3/24/2026
Elo Memory – Bio-inspired episodic memory for AI agents (free, open source)
Elo Memory: open-source episodic memory system for AI agents inspired by biological memory. Free research implementation for agent architecture.
3/24/2026
LLM drift-Claude vs. Calmkeep: 25-turn Code (60% vs. 85%) & Legal (50% vs. 100%)
Evaluation comparing Claude and Calmkeep LLM performance on code and legal tasks across 25-turn conversations. Benchmarking study with transcript analysis.
3/24/2026
We built observability layer for OpenClaw and AI coding agents
Observability layer built for OpenClaw AI coding agents to improve monitoring and debugging.
3/24/2026
When lying is the best strategy for AI [video]
Video exploring scenarios where deception emerges as optimal strategy in AI systems. Game-theoretic analysis of AI agent behavior.
3/24/2026
Train the smallest LM you can that fits in 16MB. Best model wins
Challenge to train smallest language model fitting in 16MB. Minimal details provided.
3/24/2026
Show HN: Named failure modes that stop AI agents from cutting corners
Framework documenting specific failure modes in AI agent behavior to prevent corner-cutting. Agent safety and failure analysis.
3/24/2026
LocalRouter: MCP via LLM
LocalRouter: implements Model Context Protocol routing through LLM. Tool integration layer for AI agents and LLM applications.
3/24/2026
Show HN: AI agent got 237 rules from another agent, still made the same mistakes
Case study showing AI agent receiving 237 rules from another agent but still making identical mistakes. Agent learning and constraint enforcement analysis.
3/24/2026
We scanned 900 MCP configs on GitHub. 75% had security problems
Security audit of 900+ MCP (Model Context Protocol) configurations on GitHub found 75% have security issues. Research-backed security analysis.
3/24/2026
Show HN: Crawdad – Runtime security layer for autonomous AI agents
Crawdad: runtime security API for autonomous AI agents addressing prompt injection, data exfiltration, and access control. Framework-agnostic security tool.
3/24/2026
Show HN: Drop – Linux sandboxing for LLM agents and untrusted code
Linux sandboxing tool for executing LLM agents and untrusted code safely. Preserves local environment while isolating programs.
3/24/2026
LiteLLM PyPI has been compromised an hour ago, do not update
Security alert: LiteLLM PyPI packages compromised with malicious code stealing credentials and targeting Kubernetes clusters.
3/24/2026
A Tom-Inspired Agenda for AI Safety Research
Research on Theory of Mind in AI models to address agent ecosystem fragility, manipulation risks, and reward misspecification.
3/24/2026
Show HN: Running AI agents across environments needs a proper solution
Discussion on building AI agent systems with tools, memory, and fine-grained capabilities. Argues current systems aren't ready for true agency across environments.
3/24/2026
Show HN: Claude's Code – tracking the 19M+ commits generated by Claude on GitHub
Dashboard tracking 19M+ commits generated by Claude Code on GitHub with statistics about AI-assisted code generation.
3/24/2026
Turn any REST API into an MCP server. No code required
APIFold converts OpenAPI/Swagger specs into production MCP servers enabling AI agents to call REST APIs without code.
3/24/2026
Zbot – an embedded AI agent running on Zephyr RTOS
ZBot is an open-source embedded AI agent running on Zephyr RTOS. Implements ReAct loop, connects to OpenAI-compatible LLMs, controls hardware, maintains memory across reboots.
3/24/2026
Show HN: Danube – AI Tools Marketplace
Danube is a marketplace for AI agents to discover and execute tools securely. Developers can publish tools, and agents access them via MCP without seeing API keys.
3/24/2026
Show HN: Kanna – OSS Web UI for Claude Code and Codex with Embedded Terminals
Open source web UI for Claude and Copilot with embedded terminals, one-click LLM switching, running on localhost.
3/24/2026
Show HN: Using Claude Code history to predict your next messages while you sleep
Overnight is an open-source CLI tool that runs Claude Code autonomously by reading conversation history and predicting next steps. Enables 24/7 execution.
3/24/2026
Show HN: AgentContract – behavioral contracts for AI agents
AgentContract defines behavioral contracts for AI agents, declaring must/must-not/can-do actions and enforcing them. Enables control and predictability for enterprise deployment.
3/24/2026
I built Rubric, an open source Sentry for AI. Looking for beta testers
Rubric is an open-source LLM monitoring tool that logs API calls, scores output quality, and alerts on drift. Supports multiple providers and frameworks.
3/24/2026
Helping developers build safer AI experiences for teens
OpenAI releases open-weight safety model gpt-oss-safeguard and prompt-based policies for teen-safe AI applications.
3/24/2026
Open-Source AI Text-to-Speech Models You Can Run Locally for Natural Voice
Overview of open-source text-to-speech models deployable locally. Compares quality, cost, and control versus cloud APIs.
3/24/2026
AI Agents Can Autonomously Perform Experimental High Energy Physics
Research on using AI agents to autonomously perform high energy physics experiments. Demonstrates autonomous scientific research capabilities.
