12d ago
Tutorial: Algebraic Foundations Powering FlashAttention
Technical tutorial on algebraic foundations underlying FlashAttention algorithm for efficient transformer computation.
12d ago
Show HN: Ranking 19 LLMs on Flutter code by compile pass and hidden-test pass 1
Benchmark comparing 19 LLMs on Flutter code generation, measuring compile pass rates and hidden-test pass rates.
12d ago
Show HN: Production-grade LangGraph template
Production-grade LangGraph multi-agent template with FastAPI, Helm, Terraform, CI, and observability. Deployable agent framework for ML engineers.
12d ago
Show HN: Fast NF4 dequantization Triton kernel (1.41x faster than bitsandbytes)
Custom Triton kernel optimization for NF4 dequantization in 4-bit LLM inference, achieving 1.41x speedup over bitsandbytes with code and benchmarks.
12d ago
Show HN: Aict – Unix coreutils that output XML/JSON, built for AI agents
Unix coreutils reimplemented to output XML/JSON format, designed specifically for AI agent tool integration.
12d ago
Show HN: Goku – WASM (wllama)-powered LLM inference and model manager
WASM-based LLM inference engine using wllama with integrated model manager for local deployment.
12d ago
Grok Build CLI uploads full Git history and .env secret to xAI's cloud
Security report: Grok Build CLI uploads full Git history and .env secrets to xAI cloud without user awareness. Privacy/security concern.
12d ago
Show HN: Agent's Design – Claude/Codex copy-paste templates that kill AI-slop UI
Copy-paste UI design templates for Claude/Codex integration to reduce generic AI-generated interfaces.
12d ago
Show HN: Grepathy – Claude made a decision nobody approved
Open-source tool preserving Claude agent decision transcripts locally to prevent loss of conversation history.
12d ago
Show HN: Nable – Local-first cost intelligence tool for cloud and AI
Open-source MCP tool for cloud/AI cost optimization with bill normalization and savings verification across providers.
12d ago
Im building PixelKit: A Linux alternative for Microsoft Powertoys
PixelKit: Linux native alternative to Microsoft PowerToys written in Rust, supporting X11/Wayland. Open source developer tool.
12d ago
Mosaic Bridge, Unity Editor MCP with 90 research-paper algorithms
MCP tool for Unity Editor integrating 90 research paper algorithms for game development AI.
12d ago
Verse: Build and hire autonomous AI employees from a single prompt
Platform for building and deploying autonomous AI agents from natural language prompts.
12d ago
Free local AI video clipper – No Upload, All in browser
Browser-based AI video editing tool that processes locally without cloud uploads.
12d ago
Show HN: An AUI for Nix-Darwin
NixMC: macOS native app using Claude Code to manage nix-darwin configuration via natural language. AI agent for system administration.
12d ago
Show HN: A TypeScript repo where AI agents can't break the architecture
TypeScript repository demonstrating architecture patterns that constrain AI agent behavior.
12d ago
If you want Claude to speak nicely to you, try Hindi or Arabic
Anthropic research mapping how Claude exhibits different values across languages, identifying four key behavioral axes.
12d ago
Claude Managed Agents add per-session overrides and lifecycle webhooks, Gemini
Daily AI news update: Claude Managed Agents lifecycle improvements, Gemini 3.5 Flash GA release, Genkit Python support.
12d ago
Html2realpdf creates selectable, searchable PDFs on the web instead of PNGs
Html2realpdf: Browser-based PDF generator written in Zig/WebAssembly, produces selectable vector text instead of screenshots. Developer tool with TypeScript API.
12d ago
Show HN: PortalJS – AI-native, open-source framework for data portals
Open-source AI-native framework for building data portals with agent-assisted scaffolding and Next.js architecture.
12d ago
Show HN: An uptime monitor with an MCP server, create monitors by asking Claude
Uptime monitoring service with MCP server enabling Claude/ChatGPT to create and manage monitors via natural language prompts, curl, or GitHub Actions.
12d ago
GPT-5.6-terra used 48.5% more context than Mimo-2.5-pro
User benchmark comparing GPT-5.6-terra vs Mimo-2.5-pro LLM agents on token usage and task efficiency. Real-world performance data from agent task history.
12d ago
Build an AI Price Quote Phone Agent Real-Time Custom Quotes with Telnyx Voice AI
Tutorial building voice-based AI agent for phone quotes using Telnyx Voice AI API.
12d ago
I Configure the Ryzen AI Halo (Strix Halo) for 10-15% Faster Local Inference [video]
Tutorial optimizing Ryzen AI Halo hardware for 10-15% faster local LLM inference performance.
12d ago
Aliens: Explaining LLMs and AI agents to my children
Explainer comparing LLMs and AI agents to aliens. Educational but low technical depth.
