2/25/2026
Large-scale online deanonymization with LLMs (using HN posts)
Research on large-scale deanonymization using LLMs on HN posts, explores privacy implications.
2/25/2026
Show HN: Engram – Open-source agent memory that beats Mem0 by 20% on LOCOMO
Engram: open-source TypeScript agent memory system using metadata-rich storage and read-time intelligence, achieving 80% on LOCOMO benchmark vs Mem0/Zep.
2/25/2026
Show HN: Sgai – Goal-driven multi-agent software dev (GOAL.md → working code)
Sgai: multi-agent software development framework that decomposes goals into coordinated agent roles (developer, reviewer, safety analyst) and iterates to working code.
2/25/2026
Ask HN: What would happen if unleashed an AI agent with no limits?
Hypothetical discussion on HackerNews about unconstrained AI agent behavior on a PC.
2/25/2026
PromptFast – Test and compare prompts across different LLMs without any setup
PromptFast: web tool for testing and comparing prompts across 13+ LLM providers without setup.
2/25/2026
GPT Bot Ignoring Robots.txt on my cloudflare worker
Report of GPT Bot crawler ignoring robots.txt on Cloudflare workers, causing unexpected quota usage on private apt mirror.
2/25/2026
Show HN: Oxynote – Technical knowledge base with live Prometheus charts
Knowledge base platform for technical teams with live Prometheus metrics, collaboration, and PR-style reviews. Developer tool with some technical depth.
2/25/2026
LadybugDB: DuckDB for Graphs
LadybugDB: embedded graph database optimized for analytical workloads with full-text search and vector indices.
2/25/2026
Show HN: Polos: Open-source runtime for AI agents with sandbox and durable exec
Open-source runtime for AI agents with sandboxing, durable execution, and workflow management. Production infrastructure for deploying agentic systems.
2/25/2026
I asked Claude for 37,500 random names, and it can't stop saying Marcus
Empirical study finding Claude disproportionately generates name 'Marcus' when asked for random names across 37,500 trials. LLM behavior research with reproducible code.
2/25/2026
Show HN: CodeSeeker – Knowledge graph code intelligence for AI coding assistants
MCP server providing knowledge graph-based code intelligence for Claude, Copilot, and Cursor. Semantic search and graph traversal for AI coding assistants.
2/25/2026
Show HN: Orca, open-source AI agent for deep LinkedIn profile analysis
Open-source AI agent for LinkedIn profile analysis. Scrapes and reasons over post/interaction data to extract structured insights autonomously.
2/25/2026
Our Computer Using agent just solved CAPTCHA up to Level 6
Computer-use AI agent (Coasty) achieves 82% on OSWorld benchmark, handles CAPTCHAs and browser interactions without explicit training.
2/25/2026
Show HN: Dance of Tal V2 – Dependency injection and lockfiles for AI agents
Dependency management and versioning system for AI agent system prompts, addressing modularity and maintainability in agentic AI development.
2/25/2026
Show HN: Nosh – Fast native shell with built-in AI (natural language commands)
Fast native shell replacement with optional built-in AI for natural language commands and agentic mode. Open-source developer tool with LLM integration.
2/25/2026
"A file is an app" – Selfware, a unified file format for the Agent eramory)
Selfware Protocol: unified file format for the agent era, enabling agents as executable files with draft v0.1.0 specification.
2/25/2026
AncestorTree – Open-source genealogy for Vietnamese families
AncestorTree is an open-source genealogy platform for Vietnamese families built with 8 AI agents orchestrated through Claude Code in 24 hours.
2/25/2026
AI Dev Tool Stack for 2026
Overview of 2026 development tools evolved toward AI-driven systems for code understanding, generation, testing, and deployment guidance.
2/25/2026
Show HN: Memograph CLI- A tool to diagnose 'memory failures' in AI agents
Memograph CLI is a debugging tool for AI agents that diagnoses memory failures, forgotten context, and token waste in conversations.
2/25/2026
Show HN: Dola Seed 2.0 – AI video generator with multi-shot narrative control
Dola Seed 2.0 is an AI video generator supporting multi-shot narratives with reference-driven consistency across text, images, video, and audio inputs.
2/25/2026
Show HN: SeeVideo – Access Seedance 2.0 and Kling 3.0 without a subscription
SeeVideo.dance is a web interface providing free access to Kling 3.0 and Soudance 2.0 video generation models without subscriptions.
2/25/2026
Ask HN: What if your LLM violates a patent?
Discussion thread questioning hypothetical patent violation risks in LLMs and proposed defense mechanisms. Speculative, no concrete solutions.
2/25/2026
Ephemeral Sandboxes Break Down for Long-Running AI Agents
Technical analysis of ephemeral sandbox limitations for long-running AI agents, drawing from RisingWave's experience with stateful computation.
2/25/2026
Show HN: TMDD – continuous threat modelling that makes your code more secure
TMDD is a CLI tool that generates continuous threat models in YAML and creates security-aware prompts for AI coding agents.
