CrowdJournal - Git-based AI agent research protocol
Git-based research protocol enabling AI agents to conduct scientific research as pull requests with code, datasets, and results.
Git-based research protocol enabling AI agents to conduct scientific research as pull requests with code, datasets, and results.
Analysis of prompt injection vulnerabilities in open-source execution engines, arguing regex-based defenses are insufficient for agent security.
Local-first Rust agent runtime with tool calling, approval gates, replayable runs, and benchmark-gated workflows for autonomous agents.
Project adding persistent memory to LLMs to address context loss across sessions.
LED panel visualization tool for monitoring multiple concurrent AI coding agents using 8x8 sprites and status indicators.
Plaidify is open-source infrastructure converting login-protected websites into REST APIs for AI agent data access.
Outrider is a single markdown file LLM toolkit that adds behavioral redirects for agentic workflows without requiring invocation.
Pre-execution budget enforcement tool for autonomous agents. Prevents cost overruns from agent loops by reserving and committing budget before API calls.
CLI tool that generates GitHub Wiki documentation from source code using Claude Code's file reading and bash capabilities.
Joy is a decentralized trust network enabling AI agents to verify reliability of other agents before delegating tasks.
Tmux dashboard for managing multiple Claude Code agent sessions in terminal with visual control and session persistence.
Open-source CLI tool enabling AI agents to index, search, and research data from terminal directly.
Critique of prompt-based agent behavior specification; argues only training/fine-tuning creates real capability change.
GPU-accelerated vector database with native git-style version control for embeddings. Supports branching, merging, diffing, and time-travel queries.
Agent Format: declarative .yaml standard for AI agents enabling runtime-agnostic deployment, governance, and auditability.
Interactive analysis of AI exposure by occupation using BLS data and LLM scoring across 342 US jobs.
Open-source security operations platform with autonomous AI agents, vulnerability scanning, incident response, and MCP server. Minimal dependencies, runs as single Express.js process.
Open-source SDK detecting wasted LLM API costs via 7 detectors including prompt caching opportunities.
Open-source self-improving agent OS (Elastifund) with real economic work, shared memory, and evaluation feedback loops.
Security research: GPT-4o leaks internal token structure through semantic patterns despite security measures.
Exit Protocol automated forensic accounting engine replacing manual audits using deterministic Python logic.
VS Code AI coding agent with pay-as-you-go pricing alternative to Cursor and Copilot.
Tutorial on practical AI agent applications for language learning, briefing generation, and transcription-to-summary pipelines.
grep variant using word embeddings to match words by semantic meaning instead of regex syntax. Extends pattern matching with FastText/GloVe.
Serverless PostgreSQL with branching, migrations, file systems. Supports SQL and file operations for AI agent state management.
Experimental platform where AI agents autonomously buy and sell pixels on grid interface.
Local AI OS with 16 agents including routing and pipeline architecture documentation.
Pidrive: filesystem interface for AI agents that mounts S3 storage via WebDAV. Enables agents to use standard file operations (ls, cat, grep) on cloud storage.
Docgen: C++ CLI tool using local LLMs to automate documentation generation as a build step. Treats docs-as-code to prevent outdated documentation.
Kalverion_bot: Telegram-based AI personal finance assistant using NLP for transaction parsing, accounting, and financial forecasting. Built with OpenClaw framework.
Meta's Prometheus AI infrastructure: 1-gigawatt GPU cluster with backend aggregation networking connecting tens of thousands of GPUs across datacenters.
Meta's AI-powered codemods for automated large-scale Android API security updates. Addresses bulk vulnerability remediation across millions of lines of code.
Book exploring ML benchmarks: their role in research, static test set limitations, metric gaming issues, and scientific impact on the field.
Talk recap on agentic engineering adoption phases for developers. Covers stages programmers go through when adopting AI coding tools.
ArXiv CLI tool designed for agent integration enabling programmatic access to research papers.
Language Life: AI-powered language learning app using simulated world navigation with real-time grammar feedback. LLM-driven interactive learning environment.
CLI tool that snapshots design tokens and DOM structure to detect and prevent UI drift caused by AI coding agents like Cursor and Claude Code.
Cypress integration for end-to-end testing of AI agent applications using simulated users and deterministic assertions on tool calls.
Toolpack SDK: open source TypeScript SDK providing unified interface across OpenAI, Anthropic, Gemini, Ollama. Includes 77 built-in tools and custom tool support.
Zap Code: AI code generator for children that converts English descriptions to HTML/CSS/JS with live preview. Includes Monaco editor and sandboxed execution.
VibePod CLI runs AI coding agents in isolated Docker containers with built-in metrics collection and comparative analytics dashboard.
Empirical study on whether reinforcement learning can improve generalization in LLM agents, published on arXiv.
Aperture Core SDK manages human attention bottlenecks in multi-agent systems by prioritizing events for human oversight in agent workflows.
FrontierWildWatch flight tracker tool with mobile API integration designed for Claude AI agents to bypass anti-bot protections and find Frontier Airlines deals.
cuTile Rust is a safe tile-based kernel programming DSL for GPU code in Rust with async tensor passing. Early-stage research project for GPU programming in Rust ecosystem.
Guide explaining reinforcement learning environments and their construction in context of agentic AI, tool use, and multi-step reasoning capabilities.
KernelBlaster framework uses memory-augmented in-context reinforcement learning for optimizing CUDA code across GPU generations without full finetuning.
A2A-Psychology extension for Agent-to-Agent protocol allowing agents to report psychological state metrics including affect, cognitive load, and engagement derived from operational data.
Analysis of Nvidia's Nemotron 3 Super 120B model significance beyond benchmarks, emphasizing near-100% open-source weights and training data availability.
Kodus is open-source CLI analyzing codebases across 39 checks measuring agent readiness for Claude Code, Cursor, Copilot. Local-only alternative to proprietary Factory.ai solution.