compiled to WebAssembly. The real Firefox interface, rendering entirely inside this browser tab. GPU acceleration WebGL-based rendering for web content. Enable JIT Experimental JS → WASM JIT. Expect bugs. Wisp server WebSocket proxy that tabs load web content through. Networking in the browser powered by Puter.js

Sleep regularity is a stronger predictor of mortality risk than sleep duration (2023)

611 bilsbie 2026-07-15 12:32:16 UTC academic.oup.com Full Share 306

Show HN: Capn-hook for coding agents – don't grep the same mystery twice

22 knrz 2026-07-15 18:32:16 UTC github.com/cyrusnuevodia Qwen Share 5

Capn saves coding agents' discovered codebase answers, auto-deleting outdated entries, reducing token use by 77% and rework. It integrates with Claude Code/Codex, enabling fast, accurate recall via local, disposable memory.

Show HN: Microcosm Industries – Simulation toys and software microcosms

4 arbesman 2026-07-15 20:32:16 UTC microcosm.industries Prefix Share 0

Microcosm Industries is devoted to fostering simulation toys: software that allows you to play with a complex miniature world. We are inspired by computing history, interdisciplinary science and mathematics, and open-ended play (read more here). If you are building a simulation toy that Microcosm Industries can help with, please reach out. And if you are interested in submitting one that ...

Launch HN: Coasty (YC S26) – An API for computer-use agents

27 nkov47 2026-07-15 16:32:16 UTC coasty.ai Qwen Share 5

Coasty automates tasks via managed VMs, workflows, and primitives, with autonomous runs for single goals and workflows for multi-task programs. API keys control access, billing tracks usage, and BYOK allows custom models.

The Memory Heist

77 eieio 2026-07-15 14:32:16 UTC ayush.digital Qwen Share 9

A researcher found that Claude's memory system could leak sensitive user data, like names and employers, by tricking the AI into navigating a malicious website's links to spell out personal information. Anthropic patched the vulnerability by limiting web navigation, but the exploit underscores risks in AI systems storing vast user profiles.

My midlife crisis Corolla is fast, furious, and modded

135 gmays 2026-07-15 14:32:16 UTC zocalopublicsquare.org Qwen Share 283

The author’s GR Corolla, a modified Japanese sports car, symbolizes Asian American ingenuity and midlife reinvention, blending personal nostalgia with the 1990s import car scene’s cultural legacy. By rejecting Hollywood’s whitewashing, it celebrates a community that created its own identity through underdog car culture, proving that heroes are forged in the garage, not the screen.

Briar is in maintenance mode

130 ristello 2026-07-15 13:32:16 UTC briarproject.org Qwen Share 87

Briar continues in maintenance mode with essential security updates and bugfixes, addressing longstanding issues like high battery use and Android reliability. After initial shutdown plans due to funding and technical challenges, community support and new users led to its continuation, dispelling recent rumors.

Codex Micro

236 davidbarker 2026-07-15 16:32:16 UTC openai.com Qwen Share 200

Agent Key offers live RGB status from Codex, a joystick for workflows like PR reviews/debugging, and command keys for shortcuts. It features a dial for adjusting reasoning levels and uses CNC aluminum, PBT keycaps, and mechanical switches.

Unsolved Problems in MLOps

39 gnyeki 2026-07-15 16:32:16 UTC spawn-queue.acm.org Qwen Share 4

The website's security service blocked your action, possibly due to specific keywords, SQL commands, or malformed data. Contact the site owner with your activity details and the Cloudflare Ray ID for resolution.