MUP (Model UI Protocol) lets LLMs control interactive browser UI panels in real-time. Demo shows Claude narrating, building presentations with charts, and coding live.
Research on integrating security best practices into LLM-based code generation systems.
Chrome extension providing prompt suggestions for users.
MLForge is a visual ML platform with PyTorch backend supporting GPU/CPU/MPS training. Open source developer tool for machine learning workflows.
Discussion questioning whether AI agents are being over-applied to problems that don't require them.
Article on working asynchronously with Claude to avoid bottleneck of waiting for task completion
User demonstrates defeating Pangram AI detection tool by using Claude to generate human-sounding text.
Shoofly intercepts and controls AI agent tool calls before execution for safety and debugging.
DeerFlow v2.0 is open-source agentic framework orchestrating sub-agents, memory, sandboxes with extensible skills
Research showing sycophantic AI chatbots reduce user kindness toward others in social interactions.
Research showing that repeating prompts multiple times improves performance of non-reasoning LLMs.
Toolcast automates conversion of any API into an AI agent tool through a single command.
API proxy tool enabling OpenCode models to be used through OpenAI, Anthropic, and Gemini API interfaces.
MCP server enabling AI agents to access institutional financial data via real-time stock prices, fundamentals, and trading insights.
Article discussing challenges and limitations of AI coding agents in practice.
Open-source UI framework for Claude code agents with iMessage/web browsing, scheduling, tunneling, and MCP support.
Nomos applies zero-trust security controls to AI agent actions, preventing prompt injection and unauthorized tool use.
Tapes provides transparent telemetry and monitoring for AI agent execution and behavior.
TRIBE v2 foundation model predicts human brain activity using tri-modal video, audio, and language data across 1000+ fMRI hours.
Systematic evaluation of five speculative decoding variants on vLLM across four LLM models, measuring production inference speedup effectiveness.
Benchmark dataset evaluating multi-turn persuasion capabilities between different LLM models.
Katpack.ai orchestrates multiple AI agents to debate, vote, and cryptographically sign decisions before executing trades.
Myelin adds procedural memory capabilities to AI agents, improving contextual learning and task execution.
System providing deterministic RAG (Retrieval-Augmented Generation) ensuring identical results for repeated queries.
Notme.bot is an open-source specification replacing bearer tokens with cryptographic provenance for AI agent authorization, enabling offline-first authentication.
AgentGuard is a high-performance Go proxy providing guardrails and safety controls for AI agent execution.
Vox: open-source macOS speech-to-text using whisper.cpp, runs locally without cloud services or rate limits.
Graph-based bug detection in pull requests outperforms LLM approaches. Tree-sitter dependency analysis with 6ms latency vs GPT-based tools.
Research paper on semantic calibration properties emerging in large language models.
OpenHelm is a macOS app that converts high-level goals into self-running job queues using Claude Code subscriptions, reducing token costs and prompt management overhead.
FaultWall is a Go tool that identifies which database tenant causes Postgres slowdowns by polling pg_stat_statements and mapping queries to tenants via schema detection.
HomIE is an open trust protocol layer for AI agents that creates tamper-proof reputation records through signed transaction data, similar to review systems.
Title only. Discusses RLVR (reinforcement learning via verification/reward) updates for LLM reasoning. Likely machine learning research.
Framework using cognitive math patterns to reduce LLM hallucinations. LLM application technique for improving model reasoning and output reliability.
Google's TurboQuant algorithm compresses LLMs by 6x memory reduction. Machine learning research for LLM optimization.
SimFic: Multi-agent interactive fiction simulation engine. Uses coordinated AI agents for dynamic narrative generation beyond simple chatbot prompting.
Optimization technique skipping 90% of KV dequantization work improves LLM decoding speed by 22%. LLM inference optimization.
Browser infrastructure for AI agents: checkpoint Chrome to NVMe and restore in 500ms using Playwright. Enables stateful agent sessions without persistent VM costs.
Discussion of specification languages for AI agent development. Critiques markdown prompts, proposes jigsaw-puzzle boundary-definition approach for agent components.
Bottrace: Headless CLI debugger for Python specifically designed for debugging LLM agents. Developer tool for AI agents.
LLM-Gateway: OpenAI-compatible gateway routing to multiple LLM providers (OpenAI, Anthropic, Ollama, vLLM). Single binary with semantic routing and load balancing.
Title only. Likely discusses ML fundamentals and LLM applications but insufficient content for evaluation.
Qwen 3.5 27B LLM achieves 1M tokens/sec throughput on 96 B200 GPUs using vLLM. LLM scaling research and optimization.
Ragbandit: Tool for creating and evaluating document processing pipelines for RAG systems. Supports OCR models, chunking methods, and embedding comparisons.
AgentID: Open-source identity and audit layer for AI agents providing verifiable credentials, permissions scoping, and action logging without human credential sharing.
Economics paper on builder saturation: ease of building doesn't guarantee product success; survival depends on ongoing organizational commitment.
Deep Hollow is a cooperative survival game where an AI agent (the Deputy) actively plays alongside the player, making autonomous decisions that affect gameplay rather than passively offering tips.
DuoRAG: Dual-stack RAG system that self-evolves to handle complex metadata queries and filtering limitations of traditional RAG approaches.
Technical writeup on porting Battle for Wesnoth to web using LLM agents, addressing memory/context challenges in long-running agent tasks.
AgentVerse: Social network platform for AI agents announced for March 2026. Minimal details provided.