3/29/2026
I built a better, human like memory, for Agents
elfmem: adaptive memory system for LLM agents with decay and relevance tracking. Single-file, no external infrastructure required.
3/29/2026
Show HN: Think Fu – Metacognition as a service
ThinkFu is a service addressing LLM creative limitations by reducing RLHF constraints to improve creative reasoning capabilities beyond predictable outputs.
3/29/2026
Local-first AI knowledge layer. Extract architecture, query from any AI tools
TaskWing extracts code architecture into local SQLite knowledge base for AI tools, enabling offline sessions without cloud dependency or account requirements.
3/29/2026
Are LLMs a Dead End? [video]
Multi-agent orchestration platform managing concurrent LLM agents in containers with isolated identities, credentials, and workspaces.
3/29/2026
DaVinci-MagiHuman: Open-source AI model for realistic video generation
Open-source DaVinci-MagiHuman model generates realistic human videos with improved lip-sync and audio alignment. Video generation research.
3/29/2026
Vibe physics: The AI grad student
Harvard physicist supervises Claude through theoretical physics research calculation without touching code. AI agents for scientific research.
3/29/2026
Tried a New AI Image Tool for Real-World Design Work (Nano Banana Pro)
Review of Nano Banana Pro image generation tool for design work. Commercial tool application.
3/29/2026
ARC-AGI solver via Plücker geometry – 316 tasks solved with zero learning
OpenClaw skill enabling GPU machines to run parallel autonomous ML research agents with nightly experiments and knowledge synthesis.
3/29/2026
LLMs Can Write Production Quality Code
Developer at Airbnb writes 99% production code using LLMs across 1000+ microservices. Practical LLM application at scale.
3/29/2026
What if AI doesn't need more RAM but better math?
Google's TurboQuant compression algorithm reduces LLM memory usage by 6x without quality loss. ML optimization research.
3/29/2026
Show HN: Liter-LLM, Universal LLM client in Rust with bindings for 11 languages
Liter-LLM: universal LLM client in Rust with 11 language bindings and 142 provider support. Open-source developer tool.
3/29/2026
Show HN: Built tool to turn n8n workflows into OpenClaw-compatible agent skills
CLI tool converting n8n workflows into OpenClaw-compatible agent skills using LLM for transpilation.
3/29/2026
Google's TurboQuant AI-compression algorithm can reduce LLM memory usage by 6x
TurboQuant compression algorithm reduces LLM memory 6x while maintaining quality and improving speed. ML optimization.
3/29/2026
Show HN: Shoofly – pre-execution security for Claude Code Cowork and OpenClaw
Shoofly: pre-execution security layer for Claude Code and OpenAI agents, intercepts tool calls to block injection and malware. AI agent security.
3/29/2026
Mr. Chatterbox is an LLM trained exclusively on Victorian-era British texts
LLM trained exclusively on Victorian-era British texts. Specialized language model with narrow training data.
3/29/2026
AI Hot Takes from a Platform Engineer / SRE
Platform engineer critiques AI hype, noting lack of real-world production agentic systems that perform reliably.
3/29/2026
I Built CLI for Bullmq
CLI dashboard for BullMQ job queue management. Developer tool not AI-related.
3/29/2026
Why question-space can't be baked into LLM weights (preprint)
Preprint: LLMs have structural attention-based convergence limiting open-ended exploration. Proposes Knowledge Innovation System v2.0.
3/29/2026
OpenYak – An open-source Cowork that runs any model and owns your filesystem
Open-source desktop AI agent running locally, manages files, analyzes data, automates office workflows without cloud upload.
3/29/2026
Eval-Driven Development: Applying TDD Principles to AI Agent Prompts
Framework: Apply test-driven development principles to AI agent prompt engineering. Eval-driven development methodology.
3/29/2026
Show HN: Phantom – Let AI use your API keys without leaking them
Tool: Phantom lets AI models use real API keys safely without exposing credentials to LLM context windows or prompt injection.
3/29/2026
Wikipedia officially bans AI-generated content
Wikipedia bans AI-generated content from its 260k editors, citing accuracy and sourcing concerns over LLM output.
3/29/2026
Show HN: Local-first dictation for macOS – Parakeet TDT, zero cloud calls
Local-first voice-to-text tool for macOS. Processes audio entirely on-device, no cloud calls. Free, open offline.
3/29/2026
Show HN: Anamnesis – Open-source 4D strategic memory engine for AI agents
Open-source memory engine for AI agents storing strategically-weighted, persistent context across sessions with reasoning fields.
3/29/2026
Ask HN: How are you keeping AI coding agents from burning money?
Discussion on cost control strategies for AI coding agents including token tracking and per-call logging layers.
3/28/2026
Missing layer between LLMs and engineering decisions: evidence-graded reasoning
Tool for evidence-graded technical decision-making using AI, integrates with Claude and CI/CD workflows.