12d ago
Show HN: LoopGain – Stop agent loops with control theory, not max_iterations
LoopGain tool uses control theory to prevent AI agent loops as alternative to max_iterations parameter.
12d ago
Show HN: AI skills appear in 0.3% of entry-level data jobs, 3.6% of senior
Analysis of AI skill adoption in tech job market, tracking mentions of AI/ML skills by seniority level and job category.
12d ago
TermCanvas – a tmux-backed canvas for steering AI agents (macOS)
TermCanvas tool provides tmux-based interface for steering and controlling AI agents on macOS.
12d ago
Wrap an LLM, charge $20, lose money: the pricing trap the model owners hit first
Economic analysis of LLM pricing models, examining why flat-rate pricing (GitHub Copilot $10/month) became unsustainable and how model owners shifted to usage-based pricing.
12d ago
I reverse-engineered the three biggest agent-memory tools
Reverse-engineered analysis of three agent-memory tools (Cognee, Graphiti, Neo4j) revealing convergent knowledge-graph architectures and discussing limitations of existing implementations.
12d ago
DSLs Enable Reliable Use of LLMs
Domain-Specific Languages enable reliable LLM code generation through clear boundaries and abstractions. Tickloom example with distributed systems DSL.
12d ago
ConlangCrafter: Constructing Languages with a Multi-Hop LLM Pipeline
ConlangCrafter: LLM multi-stage pipeline generating complete artificial languages with phonology, grammar, lexicon. Open dataset and research project.
12d ago
Show HN: GWZ – Git Workspace Zone (multi-repo that feels like plain Git)
Python tool for managing multiple Git repositories as a unified workspace, addressing limitations of Git submodules.
12d ago
AI Agents: Hype vs. Reality (2024)
Analysis of AI agent capabilities vs. hype, citing WebArena benchmarks showing 45.7% success rates and definitional ambiguity in the field.
12d ago
Ptolemaic and LLM Architecture Compared (In 3D)
Interactive 3D visualization comparing Ptolemaic cosmology to LLM architecture layers, illustrating forward/backward passes.
12d ago
vLLM prefill paired with TileRT decode
Technical analysis of disaggregated LLM serving with vLLM prefill and TileRT decode, separating compute-bound and memory-bound phases.
12d ago
Cicada- an agentic Python IDE Free to use ( comes with built in small model)
Agentic Python IDE with built-in small LLM, emphasizing error recovery and self-healing pipelines.
12d ago
Show HN: I built an AI agent memory engine because Obsidian wasn't cutting it
AI agent memory engine designed to replace Obsidian, achieving perfect recall across thousands of facts in 8MB, runs offline.
12d ago
TPU and GPU Clusters: The Anatomy of Collective Communication
Deep technical dive into TPU/GPU cluster topologies and collective communication patterns for transformer training and inference at scale.
12d ago
Show HN: Vehir – a platform built for AI agents: compiler, microkernel, CAS
Vehir is an experimental AI-native computing platform designed with agent-computer interaction at its core, featuring a compiler, microkernel, and CAS built for structured I/O for AI agents.
12d ago
I scanned 10 MCP servers – here's what agents can't know before connecting
Security audit tool for Model Context Protocol servers, providing trust cards for agent-server connections before deployment.
12d ago
Opensourcing Multiplayer AI in Discord
Open-source multiplayer AI development tool for Discord that enables collaborative coding with shared shells, live previews, and chat-native workflows in self-hosted environments.
12d ago
Open-sourced a dataset of companies and their detected tech stacks (CSV/JSON)
Open-source dataset of 58.9M+ companies and their technology stacks in CSV/JSON format, updated daily.
12d ago
Legal AI, not a coding agent with scaffolding
Framework for legal AI systems emphasizing agent workspaces, audit trails, and domain-specific design versus generic coding agent scaffolding.
12d ago
Why organizing knowledge in the age of AI sucks, and how I solve it
Personal exploration of knowledge organization challenges when using AI chatbots and proposal for improved systems to manage AI conversations.
12d ago
Maincode launches Matilda, an AI assistant running on Australian infrastructure
Maincode launches Matilda, an Australian-built AI assistant running on local infrastructure with beta access.
13d ago
Ask HN: How useful is Gemini for Google Apps?
User reports limitations of Gemini AI integration in Google Workspace apps, unable to create new documents across Google Drive.
13d ago
LeMario: Training a JEPA World Model on Super Mario Bros
Reproduction and extension of JEPA world model architecture trained on Super Mario Bros from pixels and actions for reward-free planning.
13d ago
OpenAI's first hardware device will be a portable desktop robot
OpenAI developing portable voice-controlled smart speaker robot with personality, no screen, learns from user interactions and email.
13d ago
Same model, same Q4_K_M label: 5.02, 5.07 and 5.27 bits per weight
Picchio measurement tool analyzes local LLM quantization accuracy and GPU performance across prefill, decode, and wallclock metrics.