2/25/2026
Show HN: Stintly – Offline-first app for freelancers with on-device AI
Stintly is an offline-first business management app for freelancers using on-device AI for invoicing, expense tracking, and tax calculations.
2/25/2026
Show HN: An Occam to Go transpiler (LLM-generated)
LLM-generated transpiler converting Occam to Go, leveraging shared CSP concurrency models.
2/25/2026
Why "Garbage In, Garbage Out" is the biggest bottleneck in AI coding
Service generating executable engineering specs from single-sentence product ideas using AI. Targets founders building with AI coding assistants.
2/25/2026
Show HN: Open-Source EU AI Act Scanner for Python AI Projects
Open-source CLI tool scans Python AI projects for EU AI Act compliance, detecting LangChain, CrewAI, OpenAI patterns against Articles 9-15 requirements.
2/25/2026
BrAIn: Persistent, Human-Inspired Memory for LLM Agents
brAIn: persistent memory system for LLM agents inspired by human brain architecture (working, episodic, semantic, procedural memory).
2/25/2026
Show HN: CivBench a long-horizon AI benchmark for multi-agent games
CivBench/ClashAI is an open agent scoreboard benchmarking frontier models playing multi-agent strategy games with observable reasoning in real-time.
2/25/2026
How to Boost Your OpenClaw Bot 10x
Bifrost is an open-source enterprise framework for OpenClaw bots claiming 5,000 req/s throughput with 3.3GB peak memory at P99 latency.
2/25/2026
If AI is mission-critical for your business, build your own feedback loop
Technical recommendation for building custom feedback loops and prompt optimization rather than relying on rigid LLMOps platforms for quality improvement.
2/25/2026
Show HN: MCPSpec – Ship reliable MCP servers without writing test code
MCPSpec: open-source CLI for testing MCP servers with session recording, mock generation, security auditing and CI integration.
2/25/2026
Ask HN: Is RAG an antipattern for AI agents?
Discussion questioning RAG as an antipattern for AI agents and proposing file-based retrieval abstraction instead of vector stores.
2/25/2026
Show HN: Crewly Codes – AI crew that specs, builds, and QAs your features
Multi-agent AI workflow system with PM agent (Morgan) that generates specs, code agents build, QA validates. Orchestrated agents for feature development.
2/25/2026
Ask HN: Have top AI research institutions just given up on the idea of safety?
Question about AI research institutions' commitment to safety. Philosophical inquiry without new data or technical analysis.
2/25/2026
Ask HN: Replacing RAG pipelines with a filesystem interface for AI agents
Filesystem interface replacing RAG boilerplate for AI agents. Virtual directory abstraction for document retrieval without embedding/vector store setup.
2/25/2026
Benchmarking the best base small model for fine-tuning
Title-only submission on benchmarking base models for fine-tuning. Insufficient content for evaluation.
2/25/2026
Show HN: AIP – Open protocol for AI agents to discover and collaborate
Open protocol for agent-to-agent discovery and coordination. Addresses inter-agent communication beyond human-to-agent interfaces.
2/25/2026
Ask HN: What would you want a daily AI portfolio briefing to tell you?
AI-powered portfolio briefing tool that analyzes SEC filings and earnings for personalized daily summaries. LLM application for financial analysis.
2/25/2026
TinyTTS: Ultra-light English TTS (9M params, 20MB), 8x CPU, 67x GPU
Compact 9M parameter text-to-speech model (20MB) optimized for local voice assistants. 0.45s CPU inference, 8x faster than baselines.
2/25/2026
Show HN: Automatic context rotation for Claude Code (no manual steps)
Context rotation system for Claude coding agents to prevent context window overflow. 3-hook pipeline with local dry-run replay capability.
2/25/2026
ActivationKit – AI agent replaces manual tooltip tours with one script tag
ActivationKit: AI-powered user guidance tool replacing manual tooltip tours with single script tag, no configuration needed.
2/25/2026
Show HN: EvidentTrail – Turn GitHub activity into continuous SOC2 audit evidence
GitHub App that converts PR activity, approvals, and CI results into SOC2/ISO audit evidence automatically.
2/25/2026
Current Large Audio Language Models largely transcribe rather than listen
arXiv research discussing limitations of large audio language models, noting they transcribe rather than truly listen to audio.
2/25/2026
One Misconfiguration, One AI Agent, 13 Hours of Downtime
Title-only submission on AI agent causing 13-hour outage due to misconfiguration. Insufficient content for evaluation.
2/25/2026
Launch HN: TeamOut (YC W22) – AI agent for planning company events
TeamOut: AI agent for autonomous company event planning handling venues, vendors, flights, and itineraries via conversation.
2/25/2026
Show HN: Md files as B2B AI agent sandbox – no production data needed
Markdown-based sandbox for B2B AI agents using live browser telemetry instead of production data for pilots.
2/25/2026
Show HN: I cut LLM API bill by 55% with a Python text compressor, no AI involved
Python text compression tool reducing LLM API costs by 55% without using AI.
2/25/2026
Spanish company releases free compressed AI model
Multiverse Computing releases CompactifAI-compressed LLM models to reduce deployment costs for developers.