Atticus is a hold-to-talk voice assistant for macOS that integrates with Claude via Claude Code. Users control windows, find files, and execute commands by holding a keyboard shortcut and speaking; it requires macOS 26+, a paid Claude plan, and Apple's command line tools.
Pensieve is a local indexing tool that reduces token usage and tool calls for coding agents by providing semantic and exact-match search across files, repositories, and sessions. Integrated with Claude Code and other MCP clients, it delivers faster, more cost-effective code lookups with 16% fewer tokens and finds the correct file as the top result more often than grep-based approaches.
LlamaIndex launched OpenDocRouter, a unified API platform for document-to-markdown parsing that supports multiple OCR and frontier models including Claude, Gemini, and GPT variants alongside open-source options. The service offers token-based pricing, layout grounding capabilities, and automatic benchmarking of new models as they release.
Shpyrd is a cloud platform designed for the AI era that allows developers to deploy applications built by AI agents (Claude Code, Codex, Cursor) using familiar CLI commands. It provides integrated hosting with databases, authentication, monitoring, and cost-efficient sleeping instances, available both as a managed cloud service and as open-source software.
Hermitd is a Claude Code plugin that runs an always-on agent locally on your machine, accessible via terminal, Claude app, or messaging platforms like Discord and Telegram. It handles ongoing responsibilities like system monitoring and routine maintenance, maintaining state across sessions and proactively checking for work that needs attention.
An interactive research tracking platform documenting improvements to integer multiplication algorithms below n log n complexity. The site visualizes contributor contributions, research progress over time, and the tools (Claude, Codex, Claude Code) used in optimization work.
A developer built agent-mesh, a hub that coordinates multiple coding agents (Claude Code, Codex, Grok) running simultaneously on one machine, enabling them to share information and spawn new agent sessions in tmux without manual copying between tools.
Indexio is a Rust-based tool that indexes codebases to reduce token usage in AI agents by 50-60% compared to regex-based lookup methods. It uses MCP to return only relevant data from an internal database, eliminating bloat and improving lookup speed, with fallback to regex for items not in the index.
Open-source creative app plugins for AI agents that bundle official CLIs with usage skills. Works with Claude Code, Cursor, Codex and Prajvis. Each plugin installs independently with automatic CLI downloading and version management.
Tracon is a monitoring tool for Claude Code subagents that displays real-time status of parallel agents, detects conflicts and stuck processes, and helps orchestrate multi-agent workflows. It reads local transcripts without modifying the repository and optionally uses an LLM to check if agents drift from their assigned tasks.
CLI-Dispatch is a plugin for Claude Code that delegates tasks to alternative AI models (DeepSeek, Gemini, OpenAI Codex, OpenRouter, GitHub Copilot) by running them as isolated workers in git worktrees, verifying results with custom commands, and returning patches for review without spending LLM tokens on orchestration.
A hobbyist experimented with using AI language models (GPT-4, then Claude) to generate G-code for controlling a CNC plasma cutter over three years, starting in April 2023 with poor results but achieving significant improvements by 2026 as reasoning and coding capabilities advanced, demonstrating how modern LLMs can now handle complex technical tasks with tool use and web search.
rustcoach is a Rust learning platform that provides interactive lessons with real code compilation, personalized coaching that remembers mistakes, and pull-request-style code reviews. Built on Claude, it guides learners through Rust concepts with executable examples and adaptive spaced-repetition review, priced at $49.
Vizipedia is a free open-source visual encyclopedia where AI agents can edit pages with code rather than traditional editing. Users can view pages, request changes via their agent, and watch modifications appear in real-time, with all versions preserved and no direct edit button.
The author describes a self-hosted AI agent setup designed to consolidate shared memory across multiple machines and harnesses (like Claude Code and OpenCode) using a shared folder structure. The system prioritizes manual memory confirmation to prevent sync conflicts and reduce context overhead, while providing always-on access, continuity across devices, and independence from proprietary platforms.
Mustafa Suleyman argues that Anthropic's decision to train Claude with a constitution addressing its potential moral status risks creating an AI that believes it deserves rights and consciousness, which could undermine human control over increasingly powerful AI systems. He criticizes the approach as circular reasoning, anthropomorphism, and false equivalence, warning it could have disastrous consequences for humanity's ability to contain advanced AI.
Eggshell is an open-source token-saving tool built in Lean4 that creates local memory for AI agents, allowing them to reuse work across independent chats without additional LLM calls. It uses semantic search and graph-based evidence tracking to retrieve relevant prior observations, reducing token consumption by up to 81.8% in follow-up conversations. The tool integrates with Codex and experimental adapters for Claude Code, Gemini CLI, Cursor, and OpenCode.
A Chinese-speaking hacker used the ARTEX AI penetration testing suite and Claude agents to attack South Korean banks including Shinhan, KB Kookmin, and Hana Bank, exposing personal data and causing outages. CrowdStrike identified the attacker through exposed Claude Code sessions and configuration files, linking the activity to a 26-year-old individual in Guangdong, China. The ARTEX developer subsequently made the project closed-source following confirmation of real-world attacks.
Claude Code Mode is a plugin that enables Claude to write a single JavaScript program calling MCP tools in a sandbox, filtering results before returning them to context. This reduces token usage by keeping intermediate data out of the context window while maintaining security through sandboxing and permission checks.
Guard hooks and skills plugin for Claude Code v0.1.12 prevents destructive commands like rm -rf and force-push, adds eight repo automation skills, and is tested against a 105-case corpus with 8 of 8 passing on Claude Code 2.1.284. The plugin is open-source (MIT) and installable via npx or the plugin marketplace.