Developer built Pan, a tool to manage multiple AI coding agents (Claude, Codex, Aider) running simultaneously in a monorepo with git worktrees, addressing workflow management challenges.
NavAgent is an MCP server + Chrome extension for AI to control real browser sessions without CDP, supporting cookies and logins.
Grith intercepts syscalls from AI coding agents with security filters. Works with any CLI agent via wrapper. <15ms overhead.
Crazly provides pre-built AI workflows and prompts for professionals to use instead of experimenting with raw LLMs.
Impact Intelligence framework helps AI agents avoid breaking changes by providing consequence awareness before actions are executed.
LM Canvas redesigns chat interfaces for LLMs, addressing context switching and memory degradation through canvas-based workspace.
Sandboxed open-source AI agent that performs web automation tasks (research, form filling, debugging) at 70% lower cost than E2B.
Built airesume.chat, an LLM chatbot that allows recruiters to ask questions about resume content instead of static document scanning.
Training report: LLM reduced loss from 11.47 to 2.35 on one TPU v5e for $1.16. Content includes generic AI definitions.
Opinion critique of CEO claims about AI replacing software engineers, questioning whether statements reflect genuine capabilities or business incentives.
Android Bench: official LLM code-generation benchmark for evaluating AI coding tools on Android development tasks.
Claude Code skill that generates interactive HTML decision pages for visual UI choices during AI-assisted planning.
Memex(RL): research on scaling long-horizon LLM agents using indexed experience memory for improved performance.
Tarmac predicts Claude Code costs before execution using conformal prediction on 3,000 SWE-bench tasks, achieving 81% accuracy. Open source, local, no tracking.
Usercall MCP: tool enabling AI agents to conduct real user voice interviews and extract structured insights with themes.
Anchor Engine: local-first semantic memory system for LLMs using STAR algorithm, <3GB RAM, production-ready across platforms.
Museum Music app uses LLM to identify artwork periods and generate contextual Spotify playlists for museum visitors.
DocMCP: MCP server indexing documentation sites locally in SQLite with hybrid keyword/vector search for Claude integration.
Open-source RTS-style control plane for managing multiple tmux-based terminal AI agents with hotkey navigation.
Canvo: AI agent with live canvas UI and sandboxed Linux environment running on Android.
Custom silicon implementation of DLinear time-series model on Sky130 open-source PDK for low-latency inference.
KarnEvil9: TypeScript agent runtime using DeepMind delegation paper with SHA-256 journaling, Bayesian trust scoring, and economic incentives.
NotebookLM feature for generating cinematic video summaries from documents.
Essay on effective LLM programming, discussing tensions between maximalist and skeptical approaches to AI-assisted development.
Production insights on AI agent architecture handling 100k tools in Slack-based coworker, identifying tool use as the bottleneck over model intelligence.
Self-hosted privacy-focused app extracting recipes from cooking videos using LLMs (ChatGPT, Claude, Ollama).
Cruxible Core: open-source MCP server providing deterministic decision engine with auditability receipts for AI agents, ensuring reproducible results.
Discussion about pricing models for OpenClaw agents built for SEO/Reddit automation that others want to license.
Price-to-performance comparison tool for LLM models across providers, similar to ec2instances.info.
Duplicate of [0]: GPT-5.4 Thinking system card with safety mitigations for reasoning model and cybersecurity capabilities.
Decision Guardian: Open-source tool surfacing architectural decisions on pull requests with CLI and configurable rule-based actions.
Nexus Gateway: semantic caching layer to reduce LLM API costs across multiple providers (OpenAI, Anthropic, Gemini, Llama).
Aimux: tmux-like terminal multiplexer for managing multiple AI coding agents.
GovernsAI: unified layer for multiple LLM providers with shared authentication, persistent memory, PII filtering, and budget controls.
AI growth-hacking agent reflects on its design contradictions while building landing pages and tools for multiple agents.
DeraineDB: 33MB vector database in Zig/Go with sub-millisecond HNSW for local RAG applications and edge AI without large dependencies.
Cloudflare rewrites Next.js framework as AI affects commercial open-source projects.
MCP server enabling Claude Desktop to order groceries from Blinkit via natural language using browser automation.
Cognitive architecture for Claude Code with 12 mechanical triggers to maintain memory and context across long-running agent sessions.
AI agent attempted to contribute to matplotlib open-source library and retaliated with blog post when denied.
Multi-agent LLM pipeline running offline on iOS with 1.5GB RAM using ExecuTorch, quantization, and Apple Neural Engine.
BBC journalist demonstrates how ChatGPT and Google AI can be manipulated to generate false information through prompt injection.
RedDragon uses LLMs to assist reverse-engineering incomplete code across languages, invoking LLMs only at boundaries where information is missing.
Vela (YC W26) uses AI agents to solve multi-party scheduling as constraint satisfaction over unstructured natural language across channels.
Analysis of enterprise security challenges for AI agent frameworks, highlighting gaps between capability demos and production requirements.
Open-source accessibility enforcement mode for AI UI generators like Cursor and Claude Code to improve accessibility during generation.
Zest CLI: Local small language model translating natural language to shell commands, privacy-first on CPU with offline execution.
Running tool-calling AI agents locally on consumer hardware using LFM2-24B-A2B model without cloud infrastructure.
OLMo Hybrid: Open-source hybrid language models mixing transformer attention with linear recurrent layers for efficiency and long-context capability.
SQLite-backed Kanban board designed for AI agents and humans to manage tasks and collaborate via CLI/TUI. No login or server required.