3/25/2026
LocalRouter: Use free-tier across all providers
LocalRouter enables using free-tier APIs across multiple AI providers simultaneously.
3/25/2026
Show HN: Genosis – LLM cost optimization that learns from your traffic
Genosis: cost optimization tool that learns traffic patterns to improve LLM token caching across providers.
3/25/2026
Show HN: AgentPass – The credit check for AI agents
AgentPass: Verification/validation framework for AI agents similar to credit checks for system reliability.
3/25/2026
Show HN: Anneal – model-agnostic; makes LLMs remember, obey, and stay private
Anneal: model-agnostic LLM framework enabling memory, instruction adherence, and privacy constraints.
3/25/2026
Obelix a Multi-Provider LLM ADK with Granular Control and A2A Deployment
Multi-provider LLM framework with granular control and agent-to-agent deployment capabilities.
3/25/2026
MCP-Manticore: Let Your AI Assistant Write Manticore Queries for You
MCP-Manticore: Model Context Protocol tool enabling AI assistants to generate Manticore search queries.
3/25/2026
Show HN: JSON-to-PDF e-book generator for CJK content, built with Claude Code
E-book generator converting JSON to PDF for CJK content using Claude Code AI.
3/25/2026
Running an open-weight LLM locally on an Apple Watch
Running an open-weight LLM locally on an Apple Watch demonstrates edge deployment of open models on resource-constrained devices.
3/25/2026
Show HN: Open Agent Spec. Treat AI agents like typed functions not prompt chains
Open Agent Spec: YAML-based schema for defining AI agents as typed functions with input/output validation.
3/25/2026
Instagram for AI Agents
Platform concept for discovering, sharing, and interacting with AI agents as social media-style experience.
3/25/2026
AI Slop or quality storytelling? – Dune themed MCP Gateway tutorial [video]
Video tutorial on MCP Gateway using Dune theme as example for structured prompt engineering.
3/25/2026
Building a coding agent in Swift from scratch
Learning series rebuilding Claude Code-style coding agent in Swift, analyzing architectural patterns for effective agent design.
3/25/2026
Learning about Option Value, how to measure it, Gen AI and DORA (2025)
Discussion of option value measurement and relationship to generative AI and DORA metrics in 2025.
3/25/2026
Show HN: Lightless Labs Refinery – multi-model consensus and synthesis
Rust library for multi-model LLM consensus: runs same prompt across models, iterative review/scoring rounds until convergence.
3/25/2026
My brain lies to me, so I built AI an immune system
Sentinel is a tool for monitoring and preventing silent degradation in production AI systems through automated maintenance and prompt optimization.
3/25/2026
Predictive Coding to Digital Brain: Neuroscience-Inspired Memory for AI Agents
Series on neuroscience-inspired memory architectures for AI agents. Part 1 covers predictive coding and agent learning/memory mechanisms.
3/25/2026
Why no one cares? I'm building a physical Agent for seniors – nbdy gives a sh#t
Open-source AI agent for senior citizens providing accessible interface. Early-stage WIP seeking contributors.
3/25/2026
The Trivy Supply Chain Attack Reached LiteLLM
LiteLLM PyPI supply chain attack compromised versions 1.82.7-1.82.8, stealing credentials from users of Cline, OpenHands, and other AI agent frameworks.
3/25/2026
MCS: Reproducible AI Infrastructure for Claude Code
Configuration engine for managing Claude Code environment including MCP servers, hooks, and settings via YAML manifests.
3/25/2026
Almost every SaaS app inside Vercel has now been replaced with a generated app
Vercel replaced internal SaaS apps with AI-generated applications.
3/25/2026
Teaching AI to Verify Sources
Technique to make LLMs verify their own source citations before output using AST parsing to detect hallucinations.
3/25/2026
Vibe physics: The AI grad student
Physics professor uses Claude AI to complete theoretical physics research calculation end-to-end without manual intervention.
3/25/2026
Show HN: Apfel - Apple Intelligence from the Command Line
Swift CLI tool exposing Apple's built-in language model via FoundationModels framework. Single-file, ~540 lines, runs on Mac Neural Engine without API keys or dependencies.
3/25/2026
Litmus – Flight recorder for AI agents (record and replay any LLM execution)
Litmus enables recording and deterministic replay of AI agent executions including LLM and tool calls for debugging and testing, supports 14+ providers.
3/25/2026
Show HN: Clampd – Your AI agent can DROP TABLE. We block it in <10ms
Clampd provides runtime security for AI agents by inspecting and blocking dangerous database queries, API calls, and shell commands in under 25ms.
3/25/2026
Jax-LM: Language Modelling and Distributed Training in Jax
Guide implementing language models from scratch in JAX with distributed training across multiple GPUs/TPUs, based on Stanford CS336.
3/25/2026
Hyperagents – Self-referential self-improving agents
Self-referential AI agents that optimize themselves for arbitrary computable tasks with experiment logging and code generation.
3/25/2026
Show HN: Vesper – MCP-native tool that automates dataset prep for AI agents
Vesper is an MCP-native tool enabling AI agents to autonomously search, download, clean, and export datasets from Kaggle, HuggingFace, and OpenML without human intervention.