3/24/2026
LLM Neuroanatomy II: Modern LLM Hacking and Hints of a Universal Language?
Technical research on LLM internals using layer duplication and probing methods. Discovered RYS method achieving top HuggingFace leaderboard ranking without training.
3/24/2026
Yann LeCun's LeWorldModel: Stable End-to-End JEPA from Pixels
LeWorldModel: Joint Embedding Predictive Architecture for stable end-to-end world model learning from raw pixels without auxiliary supervision.
3/24/2026
Review AI agent changes locally before you push
Tool for reviewing AI coding agent changes locally before pushing, solving code review bottleneck with PR-like workflows.
3/24/2026
We almost bought an automation platform. Cowork was one
Case study of evaluating Cowork automation platform alongside Claude Enterprise for employee productivity integration with business tools.
3/24/2026
Powering product discovery in ChatGPT
ChatGPT shopping assistant using Agentic Commerce Protocol for product discovery and comparison.
3/24/2026
Modular 26.2: Image Generation and Upgraded AI Coding with Mojo
Modular 26.2 release adds image generation/editing workflows and improves Mojo for GPU kernel AI development.
3/24/2026
Show HN: ProofShot – Give AI coding agents eyes to verify the UI they build
ProofShot: CLI tool enabling AI coding agents to view browser output, capture screenshots/video/errors for verification during UI development.
3/24/2026
Show HN: Kern – One agent. One folder. One mind. Every channel
Kern: AI agent framework maintaining single continuous session across CLI, Telegram, Slack with persistent local memory in plain folder.
3/24/2026
Show HN: Prompts Directory for Data Analyst
Prompt repository with curated templates for data analysts covering exploration, cleaning, SQL, dashboards, and visualization.
3/24/2026
Long-Running Sandbox in Dockers for Coding Agents (100% POSIX Compatible)
Sandock: Docker-based sandbox for running AI coding agents with persistent volumes, POSIX compatibility, cost-effective alternative to VMs.
3/24/2026
Ask HN: System Programming as a LLM shelter
Discussion on whether system programming offers refuge from LLM-assisted development trends due to different performance priorities.
3/24/2026
Ask HN: HSL 0.1 – The Human Source License. Please help refining
Discussion of Human Source License for open source stream processing library built on PyDBSP.
3/24/2026
Expert Personas Improve LLM Alignment but Damage Accuracy
Research showing expert personas improve LLM alignment but reduce accuracy, trade-offs in prompt engineering.
3/24/2026
The 53-Year Evolution of AI Agents: A Comprehensive Reading List
Comprehensive reading list tracing 53-year research lineage of AI agents from 1970s theoretical foundations through modern implementations.
3/24/2026
Show HN: Shard-based scheduling for 100x more fine-tuning experiments on 4 GPUs
Shard-based scheduling system for scaling LLM fine-tuning experiments on limited GPU resources, enabling 100x more experiments on 4 GPUs.
3/24/2026
Financial Benchmarks – Evaling LLMs on Financial Tasks
Benchmark dataset for evaluating LLMs on financial tasks. Minimal content provided but addresses important evaluation gap.
3/24/2026
Beyond the Birkhoff Polytope: Spectral-Sphere-Constrained Hyper-Connections
arXiv paper generalizing residual connections into multi-stream hyper-connections with spectral-sphere constraints to maintain identity mapping stability.
3/24/2026
Natural Gradient Descent for Online Continual Learning
arXiv paper on natural gradient descent for online continual learning in image classification addressing catastrophic forgetting in non-i.i.d. data streams.
3/24/2026
Bayesian Scattering: A Principled Baseline for Uncertainty on Image Data
arXiv paper proposing Bayesian scattering as interpretable baseline for uncertainty quantification on image data using wavelet transforms and probabilistic modeling.
3/24/2026
LLM-ODE: Data-driven Discovery of Dynamical Systems with Large Language Models
LLM-based approach for automated discovery of governing equations in dynamical systems, replacing genetic programming with language models for efficiency.
3/24/2026
Enhancing LIME using Neural Decision Trees
Interpretability method combining LIME with neural decision trees for more stable and faithful explanations of complex models on tabular data.
3/24/2026
Discriminative Representation Learning for Clinical Prediction
Clinical prediction framework using discriminative representation learning aligned to outcomes rather than reconstruction objectives.
3/24/2026
Causally-Guided Diffusion for Stable Feature Selection
Feature selection method using causal principles and diffusion models to improve stability under distribution shifts.
3/24/2026
Understanding Contextual Recall in Transformers: How Finetuning Enables In-Context Reasoning over Pretraining Knowledge
Investigation of how contextual recall emerges in transformers during pretraining vs. finetuning, examining in-context learning mechanisms for fact retrieval.
3/24/2026
Detection of adversarial intent in Human-AI teams using LLMs
Detection framework using LLMs to identify adversarial attacks against human-AI teams, covering data poisoning, prompt injection, and prompt engineering threats.