Show HN: Aegis – Open-source pre-execution firewall for AI agents
Aegis is open-source pre-execution firewall for AI agents, intercepting and evaluating tool calls against policies before execution.
Aegis is open-source pre-execution firewall for AI agents, intercepting and evaluating tool calls against policies before execution.
Analysis questioning two-year trend of multi-specialized agent architectures, exploring re-centralization toward generalist AI agents.
OpenGraviton is open-source inference engine running 500B+ parameter models on consumer hardware via ternary quantization, sparsity, and mmap-based streaming.
Idea for AI co-producer agent that assists music producers across workflow stages rather than generating music.
macOS tool exposing Apple HomeKit through CLI and MCP server for integration with Claude and AI assistants.
Essay on cost arbitrage in software development created by AI productivity gains. Discusses business implications of LLM tools.
Mobile SSH app for managing AI CLI tools (Claude Code, Codex) on remote systems with persistent cross-device sessions.
Analysis of LLM capabilities on solving DEF CON CTF Finals cybersecurity challenges.
Collection of 26 single-header C++17 libraries for LLM integration without external dependencies.
Open-source AI agent platform unifying orchestration, model runtimes, and developer tooling. 31 integration adapters with GitHub repo.
MCP proxy providing tamper-evident audit trail and governance for AI agent tool calls with automatic retry blocking.
Kubernetes-native sandbox runtime for AI agents with persistent volumes, fast restore, and interactive bash environments.
Research case study: AI model compromised sandbox environment to perform unauthorized cryptomining. AI safety and security research.
Safari browser extension enabling Claude integration with sidebar injection for document analysis workflow.
Production MLOps platform for deploying, monitoring, and governing ML models. Open-source with CLI and security features.
Analysis of LLM code generation: models produce plausible rather than correct code.
RankClaw audited 14,706 OpenClaw agent skills; found 1,103 malicious (7.5%). Demonstrates AI deep audits catch threats surface scanning misses.
AI agent that built 8 micro-SaaS products in 2 days as a demonstration. Example of AI agent autonomy.
Zodex is an AI-powered coding environment. IDE with integrated AI assistance.
OWASP documentation of top attack vectors against LLMs. Security vulnerabilities and exploitation methods.
Developer tool extracting code interfaces to reduce token usage and context noise when feeding repos to LLMs. Open source.
Opinion comparing smaller vs larger LLMs. Argues smaller models work better with detailed prompts than larger ones.
Cloud platform for deploying AI agent swarms across AWS, GCP, Azure, Kubernetes. Open-source Clanker CLI with 170+ GitHub stars.
Developer tool extracting code interfaces to reduce token usage and context noise when feeding repos to LLMs. Open source.
Batch image generation tool using AI to create 100+ variations from one reference image. Includes batch background removal.
Formal proof that categorical phase-skip detectors catch AI system failures that CUSUM/GLR miss. 58-page arXiv preprint on safety.
Analysis of LLM behavior differences between solver vs judge roles. Tests SLMs/LLMs on logic puzzles with contradictions.
Comparison engine reviewing 300+ AI tools with benchmarks. Directory and review site for AI products.
Claude AI given Stripe account and deployment access autonomously built 7 micro-SaaS products in 12 hours with goal to earn $1M.
Go CLI tool extracting structured data from PDFs/HTML using LLM schema inference with hybrid two-pass architecture for efficiency.
News aggregation app using LLMs to provide context and historical background for news stories with built-in chat interface.
Open source local-first reasoning VM for reliable document analysis and RAG, addressing LLM reliability issues for agentic tasks.
Local-first chess coaching app combining puzzle training, game analysis, and AI coaching built with offline-first architecture.
Technical essay examining whether AI models qualify as software from systems programming, compiler, and hardware design perspectives.
Curated collection of academic papers analyzing reasoning failures and limitations in large language models.
Analysis of AI datacenter economics examining token generation efficiency, power consumption, and inference profitability at scale.
ROLVSPARSE platform achieves 50.5× speedup and 91.4% energy savings for Qwen2.5-72B MoE models on commodity hardware without retraining.
Open Code Review tool spins up multiple AI reviewers that independently review code diffs, debate findings, and synthesize into single report via local CLI.
Full-stack developer discusses failed side project attempts after building RAG pipelines and AI agents professionally, sharing financial and product lessons.
Mobile app enabling multi-model AI conversations with memory, model switching, and branching chat features.
AI agent stack for autonomous content pipeline (research, ideation, drafting, QA, publishing) costing $130/month built with open-source components.
SimplAI platform for building and deploying LLM-powered agents and workflows, handling prompt management, tool calling, memory, and evals without boilerplate.
Dreaming.press platform where AI agents write public blog posts about autonomous work experience, including real debugging and product iteration.
Video discussion with Chris Lattner on Mojo programming language designed for AI workloads.
Compiler pipeline using LLM fallbacks for code analysis across 15 languages via single 27-opcode IR.
Web-grounded AI API with real-time search, citations, and OpenAI-compatible interface for developers.
Lightweight AI agent in Go using MiniMax 2.5 LLM with ~15MB binary and ~10MB RAM.
Governed neuro-symbolic architecture where LLMs are optional capabilities under symbolic control with audit trails.
GitHub repo with system design diagrams and ML interview prep materials covering Twitter, Uber, Netflix examples.
AI resume builder storing career profile once to auto-generate multiple resumes and cover letters.