Show HN Daily Selection (2025-08-20)
- OpenAI/reflect – Physical AI Assistant that illuminates your life
- Twick - React SDK for Timeline-Based Video Editing
- Python file streaming 237MB/s on $8/M droplet in 507 lines of stdlib
- Wake word detection with custom phrases without model training
- Lemonade: Run LLMs Locally with GPU and NPU Acceleration
- Built a memory layer that stops AI agents from forgetting everything
- Sam Altman:"Scaling LLMs won't get us to AGI" – maybe we found a path
- GiralNet – A Privacy Network for Your Team (Not the World)
- AI-powered CLI that translates natural language to FFmpeg
- OnPair – String compression with fast random access (Rust, C++)
- I built a simple restaurant bill splitting app with receipt reading
- Skilfut – 138 UI components to help devs build faster and prettier
- I'm building a "work visa" API for AI agents
- Hacker News London Meetup #4
- Bells and Whistles – a new type of daily logic puzzle
- Memeclip.ai – AI-powered meme maker that turns text into memes
- Twitter Condom – Control your Twitter feed
- Side Space – An Arc-like AI-powered Vertical tabs manager for Chrome
OpenAI/reflect – Physical AI Assistant that illuminates your life [Life Assistant]
36 points, 19 comments
A physical AI assistant that uses light to reflect mood and guide daily activities.
- Extendable/hackable code for easy modification
- Communicates via light changes based on mood or feelings
- Provides guidance by populating info from controlled clients
- Uses phone as a control hub, avoiding WiFi setup and keeping sensitive data off the device
Discussion Link, Original Link
Twick - React SDK for Timeline-Based Video Editing [Developer Tools]
5 points, 1 comments
React SDK for timeline-based video editing.
- Enables timeline-based video editing in React applications
- Provides an SDK for developers to integrate video editing features
- Focuses on a React-compatible solution for video editing workflows
Discussion Link, Original Link
Python file streaming 237MB/s on $8/M droplet in 507 lines of stdlib [Developer Tools]
17 points, 12 comments
A lightweight Python WSGI framework achieving high-speed file streaming with minimal memory usage.
- Pure Python stdlib implementation with no external dependencies
- Supports HTTP range for partial content delivery
- Generator-based streaming with constant memory usage (<225MB)
- Request batching via query parameters
- Efficient routing using match statements
- Built-in sanitization and structured logging
- Benchmarked at 237MB/s on $8/month hardware
Discussion Link, Original Link
Wake word detection with custom phrases without model training [Audio]
3 points, 0 comments
A wake word detection system that supports custom phrases without requiring model training, combining Google's speech-embedding with DTW for improved accuracy and efficiency.
- Uses Google's speech-embedding for noise-resilient feature extraction
- Dynamic Time Warping (DTW) for temporal alignment
- Achieves 98.6% same-speaker and 81.9% cross-speaker accuracy
- Optimized for Raspberry Pi with ONNX conversion (10% CPU usage)
- Open-source and customizable for user-defined wake words
Discussion Link, Original Link
Lemonade: Run LLMs Locally with GPU and NPU Acceleration [Developer Tools]
12 points, 0 comments
Lemonade is an open-source SDK and local LLM server designed to simplify running and experimenting with large language models on personal PCs, with optimized support for NPUs and GPUs.
- Provides an OpenAI-compatible API for seamless integration with existing tools
- Optimized performance using llama.cpp (Vulkan/ROCm) and ONNXRuntime (Ryzen AI)
- Cross-platform support with one-click Windows installer and Linux pip/source installation
- Supports GGUF and ONNX models (Gemma, Llama, Qwen, Phi, etc.)
- Includes Python API and CLI for model management and benchmarking
- Fully open-source (Apache 2.0) with community collaboration
- Privacy-focused, enabling local execution without cloud dependencies
Discussion Link, Original Link
Built a memory layer that stops AI agents from forgetting everything [Developer Tools]
4 points, 0 comments
In Memoria is a local MCP server that provides AI coding tools with persistent memory to retain context across sessions.
- Stores coding patterns, architectural decisions, and context for AI tools like Claude and Copilot
- Simple setup with
npx in-memoria server—no accounts or external data sharing - Built with TypeScript + Rust, using tree-sitter for parsing and vector storage for semantic search
- Supports JavaScript/TypeScript, Python, and Rust, with plans to expand language support
- Runs locally, ensuring privacy and control over data
Discussion Link, Original Link
Sam Altman:"Scaling LLMs won't get us to AGI" – maybe we found a path [Research]
4 points, 1 comments
AGI Grid proposes a collective, civilizational approach to achieving AGI through open, multi-agent ecosystems rather than scaling individual models.
- Inspired by human intelligence's evolution through collaboration and culture
- Focuses on building diverse AI networks, institutions, and shared participation
- Aims to evolve new knowledge and thinking beyond text prediction
- Includes 12 open-source projects as part of the AGI Grid ecosystem
- Seeks community feedback, critique, and collaboration
Discussion Link, Original Link
GiralNet – A Privacy Network for Your Team (Not the World) [Business]
6 points, 2 comments
GiralNet is a private, team-oriented network inspired by Tor, designed for small groups needing privacy with trusted, verifiable nodes.
