Category: tech

Tencent's Hunyuan World turns a text prompt into a 3D scene where every object is a separate editable asset, not one fused mesh or splat.

Starting from an open-source base and spending $40M, Thomson Reuters built a legal and tax model matching frontier performance on under 10% of its content.

An open-source Claude Code skill enforcing 10 output rules: lead with the action, number steps, cap lists at five, no preamble and no closing pleasantries.

A free modular menu-bar toolkit fixes clamshell sleep on battery, per-app volume, inverted mouse scrolling, and file cut-and-paste on macOS.

Open-source macOS menu bar app that logs focused-window text via the accessibility API into deduplicated daily markdown files for LLM agents to read.

Colibri, a 1,300-line C engine, runs a 370 GB mixture-of-experts model on 25 GB of RAM by streaming experts off SSD at 0.1 to 1 token per second.

A bookmarklet that captures authenticated pages into Figma as editable frames by inlining Figma's own capture library, which lets it bypass strict CSPs.

A desktop assistant that passively watches the active window and listens in meetings to build running work context, captured locally but processed in the cloud.

A local, open-source memory layer for coding agents that stores conversations verbatim while indexing them with a compact symbolic notation for fast recall.

Munder Difflin wraps your agent CLI into a free, local, always-on clone that messages teammates' clones to hand off work and share an org knowledge base.

A TikTok creator claims a new open-weight Qwen model matches Opus-level intelligence and already has an uncensored fork, raising AI safety questions.

The Arch and Hyprland desktop unifies bar, launcher, and notifications into one themable layer, and adds zero-touch provisioning plus offline voice dictation.

Self-hosted platform where each agent gets its own container, browser, and workspace, and every action passes a policy gateway before it happens.

A promotional TikTok pitches an MCP server indexing 80,000+ AI skills, so agents search and load only the relevant skill instead of preloading everything int...

The tode CLI pipes code-server through a terminal-based browser renderer, giving a full VS Code UI over SSH with no local GUI or forwarded display.

The MCP core team's new roadmap covers agentic messaging, transport unification, agent identity, tool-result contracts, and SDK developer experience.

Cost math on self-hosting a frontier LLM: a $112,000 rig can crawl at 0.1 tokens per second because memory bandwidth, not price, is the real bottleneck.

ChatGPT for Mac adds an Apple Messages plugin that reads, searches, and sends iMessage, SMS, and RCS locally, gated by per-message approval by default.

DeepSeek's vision model resizes every image toward 800x800 before inference, so image cost is a flat 384 tokens regardless of original resolution.

A hyped TikTok promo for an unverifiable coding tool still points to a real pattern: split planning, building, and review across separate models.