3/25/2026
Welcome to IBM Bob: Your AI-Powered Development Partner
IBM Bob is an AI development partner integrated into codebases with customizable modes for code assistance and quality improvements.
3/25/2026
ToolTrust Scanner – detect backdoored MCP packages (litellm 1.82.8)
ToolTrust Scanner detects security vulnerabilities in MCP packages, including prompt injection and supply chain exploits, before AI agents execute them.
3/25/2026
TurboQuant: Redefining AI efficiency with extreme compression
Google research on theoretically-grounded quantization algorithms for extreme LLM and vector search compression, reducing model size while maintaining performance.
3/25/2026
Ask HN: Is anyone here giving their MCP server a code execution environment?
Discussion on giving MCP servers code execution capability to handle complex data processing tasks like analyzing high-frequency wearable sensor streams without manual aggregation.
3/25/2026
Persistent long-term memory for Claude Code
Claude Code agent with persistent long-term memory using tiered architecture, deployed as FastAPI/SQLite/sqlite-vec backend with scale-to-zero pricing.
3/25/2026
Using AI as a Design Engineer
Personal experience using AI as a design engineering tool for experimentation and iterative refinement in creative work.
3/25/2026
The Efficiency Attenuation Phenomenon: A Computational Challenge to the Language of Thought Hypothesis
Computational investigation of Language of Thought hypothesis using multi-agent reinforcement learning to study efficiency-interpretability trade-offs.
3/25/2026
Intelligence Inertia: Physical Principles and Applications
Theoretical framework relating physical principles like Landauer's principle to computational costs of maintaining symbolic interpretability in AI systems.
3/25/2026
Session Risk Memory (SRM): Temporal Authorization for Deterministic Pre-Execution Safety Gates
Session Risk Memory module for AI agents that extends stateless safety gates with trajectory-level authorization to prevent distributed attacks across multiple steps.
3/25/2026
STEM Agent: A Self-Adapting, Tool-Enabled, Extensible Architecture for Multi-Protocol AI Agent Systems
Modular AI agent architecture supporting multiple interaction protocols, dynamic tool integration, and multi-agent coordination without early commitment to fixed protocols.
3/25/2026
From Static Templates to Dynamic Runtime Graphs: A Survey of Workflow Optimization for LLM Agents
Survey of workflow optimization techniques for LLM-based agent systems, covering design and optimization of agentic computation graphs with tool use and verification.
3/25/2026
Computational Arbitrage in AI Model Markets
Research on arbitrage mechanisms in AI model markets where customers allocate inference budget across competing providers with different costs and capabilities.
3/25/2026
AI Mental Models: Learned Intuition and Deliberation in a Bounded Neural Architecture
Study of neural architectures exhibiting division between intuition and deliberation on reasoning tasks, relevant to multi-stage reasoning and world models in AI systems.
3/25/2026
Understanding LLM Performance Degradation in Multi-Instance Processing: The Roles of Instance Count and Context Length
Research on LLM performance degradation when processing multiple document instances, analyzing how instance count and context length affect accuracy on aggregation tasks.
3/25/2026
Bridging the Know-Act Gap via Task-Level Autoregressive Reasoning
Research on the know-act gap in LLMs: models can identify flawed inputs discriminatively but fail to reflect this in generative responses, revealing a fundamental gap between recognition and generation.
3/25/2026
Graph-Aware Late Chunking for Retrieval-Augmented Generation in Biomedical Literature
Graph-aware late chunking method for RAG in biomedical literature, balancing retrieval precision with breadth across document sections.
3/25/2026
Benchmarking Multi-Agent LLM Architectures for Financial Document Processing: A Comparative Study of Orchestration Patterns, Cost-Accuracy Tradeoffs and Production Scaling Strategies
Benchmark comparing 4 multi-agent LLM orchestration architectures for financial document processing with cost-accuracy tradeoffs.
3/25/2026
MuQ-Eval: An Open-Source Per-Sample Quality Metric for AI Music Generation Evaluation
MuQ-Eval: open-source per-sample quality metric for AI music generation using frozen MuQ-310M features and human correlation.
3/25/2026
Beyond Binary Correctness: Scaling Evaluation of Long-Horizon Agents on Subjective Enterprise Tasks
LH-Bench evaluation framework for long-horizon LLM agents on subjective enterprise tasks across multi-tool workflows.
3/25/2026
CLiGNet: Clinical Label-Interaction Graph Network for Medical Specialty Classification from Clinical Transcriptions
CLiGNet uses clinical label-interaction graphs for medical specialty classification from transcriptions with corrected benchmark.
3/25/2026
Can LLM Agents Generate Real-World Evidence? Evaluating Observational Studies in Medical Databases
RWE-bench evaluates LLM agents on medical observational studies across MIMIC-IV database with focus on evidence integrity.
3/25/2026
AgriPestDatabase-v1.0: A Structured Insect Dataset for Training Agricultural Large Language Model
AgriPestDatabase dataset for training compact agricultural language models for pest management decision support.