- Privacy-focused SOCKS5 proxy for small teams
- Routes traffic through encrypted layers like Tor, but with known nodes
- Nodes are controlled and accountable within the group
- Avoids reliance on large, anonymous public networks
- Provides secure, private communication without exposing data to strangers
Discussion Link, Original Link
AI-powered CLI that translates natural language to FFmpeg [Developer Tools]
3 points, 0 comments
AI-powered CLI that translates natural language into FFmpeg commands for video processing.
- Safety first: Preview every command before execution
- Smart defaults: Sensible codec and quality settings
- Context aware: Scans your directory for input files
- Interactive mode: Iterate on commands naturally
- Well-tested: 87%+ test coverage with comprehensive error handling
- Supports video format conversion, resizing, trimming, thumbnails, and more
OnPair – String compression with fast random access (Rust, C++) [Developer Tools]
3 points, 0 comments
OnPair is a string compression algorithm optimized for fast random access in large collections, balancing compression ratio and speed.
- Compression ratio comparable to BPE
- High compression speed (100–200 MiB/s)
- Fast decompression speed (6–7 GiB/s)
- Designed for low-latency point queries in databases
- Available in Rust and C++ implementations
Discussion Link, Original Link
I built a simple restaurant bill splitting app with receipt reading [Life Assistant]
4 points, 1 comments
A simple app for splitting restaurant bills with friends by reading receipts.
- Calculates bill splits automatically
- Eliminates disputes over splitting methods
- No-frills, straightforward design
- Supports receipt reading for easy input
Discussion Link, Original Link
Skilfut – 138 UI components to help devs build faster and prettier [Design]
4 points, 2 comments
Skilfut is a SaaS offering 138 UI components with associated prompts to help developers and no-code users build faster and more visually distinct designs.
- Provides 138 ready-to-use UI components with prompts
- New components added weekly (200+ planned)
- Designed in collaboration with professionals from companies like Uber
- Aims to speed up development for vibe coders and indie hackers
- Soft-launch attracted 100+ waitlist signups in 2 days
I'm building a "work visa" API for AI agents [Developer Tools]
6 points, 2 comments
AgentVisa is an API that issues secure, short-lived credentials for AI agents to perform specific tasks, replacing static API keys with temporary "work visas."
- Replaces permanent API keys with temporary, task-specific credentials
- Links agent actions to specific users and permissions
- Uses JWTs under the hood for secure authentication
- Includes a Python SDK and demo for testing
- Aims to improve agent identity and auditability
- Free tier available for immediate use
Discussion Link, Original Link
Hacker News London Meetup #4 [Entertainment]
6 points, 0 comments
A community-organized meetup for Hacker News readers in London to discuss technology, science, and business.
- Event: Hacker News London Meetup #4
- Date & Time: August 26, 2025, 6pm - 9pm
- Location: The Masque Haunt, Old Street, London
- Attendees: 64+ (via Meetup)
- Links: Meetup, Lu.ma
- Unofficial, community-driven event
Bells and Whistles – a new type of daily logic puzzle [Entertainment]
3 points, 0 comments
A new daily logic puzzle variant of nonograms called "kanogram" featuring two icons (bells and whistles) per cell with row and column clues.
- Daily puzzles that are always logically solvable
- Features two possible icons (bells and whistles) per cell
- Includes a 'mini' mode for quicker, smaller puzzles
- Named "kanogram" as a new nonogram variant
Discussion Link, Original Link
Memeclip.ai – AI-powered meme maker that turns text into memes [Image]
3 points, 2 comments
AI-powered meme maker that automatically generates memes from text descriptions.
- Transforms any text into viral memes using AI
- Uses semantic template matching to find perfect meme templates
- Integrates vision-language models for context-aware caption generation
- Automates meme creation by generating both template and caption
- Focuses on understanding meme culture and situational context
Discussion Link, Original Link
Twitter Condom – Control your Twitter feed [Productivity]
3 points, 0 comments
Twitter Condom helps users control their Twitter feed.
- Customizable feed filters
- Blocks unwanted content
- Enhances user experience by reducing noise
Discussion Link, Original Link
Side Space – An Arc-like AI-powered Vertical tabs manager for Chrome [Productivity]
3 points, 0 comments
Side Space is an AI-powered Chrome extension for organizing tabs in a vertical sidebar with smart grouping and syncing.
- Vertical Spaces for categorizing tabs (work, shopping, etc.)
- AI-powered tab grouping to reduce clutter
- Cloud sync across devices
- Tab management tools (pin, search, suspend, deduplicate)
- Customizable themes (light/dark mode, color palettes)
- Autosave & restore tabs
- Free plan (5 spaces, 1,000 URLs) and one-time paid upgrade
- Compatible with Chrome, Brave, and Edge