3/28/2026
Bluesky leans into AI with Attie, an app for building custom feeds
Bluesky launches Attie, an AI assistant for building custom feed algorithms and vibe-coding apps.
3/28/2026
Why AI Companies Want to Take Control of Your Computer
Overview of AI agent computer use capabilities, examining LLM access to desktop automation tools like Anthropic's Claude implementation.
3/28/2026
From 300KB to 69KB per Token: How LLM Architectures Solve the KV Cache Problem
Technical analysis of KV cache optimization in LLM architectures, examining memory constraints and solutions for token processing efficiency.
3/28/2026
Show HN: EnterpriseFizzBuzz – 622K lines of production-grade FizzBuzz
Satirical GitHub project demonstrating over-engineering via production-grade FizzBuzz implementation with excessive architecture.
3/28/2026
MCP Toolbox for Databases
Open source MCP server enabling AI agents and applications to connect directly to enterprise databases with dual-purpose design.
3/28/2026
Is AI agent discoverability a reputation graph problem, not an SEO problem?
Technical analysis proposing reputation graphs as alternative to SEO for AI agent discovery and trustworthiness evaluation.
3/28/2026
Verify_before – pre-deployment verifiability analysis for neuro-symbolic AI
Research framework for verifying neuro-symbolic AI systems pre-deployment, deriving mathematical threshold for verifiability of composed AI architectures.
3/28/2026
Show HN: Glazyr Viz – Zero-Copy MCP Vision Server via POSIX Shared Memory
Zero-copy MCP vision server using POSIX shared memory for high-performance agent perception, achieving 7.35ms latency without base64 serialization.
3/28/2026
Entroly Context engine- AI coding tools 100% codebase visibility
Entroly Context Engine reduces tokens needed for AI coding agents by showing full codebase context instead of 5%. Open-source tool integrating with Cursor, Claude, Copilot.
3/28/2026
Safari MCP: 80-tool native browser automation for AI agents (macOS)
Native Safari MCP server with 80 browser automation tools for AI agents on macOS, using AppleScript/JavaScript with 60% lower CPU than Chrome.
3/28/2026
Show HN: AI Cost Firewall – OpenAI-compatible gateway with semantic caching
AI Cost Firewall: OpenAI-compatible API gateway reducing LLM costs via exact-match and semantic similarity caching between applications and providers.
3/28/2026
Ask HN: "Claws" and Human-in-the-Loop Safety
Sandclaw: open source human-in-the-loop safety framework for AI agents with progressive firewall relaxation and default-deny write path access.
3/28/2026
Open-source CVE scanner for AI-generated code
CVE Guard: open source offline scanner detecting vulnerabilities in AI-generated code before commit, identifies 1.7x higher vulnerability rates in AI code.
3/28/2026
Data Is the Final Moat
Analysis of data as competitive advantage as LLM commoditization increases, discussing open source agent frameworks and pricing dynamics.
3/28/2026
Show HN: Vyasa – A client-side AI writing detector (WASM, no API calls)
Vyasa: client-side WASM-based AI-generated text detector using signatures from Wikipedia, runs offline without API calls.
3/28/2026
Building AI Agents: The Fundamentals
Title-only stub about AI agent fundamentals.
3/28/2026
Who Wins If AI Models Commoditize? – With Mistral CEO Arthur Mensch [video]
Video interview with Mistral CEO Arthur Mensch discussing market implications of AI model commoditization.
3/28/2026
Argus-LLM – open-source LLM output evaluation across 6 dimensions
Argus-LLM production observability framework evaluating LLM outputs across six dimensions: groundedness, accuracy, reliability, variance, cost, safety.
3/28/2026
Open-source GEO audit – test your brand visibility across 7 AI engines
XanLens open-source tool that queries 7 AI engines (ChatGPT, Gemini, Grok, DeepSeek, Claude, Llama, Qwen) to audit brand visibility and compare responses.
3/28/2026
When the bill comes due: the economics of AI coding tools
Analysis of AI coding tool economics: subsidy withdrawal risks and long-term viability as cost curves decline.
3/28/2026
Further human + AI + proof assistant work on Knuth's "Claude Cycles" problem
LLMs and proof assistants collaborated to solve Knuth's Claude Cycles problem. Links to ChatGPT conversation and academic work.
3/28/2026
Show HN: I built an OS that is pure AI
Pneuma: AI-native desktop OS where applications are generated on-demand via natural language prompts. Agents persist, communicate via IPC, and share through community store.
3/28/2026
Show HN: Nanopm – PM automation for Claude Code (audit → strategy → roadmap)
Nanopm automates product management tasks (audit, strategy, roadmap) for Claude Code using skill-based pipeline architecture with persistent memory.
3/28/2026
Show HN: GitHub Copilot Technical Writing Skill
GitHub Copilot skill for technical writing that reviews documentation using Google's technical writing principles. Available via /tech-writer command.